University of Southern California Dissertations and Theses

Active microdisk resonant devices and semiconductor optical equalizers as building blocks for future photonic circuitry

(USC Thesis Other)

Active microdisk resonant devices and semiconductor optical equalizers as building blocks for future photonic circuitry

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ACTIVE MICRODISK RESONANT DEVICES AND SEMICONDUCTOR OPTICAL EQUALIZERS AS BUILDING BLOCKS FOR FUTURE PHOTONIC CIRCUITRY b y Kostadin Dimitrov Djordjev A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOS OPY (ELECTRICAL ENGINEERING) December 2002 Copyright 2002 Kostadin Dimitrov Djordjev Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 3093754 UMI UMI Microform 3093754 Copyright 2003 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UNIVERSITY OF SOUTHERN CALIFORNIA THE GRADUATE SCHOOL UNIVERSITY PARK LOS ANGELES, CALIFORNIA 90089-1695 This dissertation, written by ________ K o sta d in D im itr o v D jo r d je v under the direction of h i s dissertation committee, and approved by all its members, has been presented to and accepted by the Director of Graduate and Professional Programs, in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY X Director Date Dorenihpr 18. 2002 DissertdtionJCbmmitvee Chair Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Acknowledgements This dissertation represents the collective effort of many people, which I would like to express my gratitude to. Without their help and encouragement the completion of this work would not have been possible. I would like to express my greatest gratitude to my advisor, Professor P.D. Dapkus. He was not only my advisor but a tutor and a friend, who made my academic research a challenging, exciting and rewarding experience. I had his help whenever I needed it; his advice was always well made and helped in overcoming problems. His ideas were ‘golden’. He used to say that research is 90% frustration and 10% excitement from the achieved results. At the end I realized that he was right and that the joys of success were absolutely worth the whole experience. I would also like to thank Dr. John O’Brien for his helpful discussions and new ideas and his willingness to help and spend as much time as necessary with me to solve a problem. I would like to thank the following members of our lab: Seung-June Choi, who is a good friend of mine and with whom I share the microresonator work. He is an excellent scientist and was always willing to help and embrace new ideas and approaches; Sang Jun Choi, who was able to grow the material which made this research possible; Ryan Stevenson, for the good friendship, discussions and his help with the high-speed measurements. I am also very thankful to the CSL alumni: Denis Tishinin, who introduced me to processing and testing and was very patient while teaching me; Won-Jin Choi, who ii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. was really helpful, experienced and knowledgeable and who had always an advice for me; David (Chao-Kum) Lin, for his help. I would also like to thank all of my friends who always supported me. Thiruvikraman Sadagopan, Yuanming Deng, Ruijuan Li, Zhi-Jian Wei and Dawei Ren for their interesting discussions. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table of Contents ACKNOWLEDGEMENTS......................................................... ii LIST OF TABLES...............................................................................vii LIST OF FIGURES............................................................................viii ABSTRACT...................................................................................... xviii CHAPTER 1 INTRODUCTION...................................................... 1 1.1 Evolution of Fiber Optic Tr a n sm issio n ....... 1 1.2 Semiconductor optical equalizer................................................................4 1.3 Semiconductor microdisk wavelength selective elem ents 7 1.4 Outline ........ 16 1.5 References............................................. 16 CHAPTER 2 THEORY OF PULSE PROPAGATION IN A SOA.. 18 2.1 Theoretical background ...........................................................18 2.2 N umerical solution ........................... 22 2.3 Results and d iscussio n..................................................................................... 26 2.3.1 Results: C W operation ...... ....27 2.3.2 Results: Pulse propagation ..............................................................28 2.3.3 Results: Pulse - effect o f gain reco v e ry............................................. 30 2.3.4 Results: Pulse - effect o f the input ch irp............................................32 2.4 N oise theory in SO A .......................................................................... 34 2.4.1 Results: N o ise.................................................. 37 2.5 C o n c lu s io n s ................................................ 39 2.6 References........................................ 40 CHAPTER 3 DESIGN OF AN OPTICAL EQUALIZER........... 42 3.1 D esign Is s u e s.................................................................................................. 42 3.2 Co n c lu sio n s ............................................ 51 3.3 References........................ 52 CHAPTER 4 SOA-EQ FABRICATION....................................... 53 4.1 Introduction to S A G ........................................................................ 53 4.2 W afer Preparation for S A G ............................................... 56 4.3 B road-area SAG devices ...................................................... 60 4.4 B uried-heterostructure (BH) d e v ic e s ...... 69 iv Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4.5 Ridge-w aveguide d ev ic es.......................................................................... 73 4.6 References........................................................ 79 CHAPTER 5 SEMICONDUCTOR OPTICAL EQUALIZER ....80 5.1 Results........................................................ 80 5.2 Co n c lu sio n s............................... 89 5.3 References.......... ..... 89 CHAPTER 6 FABRICATION AND MEASUREMENT OF VERTICALLY-COUPLED MICRODISK RESONANT DEVICES.. .................................................................................... 90 6.1 Introduction.................................................................................... 90 6.2 The fabrication pro cess ........................... 92 6.2.1 W aveguide form ation ........................................................................93 6.2.2 Wafer Bonding................................................... 98 6.2.3 Substrate rem oval and edge open in g .............. 101 6.2.4 M icrodisk mesa form ation .....................................................................103 6.2.5 Polyim ide planarization and m etal contacts fo rm a tio n ............... 109 6.2.6 Substrate thinning, n-contact definition, and mounting................112 6.3 M e asu rem ent ................................................ 116 6.4 References..................................................... ......................................................117 CHAPTER 7 PASSIVE MICRODISK DEVICES: DESIGN AND PERFORMANCE PARAMETERS.................................................. 119 7.1 Introduction........................................................................................................119 7.2 Structural parameters of interest.........................................................120 7.3 Properties of a microdisk resonant c a v it y ......................................... 123 7.3.1 D istributed loss..........................................................................................125 7.3.2 L ocalized lo ss............................................................................................126 1 A D ep en d en ce o n t h e d isk r a d iu s, R ................................................... 127 7.5 D ependence on the separation distance d c.................................. 130 7.6 D ependence on the w aveguide etch depth d wg................................. 133 7.7 D ependence on the thickness of the mem brane t............................ 138 7.8 Su m m a r y .... ................................................................... 140 7.9 References .................................................... 141 CHAPTER 8 ACTIVE MICRODISK DEVICES.........................142 8.1 D evices with a n Electroab sorptive A ctive Re g io n ....... 142 8.1.1 Theoretical ............................................. 142 8.1.2 E xperim ental......................... 148 8.1.3 Sum m ary........................ 155 8.2 D evices with a Gain A ctive Re g io n ........................................155 8.2.1 Theoretical.............................. 155 v Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 8.2.2 Experimental........................ 160 8.2.3 Summary. ........... 165 8.3 D evices with Free Carrier Injection A ctive Re g io n ..................... 166 8.3.1 Theoretical ........ 166 8.3.2 Experimental ........ 169 8.3.3 Summary........................................................... 175 8.4 D evices with Electrooptic A ctive Re g io n........................................... 176 8.4.1 Theoretical ......... 176 8.5 References................................................ 178 CHAPTER 9 SUMMARY...............................................................180 9.1 Semiconductor optical equalizer...........................................................180 9.2 M icrodisk resonant devices .................................................. 183 9.3 Future Research D irections.....................................................................188 BIBLIOGRAPHY.............................................................................. 190 APPENDICES .................................................................................. 195 A ppendix 1 SOA/EQ device process flow-c h a rt...................................... 195 Appendix 2 W afer bo n ding.................................... 202 A ppendix 3 Properties of a M icrodisk Resonant Ca v it y ................... 203 A3.1 Modal fields distribution........................ 203 A3.2 Calculation of the power coupling coefficient..............................207 Appendix 4 Coupling of modes in time fo r m a l ism ..................................216 A ppendix 5 Quantum Confined Stark Ef f e c t ...........................................223 A5.1 Finding the excitonic binding energy............................................ 223 A5.2 Absorption coefficient ................................................................227 A5.3 Solving the QCSE problem................................. 229 v i Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. List of Tables Table 4.1 Epi-structure used for growing SAG devices................................................ 59 Table 4.2 Epi-structure of the current blocking layers...................................................70 Table 4.3 Modification to the epi-structure shown in Table 4.1 as used for fabrication of ridge-waveguide devices..........................................................74 Table 7.1 An example epi-structure of a passive microdisk............... 120 Table 8.1 Epi-structure of an active microdisk resonator utilizing the Quantum Confined Stark Effect................................... 149 Table 8.2 Transmission at resonance T, quality factor Q, and the calculated modal absorption coefficient a, as a function of the applied reverse bias at A,=1584nm. The coupling coefficient, K , is assumed to be 4.5%. 151 Table 8.3 Epi-structure of a microdisk resonator with PCI active region..................170 Table 8.4 Electrooptic effect and its application to microdisk modulators................177 Table A l.l STEP 1: SAG stripe definition ................................................................195 Table A1.2 STEP 2: SAG pre-growth treatment + growth........................................ 196 Table A1.3 STEP 3: Isolation trench formation............................................................197 Table A1.4 STEP 4: Mesa stripe formation .......................................................198 Table A 1.5 STEP 5: Polyimide planarization....................... 199 Table A1.6 STEP 6: P-contact definition ........................ 200 Table A1.7 STEP 7: Substrate lapping and N-metals definition............................... 201 Table A2.1 Summary of the wafer-bonding process........................... 202 vii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. List of Figures Figure 1.1 A schematic of a simple WDM transmission system...... ..... 1 Figure 1.2 (a) Multiplexing and demultiplexing in a bidirectional system; (b) OADM selectively removing and adding wavelengths ..... 3 Figure 1.3 A schematic diagram of an optical analog-to-digital converter system. The optical equalizer is used to flatten a highly chirped Gaussian input pulse ...... 6 Figure 1.4 Microdisk switch.................................................................................................8 Figure 1.5 Channel-dropping filter......................................................................................9 Figure 1.6 WDM demultiplexer............................................. 9 Figure 1.7 Mach-Zender interferometer/ notch filter........................................................ 9 Figure 1.8 Second order add-drop filter............................................................................. 9 Figure 1.9 A vision for a future WDM PIC utilizing active microdisk devices..........11 Figure 1.10 Microdisk resonant coupler realized via lateral (a), and vertical geometry (b).............................................................................. 11 Figure 1.11 Modulation behavior of a microdisk resonator via change of the absorption coefficient (a), and change of the resonant frequency (b).........13 Figure 1.12 The frequency bandwidth is limited by the RC time constant (disk size) and the quality factor of the cavity......................................................... 14 Figure 2.1 Computational grid used for solving the problem of pulse propagation ..23 Figure 2.2 Calculated gain spectra for different carrier density (a), and the corresponding change of the refractive index (b) for an InGaAsP QW. ...25 Figure 2.3 Theoretical simulation of the saturation behavior of a TWSOA (a), and modal gain spectra for different pumping currents (b)..........................28 Figure 2.4 Effect of gain saturation and SPM on pulse shape and spectrum. The induced chirp has the shape of the output pulse.............................................29 viii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 2.5 Effect of gain recovery on input Gaussian pulses with different durations t0 .................... 31 Figure 2.6 Pulse shaping at different points along the SOA (a), and the broadened pulse spectrum at the output of the amplifier in the quazi- CW case (b)............. 32 Figure 2.7 Dependence on the output chirp on the chirp of the input pulse. The shape of the output pulse stays the same (a), but the output chirp is different, dependent on the value of the liner-chirp parameter: C=0 (b); C=5 (c); C=5000 (d)......................... 34 Figure 2.8 The calculated output noise power as a function of the relative time. The noise is greatly suppressed during the pulse propagation because of the saturated spontaneous emission ..................................................38 Figure 2.9 A tree-dimensional plot of the ASE spectrum as a function of time. The pulse arrive at T = -2, saturates the amplifier, and suppresses the output ASE..........................................................................................................38 Figure 3.1 A three-segment spectrally and spatially inhomogeneous gain medium................................................................................................................42 Figure 3.2 An example theoretical calculation of the gain spectra of two independent gain media pumped separately.................................. 43 Figure 3.3 The calculated CW spectrum of a two-section SOA. The short- wavelength section only equalizes the device spectrum at shorter wavelengths........................................................................................................ 44 Figure 3.4 A theoretical plot of the output pulse shape amplified by a single section device with a long-wavelength gain region...................................... 45 Figure 3.5 A theoretical plot of the output pulse shape amplified by a single section device with a short-wavelength gain region.................... 47 Figure 3.6 Pulse equalization from a two-section spectrally and spatially inhomogeneous optical amplifier......................... 48 Figure 3.7 The calculated output spectrum of the pulse plotted in Figure 3.6........... .49 Figure 3.8 SOA/EQ is not sensitive to variation of the input signal. Change in the input amplitude by more than 200% leads to 10% variation in the output .................................... 50 ix Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 3.9 Pulse equalization from a two two-section devices. The second amplifier completely equalizes the output pulse.........................................51 Figure 4.1 Selective area growth process: (a) the growth enhancement is due to gas-phase diffusion and surface migration; (b) the SAG profile after growth.................................................... 53 Figure 4.2 Orientation of a two-section device with respect to the SAG region 53 Figure 4.3 Cross-sectional SEM photograph of an SAG region just after the growth is performed. ........ 54 Figure 4.4 Selective area growth enhancement factor as a function of the dielectric stripe width, w......................................... 55 Figure 4.5 Pictures of six quantum wells grown at the same time within different SAG regions with mask widths w............... ...55 Figure 4.6 LI curves from samples with different pre-growth treatment. The best results are obtained with ‘B0E+02+clean’ treatment.......................... 58 Figure 4.7 Top view of the wafer surface with the orientation of the device (a), and cross-section of the SAG region close to the mask edge... ......................................................................................... 61 Figure 4.8 A process flow-chart for fabricating two-section, broad area SAG devices ................................................................................................. 62 Figure 4.9 A cross-section of a deep BA laser mesa processed within 15pm SAG region. A thick layer of polyimide is spun on top of it................................63 Figure 4.10 A SEM picture showing the SOA mesa top, after the polyimide curing and the application of short-time 0 2 plasma treatment. The PY layer is rough ................... 64 Figure 4.11 An unoptimized PY process could result in r.m.s. roughness of more than 2pm (a), and a smooth PY surface resultant from an optimized process (b). ........ 65 Figure 4.12 Resistance as a function of the RTA annealing temperature. Temperatures in excess of 420°C are required for obtaining low resistance............................. 67 Figure 4.13 A schematic of the processed BA device (a), and a SEM picture (top view) of the same device (b)..... 67 x Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 4.14 Calculation of the waveguide mode effective index and the reflectivity at the interface trench/waveguide as a function of the trench etch depth................................................................................................68 Figure 4.15 A BH mesa with grown CBLs. ...................................................................70 Figure 4.16 A cross-sectional SEM photograph of a completed BH device. A stain etch is performed to reveal the waveguide region containing the QW s. ............................................................................................... 72 Figure 4.17 Top view of a completed two-section BH optical equalizer. The mesa, the isolation trench and the gold bond-pads of the two sections are visible ....... ....73 Figure 4.18 SiNx mask overhanging on both sides of the RM-RWG laser mesa.......75 Figure 4.19 A SEM cross-sectional view of the processed RM-RWG device. The SiNx mask is still embedded into the polyimide layer.................................. 78 Figure 4.20 A SEM picture showing a top-view from the final RM-RWG device with polyimide layer and gold bond-pads...................................................... 78 Figure 5.1 Buried heterostructure SAG lasers; (a) LI curve, (b) IV curves................81 Figure 5.2 Growth enhancement ratio as a function of the dielectric stripe width w. The open space between the SiNx mask patterns is 15pm...................... 81 Figure 5.3 Dependence of the lasing wavelength (a) and the threshold current (b) on the dielectric stripe width w. The devices are broad area laser processed from the same wafer (#2526)..........................................................82 Figure 5.4 Dependence of the lasing wavelength (a) and the threshold current (b) on the dielectric stripe width w. The devices are broad area laser processed from the same wafer (#2940)..........................................................83 Figure 5.5 Spectral responses of four two-section optical equalizers with different width of the SAG region: (a) w=10pm, (b) w=15pm, (c) w=25pm, (d) w=30pm............................................ 84 Figure 5.6 A theoretical simulation of a two-section SAG device, whose experimental data is shown in Figure 5.5(d)...................................................86 Figure 5.7 Comparison between theory (dashed) and experiment (solid). The equalizer is processed from SAG regions with w=30pm stripe width and exhibit an equalized bandwidth of lOOnm................................. 87 xi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 5.8 Saturation behavior of the amplifier, shown in Figure 5.7, at 1=1.55pm with only the main section pumped. A saturation output power of lOdBm is measured. ....... ....87 Figure 5.9 Tunable two-section Fabry-Perot laser. Pumping the plain area section changes in the overall device gain spectrum and thus shifts the lasing wavelength................................... 88 Figure 6.1 Vertically coupled microdisk device with a post for improving the mechanical stability and current/field uniformity, (a), and the corresponding epi-structure, (b)....... 90 Figure 6.2 The post below the microdisk cavity improves: (a) mechanical stability, (b) current/field uniformity, and (c) forces single mode operation ...... 91 Figure 6.3 Vertical Coupler Fabrication Process Overview: (a) as-grown structure; (b) bus waveguide formation; (c) wafer bonding to transfer substrate; (d) substrate removal and disk formation..................................... 93 Figure 6.4 Bus waveguides and the circular post, which will be positioned below the microdisk. The mesa is 1.6pm deep after the dry etch with CPU/FF/Ar chemistry............................................. 97 Figure 6.5 Schematic of the bonding fixture. The small steel balls are expected to distribute the applied pressure uniformly over the samples.......................100 Figure 6.6 Bus waveguides at the bonded interface and the remaining transparent InP layer after the etch of the microdisk mesa....................................... 101 Figure 6.7 Etching experiments for obtaining vertical profiles...................................104 Figure 6.8 SEM photograph of a microdisk mesa immediately after the dry etch. The mesa top and sidewalls are still covered with polymers.....................105 Figure 6.9 SEM photos of smooth and vertical 2.3pm deep microdisk and microring mesas, etched with the optimized conditions.............................107 Figure 6.10 A photograph of a microring mesa taken with a high resolution FE- SEM. The sidewalls are very smooth with r.m.s. of less than lOnm. 107 Figure 6.11 A SEM picture showing the completed microdisk device. The bus waveguides are visible bellow the thin InP membrane left after the disk mesa formation......................... 108 xii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 6.12 A SEM picture showing the completed microdisk device. The coupling layer is completely etched, so there is no membrane left. The alignment mark is also visible to the right...................................................108 Figure 6.13 An example photograph of a polyimide layer covering the sample surface and the microdisk top contact layer is completely open and ready for bond-pad definition ..... .....................................................I ll Figure 6.14 Photographs of completed active microdisk devices. The bond-pad is far from the disk mesa improving the mechanical stability and decreasing the capacitance. On the right a cross-sectional view through the disk mesa is shown. The post and the bus waveguide are visible 112 Figure 6.15 Reflectivity as a function of the refractive index and the thickness of a single layer coating...................................................................................... 113 Figure 6.16 A schematic drawing of an active microdisk device mounted on a high-speed test mount. The mount has a 50 0 microwave strip-line and a 5 0 0 integrated resistor .... ..............................................................115 Figure 6.17 A picture of an active microdisk device mounted on a high-speed mount. The microwave strip-line is visible on the left and the terminal load is on the right.......................................................................................... 115 Figure 6.18 Test setup for microdisk characterization.................................................116 Figure 7.1 Schematic diagram of a vertically coupled microdisk resonator with a post a), and definition of the parameters of interest in this study: disk radius R; thickness of the coupling layer dc\ waveguide etch depth dw G ', thickness of the remaining thin membrane t................................................119 Figure 7.2 An example, TE, transmission characteristic of a microdisk with radius i?=12(tm. Sharp and deep dips are observed, with a quality factor in excess of 7,000. In the insert a magnified resonance around yl=1.55jim is shown with FWHM zU=0.22nm........................................... 121 Figure 7.3 The TE and TM resonance wavelengths are different by AF=5nm in the vicinity of /F=1.55fim, because of the different mode effective index. .......................................... ...........................................................122 Figure 7.4 Dependence of the quality factor, Q (a), and the power extraction efficiency (b) on the disk radius, R. The coupling separation between the disk and bus waveguides core layers is dc=Q.l\lm and 0.8|im...........128 xiii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 7.5 Distribution of the first two modes in 7?= 12pm disk. The contour line represents the cross-section of the disk cavity/core and the bus waveguide. The field is positioned very close to the semiconductor/air interface.............................................................................................................129 Figure 7.6 Effective indices of the first TE and TM disk modes as a function of disk radius, R. At small radii the difference in the effective indices between both polarizations becomes larger................................................129 Figure 7.7 The coupling coefficient k, as a function of the separation distance dc. The calculation is performed using a slab waveguide approximation and is discussed in [8]................................................. .131 Figure 7.8 Dependence of the quality factor, Q (a), and the power extraction efficiency (b) on the coupling distance, dc, between the disk and bus waveguides. The tested devices have radii of R=8 and 12jim...................132 Figure 7.9 Calculation of the effective indices and losses for the first TE and TM modes of the bus waveguide as a function of the waveguide etch depth, dwG........................................................................................ 133 Figure 7.10 Bus waveguide mode distribution. The bus cladding is lpm thick and the top InP membrane is 0.2pm. dwa= +0.5pm (left), and dwa= - 0.2pm(right)........................................................... .134 Figure 7.11 Calculation of the effective indices and losses for the first TE and TM modes of the bus waveguide as a function of the cladding layer thickness, ddad............................................................. 135 Figure 7.12 Bus waveguide mode distribution. The bus cladding, dcia ( i, and the top InP membrane are 0.2pm thick. d^G- +0pm. The mode leaks into the substrate................................................. 135 Figure 7.13 Dependence of the quality factor, Q (a), and the power extraction efficiency (b) on the waveguide etch depth, dwc• Tested devices have 7?= 12pm and dc=0.1 and 0.8pm......................................................................136 Figure 7.14 Calculation of the disk effective index and the loss as a function of the thickness of the remaining InP membrane, t. An increase of the loss is observed at large t, due to the more effective coupling of energy from the disk into the slab waveguide formed by the membrane.............. 138 Figure 7.15 Coupling of light from the microdisk cavity to the slab waveguide formed from the remaining InP membrane. The thickness of the membrane is f=0.5pm.......................................................................................139 xiv Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 7.16 Quality factor as a function of the disk radius, R, for different thickness of the remaining InP membrane, t. A decrease of Q is observed at large values of t due to the leakage of energy from the cavity into the membrane .................................................................. .139 Figure 8.1 Contours of equal power transmission T=\t\2 and quality factor Q derived with the CMT theory are plotted at resonance for R=5/im as a function of the power coupling coefficient and cavity loss. ..... 143 Figure 8.2 Example QCSE optimization of the QW structure for an active microdisk........................... 146 Figure 8.3 Loss trimming of the transmission response of a microdisk/waveguide coupler with radius i?=10pm. By applying -3V reverse bias the losses in the cavity increase by Aos=\6cml, which leads to change in the quality factor by zl<2=3200 and in the transmission by AT=QA5..............150 Figure 8.4 Dependence of the transmission and dropped coefficients on the loss in the microdisk cavity. The measured values of T at different bias are also plotted. ....... 152 Figure 8.5 Transmission coefficient as a function of the applied bias measured at different resonant wavelengths. The maximum change is achieved at zbb=29nm from the bandgap ....................................................................153 Figure 8.6 Loss trimming of the dropped response of a microdisk with radius R= 12pm. Larger contrast ratios Ci?>10dB could be achieved at the expense of larger insertion loss...................................................................... 154 Figure 8.7 Contours of equal transmission T and Q for a 5pm microdisk device with gain active region (g=-0t) (the changes in the refractive index are not included) ...................................................................................... 156 Figure 8.8 Material gain and index change as a function of the carrier density (a), and the calculated modal gain as a function of the applied voltage..........158 Figure 8.9 Contours of equal transmission T and Q for a 5pm microdisk device with gain active region (g=-d). (the changes in the refractive index are included)................ 158 xv Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 8.10 Gain trimming of the transmission response of a microdisk/waveguide coupler with radius i?=10|xm. By applying a dc current of 5 mA, the modal loss in the cavity decreases by A a = \lc m x , which leads to increase of the quality factor by zl<2=5300 and decrease of the transmission by AT=0.5...... 160 Figure 8.11 A calculation of the transmission and dropped coefficients as a function of the cavity modal gain/loss. The microdisk have radius R=10fim, coupling coefficient K= 3.6% and internal loss coefficient (X=5cml around A.=1552nm. The measured values of the transmission are also plotted with triangles. ...................................................................162 Figure 8.12 Microdisk filter with tunable bandwidth. The refractive index change due to the temperature tuning has the opposite sign and the same magnitude compared to the refractive index change due to the injected carriers, and thus the resonant wavelength stays constant......................... 164 Figure 8.13 Gain trimming of the dropped response of a microdisk/waveguide coupler with radius i?=10jiim. The dropped power is amplified above the bandgap energies and the dropped coefficient is larger that unity 165 Figure 8.14 Contours of equal power transmission T=\t\2 and quality factor Q derived with the CMT theory are plotted at resonance for R=5/um as a function of the power coupling coefficient and the injected current density................................................................................................................168 Figure 8.15 Microdisk tunable filter. The free carriers injected into the microdisk cavity change the modal refractive index, which blue-shifts the resonant frequency.................................... 171 Figure 8.16 Switching behavior of a microdisk tunable filter at )t=1598nm. Change in the drive current by /d/=200pA is enough to toggle the switch from OFF to ON state. The voltage change required for this transition is ziy=0.1V..................................................................... 172 Figure 8.17 Dropped port tuning of microdisk with radius i?=10fim. The device has a high quality factor Q of 7000............................................................... 173 Figure 8.18 Typical tuning characteristics of microdisk devices with gain (a) and FCI (b) active regions. The graphs show the achievable wavelength shift as a function of the wavelength (measured at the resonance), and the drive current is used as parameter....................... ....174 xvi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure A3.1 WGM in a microring cavity ......203 Figure A3.2 Conformal transformation method for calculating the modal field distribution........................... ...203 Figure A3.3 Example graphs from the solution of a microdisk problem. ..... 205 Figure A3.4 Coupling between curved waveguides..................... 208 Figure A3.5 Calculation of the model field distribution and the power-coupling coefficient........................................................................... 213 Figure A 4.1 CMT theory approach..................... 216 Figure A4.2 The effect of changing the cavity loss on the transmission and quality factor of a resonant coupler................ 220 Figure A4.3 The effect of changing the refractive index of the cavity on the transmission and quality factor of a resonant coupler................................ 221 Figure A5.1 Properties of the QCSE............... 231 x v ii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Abstract In this dissertation we will present novel WDM components based on multi section semiconductor optical amplifiers and ultra-small microdisk and microring resonant devices. In the first part of the dissertation, the design and fabrication of multi-section semiconductor optical amplifiers/equalizers with current-adjustable gain spectrum will be discussed. Two-section amplifiers were fabricated using selective area growth techniques to simultaneously define gain regions with different spectral properties with one growth run. Those devices showed a gain of 12dB and an output saturation power of lOdBm. A flat gain spectrum of more than lOOnm was obtained by varying the relative magnitude of the drive currents in both sections. In the second part, the design and fabrication of active semiconductor microdisk switches, filters, modulators and wavelength routers enabled by modulating the transfer characteristics of a resonant cavity are investigated. Microdisk devices are facet-free resonant cavities that can conveniently be coupled to bus waveguides to provide compact, high-spectral resolution filtering and routing capabilities to the DWDM systems.. In this project we demonstrated for the first time active, vertically- coupled semiconductor microdisk switches, filters, modulators and wavelength routers amenable to large-scale integration. xviii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 1 Introduction 1.1 Evolution of Fiber Optic Transmission Optical systems are today’s backbone of the information society due to the tremendous caring capacity of the light. The advantages of optical fiber over copper lines include low error rates, immunity to electrical interference, security, and lightweight. The development of fiber optics is closely tied to the use of the specific regions of the optical spectrum where optical attenuation is low. These regions, called windows, lie between areas of high absorption. The earliest systems were developed to operate around 850 nm, the first window in silica-based optical fiber. A second window (S band), at 1310 nm, soon proved to be superior because of its lower attenuation and zero dispersion, followed by a third window (C band) at 1550 nm with an even lower optical loss. Today, a fourth window (L band) near 1625 nm is under development and early deployment. Transmitters Receivers Separating signals Combining signals Transmission on fiber Figure 1.1 A schematic of a simple WDM transmission system. 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Systems in use today are called dense wavelength division multiplexed (DWDM), where multiple information channels/wavelengths are propagated simultaneously into a single mode fiber, see Figure 1.1. Today’s DWDM systems are capable of supporting 64 to 160 parallel channels, densely packed at 50 or even 25 GHz intervals. A simple DWDM system includes: • Generation of the signal—the source laser, must provide stable light within a specific, narrow bandwidth • Combining the signals— modem DWDM systems employ multiplexers to combine the signals. • Transmitting the signals— the combined channels are transmitted through a low-loss single mode fiber. • Separating the received signals— at the receiving end, the multiplexed signals must be separated out by a demultiplexer. • Receiving the signals—the demultiplexed signal is received by a photodetector. Owing to attenuation in the optical link, there are limits to how long a fiber segment can propagate a signal with integrity before it has to be regenerated. Before the arrival of optical amplifiers (OAs), there had to be an electronic repeater for every wavelength transmitted into the system. The OA has made it possible to amplify all the wavelengths at once and without optical-electrical-optical (OEO) conversion. An invention of a flat-gain optical amplifier, coupled in line with the 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. transmitting fiber to boost the optical signal, will dramatically increase the viability of DWDM systems by greatly extending the transmission distance. Because DWDM systems send signals from several sources over a single fiber, they must include some means to combine the incoming signals, Figure 1.2(a). This is done with a multiplexer, which takes optical wavelengths from multiple fibers and converges them into one beam. At the receiving end the system must be able to separate out the components of the light so that they can be discreetly detected. Demultiplexers perform this function by separating the received beam into its wavelength components and coupling them to individual fibers. Demultiplexing must be done before the light is detected, because photodetectors are typically broadband devices that cannot selectively detect a single wavelength. Fiber Fiber Amp. Amp. OADM MUX/DEMUX MUX/DEMUX (b) (a) Figure 1.2 (a) Multiplexing and demultiplexing in a bidirectional system; (b) OADM selectively removing and adding wavelengths Between multiplexing and demultiplexing points in a DWDM system, there is an area in which multiple wavelengths exist. It is often desirable to be able to remove or insert one or more wavelengths/channels at some point along this span. An optical add/drop multiplexer (OADM) performs this function, as shown in Figure 1.2(b). Rather than combining or separating all wavelengths, the OADM can remove some 3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. while passing others on. OADMs are a key part of moving toward the goal of all- optical networks. The task we have with this dissertation is to demonstrate novel components that can be implemented into future DWDM systems. First, a flat-gain semiconductor optical amplifier/equalizer will be designed and demonstrated. The device will exhibit a current-adjustable gain spectrum, and will be capable of equally amplifying all of the DWDM channels over span of more than lOOnm. It was also designed to be part of an optical analog-to-digital converter system. Second, novel ultra-small active microdisk wavelength selective elements will be designed and demonstrated. These components will play an important role in future photonic circuitry, as multiplexers, demultiplexers, and add-drop filters. Moreover, the devices will be active and amenable to large-scale integration, which will allow tuning, detection and switching capabilities to be monolithically integrated into the systems on a single photonic chip. 1.2 Semiconductor optical equalizer Semiconductor optical amplifiers (SOAs) amplify incident light through stimulated emission, the same mechanism used by lasers. Indeed, an optical amplifier is nothing but a laser without feedback. The end facets of the amplifier must have a reflectivity R, much smaller than in the case of a standard laser. If the single pass gain of the optical medium is G, then the lasing, threshold condition can be expressed as G*R~1. Thus, in order to utilize the high gain available in the 4 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. semiconductor media, the reflectivity should be reduced to less than 1 C T 4. The optical gain is realized when the amplifier is pumped (optically or electrically) to achieve population inversion. The small size, high gain and the electrical pumping available in SO As make them attractive for signal boosting in long-hall systems, as well as for many applications involving optoelectronic integration. SO As allow for direct and simultaneous amplification of all of the channels in DWDM systems, avoiding the optical-to-electrical regeneration, and thus reducing the overall cost of the systems. The main parameters describing a SOA are chip gain, gain flatness, gain bandwidth, polarization sensitivity, noise, and saturation intensity. The number of channels amplified in a DWDM system is closely related to the gain flatness and bandwidth. The gain flatness is an important parameter in that all channels need to be amplified by the same amount in order to simultaneously keep their level higher than the noise floor, and avoid data loss in a long-haul DWDM system. Therefore, a wide-band, spectrally-flat, high-gain SOA is needed to meet the demands of future systems with large number of channels. In this dissertation we will demonstrate such an amplifier, having two-sections with different spectral properties. The independent control of the biasing of these sections will allow us to shape the overall device spectrum, thus achieving an equalized CW gain over a large bandwidth. The equalizer is also designed as a component of a particular system - an optical analog-to-digital converter [1], based on time stretching ideas to digitize a very fast electrical signal. The idea is to use the enormous frequency chirp available in an 5 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. optical system and employ the dispersion of an optical fiber to time stretch a short Gaussian pulse. A schematic of such a system is shown in Figure 1.3. A very short, femtosecond pulse is first obtained from a mode-locked laser (MLL). As known from Fourier transform theory, the duration and the frequency bandwidth of a transform-limited Gaussian pulse are related by an ‘uncertainty’ relationship Am At ~ , and thus the shorter the pulse the larger the frequency bandwidth (more than lOTHz in this case). Next in this system, the pulse is propagated through a very long dispersive fiber. The initial linear chirp (different frequency components arrive at different time), and the dispersion of the fiber, (different frequency components travel with different velocities) result in time stretching of the input pulse and convert it from a femtosecond into a nanosecond time range. Note that the frequency bandwidth is still the same, while the pulse duration is increased 3-4 orders of magnitude. This transformation is possible only if the pulse chirp is increased by the same amount. MLL Tim e Stretching > QBLjw Time Stretching MO® OEM SOA-EQ t, C O t, C O t, C O Figure 1.3 A schematic diagram of an optical analog-to-digital converter system. The optical equalizer is used to flatten a highly chirped Gaussian input pulse. 6 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. After the time-stretching, the pulse is a highly chirped Gaussian, so that at different moments different instantaneous frequencies are present. This fact is indicated by the different shading in the figure above. The task of the device we are designing (a spectral equalizer, SOA-EQ) is to convert the highly chirped Gaussian pulse into a flat, spectrally- and temporally-equalized output (see Figure 1.3). Further within the system, this flat super-continuum is used as a carrier and a modulator (MOD) is employed to modulate the amplitude of the carrier with the signal to be converted. Additional time-stretching is used to further stretch the carrier and a conventional demultiplexer (DEM) or slow analog-to-digital converter can be employed to detect the signal of interest. In this configuration, the equalizer works in a quasi-CW regime, with the pulse duration being much longer than the carrier relaxation time. The saturation properties of the amplifier are also exploited. 1.3 Semiconductor microdisk wavelength selective elements The requirement of large-scale integration of optical devices imposes certain constraints on device dimensions. If we compare today’s edge emitting lasers (typical length of 300pm) with VCSEL’s (typical size of 50pm2 ), an improvement in the performance and decrease in the threshold currents and dissipated power can be seen. Dielectric microcavities are versatile elements for integrated optics and may serve as building blocks for future photonic integrated circuits (PICs). Photonic bandgap (PEG) lasers have been demonstrated [2], and the PEG cavity may serve as 7 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. an ultimate example for device scaling, shrinking the cavity dimensions to a volume on the order of X3. Microring/microdisk resonators are other candidates for VLSI PICs as they provide a wide range of signal processing functions. Many practical devices incorporating microrings have been demonstrated, including waveguide- coupled GaAs lasers [3], passive channel dropping filters [4], passive WDM demultiplexers [5], and switches [6], Fabricated devices have been reported to have ring radii ranging from 4 to 30pm. Such small dimensions could lead to device integration densities of up to 104 devices per square centimeter. Ring or disk resonators support traveling wave resonant modes called whispering gallery modes. By side or vertical coupling to a bus waveguide (Figure 1.4), a single ring may completely extract a particular wavelength Figure 1.4 Microdisk switch. and offer superior performance compared to the standing wave resonators. Thus, this device may serve as a microdisk switch for that wavelength. If we incorporate a gain active region in the disk cavity, the device from Figure 1.4 could be utilized as a microdisk laser coupling light into the bus waveguide. In the ideal case, if we couple a second waveguide to the disk resonator, then the light dropped out from that waveguide is complimentary to the light transmitted out of the first one, and thus the device depicted in Figure 1.5 is a WDM channel- dropping filter. Figure 1.6 shows a schematic of a demultiplexer built from microdisk resonators. If they are made small enough, so that the free spectral range 8 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (FSR) of a single disk is larger than the WDM communication window, the task of dropping only one channel per filter could be accomplished. The erbium-doped amplifiers support a communication window of 30nm, so the required size of the microdisks is less than 5p.m. Figure 1.7 shows a microdisk used in a Mach-Zender interferometer. In this configuration the microdisk is in the overcoupled regime of operation, and thus only introduces a # phase shift at resonance on the side-coupled waveguide, imbalancing the interferometer. Using this configuration, a notch filter has been reported [7] with a 100pm long, 50° splitting ratio interferometer, thus still keeping the overall device size small. Figure 1.8 shows a second order microdisk filter with improved transfer function, where the Lorenzian lineshape characteristic for a single device, is transformed into a box-like shape [8], Figure 1.5 Channel- dropping filter. Figure 1.6 WDM demultiplexer. o Figure 1.7 Mach-Zender interferometer/ notch filter. Figure 1.8 Second Order add-drop filter 9 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Passive devices such as the ones described above have many limitations when considered for incorporation into large-scale systems using a fixed manifold of wavelengths. Their inherent wavelength selectivity places unrealistic demands on fabrication uniformity and reproducibility and their temperature sensitivity suggests that a static system design would be thermally unstable. For these reasons alone, it is interesting to consider active resonant devices that can be configured as tunable elements. Active devices also enable switching and tuning elements to be incorporated into optical circuits that greatly increase the functionality of such circuits. Furthermore, with slight modifications of the device design one can easily envision that the same basic structure can be used to incorporate tunable lasers, detectors and modulators into the system, Figure 1.9. In view of the potential of these elements in WDM systems, it is surprising that so little attention has been paid to analyzing their performance. In this dissertation we will analyze the operation and design of active semiconductor microresonator devices to determine the most promising mechanisms to incorporate into the resonator to achieve the desired tunability and switching characteristics. Our design perspective is to focus on those mechanisms that maintain high system throughput and require low electrical power. Furthermore, we will incorporate them into real active microdisk devices and investigate experimentally their performance. There are two main configurations utilizing the coupling between the disk and bus waveguides. The first approach uses lateral geometry Figure 1.10(a), with waveguide/resonator separations on the order of 0.1pm fabricated by high resolution 10 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. lithography. This small separation is a result of the high index contrast, An>2, at the semiconductor/air interface and in this design the coupling coefficient is very difficult to control due to fabrication nonuniformities. The second approach uses a vertical geometry [9] and has the advantage of precise control of the coupling coefficient by epitaxial growth (Figure 1.10b). Another advantage is that waveguides and resonator can be grown with different material compositions. This facilitates the design of active microdisk devices - ON/OFF switches, modulators and microdisk lasers. Tunable laser / Power monitor / Modulator / Tunable source / and equalizer. / / multiplexer. ' t / Figure 1.9 A vision for a future WDM PIC utilizing active microdisk devices. d a ^ ~ 0 .1 u m Waveguides esonant Disk Transfer Substrate a) L ateral C oupling b ) V ertical C oupling Figure 1.10 Microdisk resonant coupler realized via lateral (a), and vertical geometry (b). The precise simulation of an active microdisk device is a rather challenging task. This includes solving for the resonant modal field distribution and for the coupling between the disk and the waveguides under given pumping conditions, and at the 11 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. same time accounting for all of the properties of the active media. The finite difference time domain (FDTD) method solves Maxwell's equations exactly but it is not intuitive, and is computationally intensive and time consuming. It is also difficult to incorporate gain/absorption in this model. It can be used, however, for the simulation of E-M fields in the final device. The conformal transformation approach for finding the modal fields is rather fast, with reasonable accuracy and can be used to solve parts of the problem such as the optical coupling coefficient between the resonator and the waveguides busses. In this thesis, we will analyze the effects of incorporating different active regions inside a semiconductor microdisk resonant cavity and will find certain design constraints and material parameters leading to optimal device performance. Gain and electroabsorption (EA) regions are considered as means to tune the resonator and to affect its losses (and therefore its quality factor Q, Figure 1.11(a)). Free carrier (FC) injection/absorption and electrooptic effects are also investigated as means to tune the resonator/switch, Figure 1.11(b). The losses that accompany the FC-induced index changes in this case are detrimental but small. In the case where practicality of the modulation mechanism is strongly dependent on the specifics of the active region design (electroabsorption), we have also included an optimization procedure for the active region. The coupling of modes in time (CMT) theory [4] is used as a main theoretical approach to predict the overall device response because of its simplicity and speed. This approach does not distinguish between polarizations (CMT looks at the devices as lumped circuits). The 12 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. effects of polarization can be calculated, however, by utilizing the appropriate coupling coefficients and effective refractive indices for each polarization and modeling them separately. The distributions of the fields are calculated by combining the effective index approach in the vertical (growth) direction with the conformal transformation in the radial direction. Coupling of modes in space theory (CMS) for bent waveguides is used to calculate the overall coupling coefficient, with the phase mismatch due to waveguide bending included. The Runge-Kutta method is applied for numerically solving the CMS equations (see Appendix 3 for details). Change of resonance Change of absorption - 4-10 " 2-10 (w-woywo 1 0.5 0 (w-wo)/wo (a) (b) Figure 1.11 Modulation behavior of a microdisk resonator via change of the absorption coefficient (a), and change of the resonant frequency (b). We will show examples of different active devices to emphasize the important role that we believe the microdisk will play in future photonic circuitry. As part of this dissertation, we would like to use a microdisk as a key element for building a new type of modulator. The use of a resonator to enhance the electroabsorption, electrorefraction, and the electro-optic effects in semiconductors has the potential to 13 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. reduce the drive voltage levels to less than IV, while maintaining high bandwidths of more than 40GHz. There are three limiting factors that determine the frequency bandwidth (the response time) of an active microdisk device: (i) the parasitic capacitance from the bond-pads and the junction capacitance; (ii) the quality factor of the cavity (the energy builds up and decays with a certain time constant); (iii) the modulating mechanism itself. 20 30 - - C-lim ited A f Q -lim ited Af • 2 5 - 2 0 - < / > 10- 5 - 0 10 20 30 40 50 60 70 80 90 100 O S_ o ■ t — i O c d LL t cd 3 o Switching Frequency Bandwidth [ GHz ] Figure 1.12 The frequency bandwidth is limited by the RC time constant (disk size) and the quality factor of the cavity. Figure 1.12 shows a plot of the bandwidth, A f imposing restrictions on the maximum allowed junction capacitance and loaded Q. Assuming a load resistance of Ri=50Q and intrinsic layer thickness of d;=0.45pm (a typical value for these devices - Af scales linearly with dt ), a bandwidth of 40GHz limits the disk radius to less than 10pm. Because this is based purely on area considerations, Af can be increased by 14 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. not covering the whole disk with an electrode (as the optical field is concentrated close to the disk edge) or by the use of a microring. Certain semiconductor devices, such as add-drop filters and switches, require both large free spectral range (FSR) and high Q. Large FSR can be obtained by decreasing the cavity dimensions. For example a cavity with f?<5pm has a FSR>5Onm, which is larger than the communication window defined by the Er- doped amplifier. On the other hand, increasing the loaded Q leads naturally to a decrease of the bandwidth/speed of the device. Figure 1.12 shows that a Q of 5000 limits the bandwidth to 40GHz. The maximum available value of Q is determined by the bending radiation loss [10], and it is less than 1% per roundtrip for diameters greater than lpm for high index contrast semiconductor devices. Therefore, for any practical devices with radii less than 10pm, A f is limited by the loaded quality factor of the cavity. The bandwidth may also be restricted by the modulation mechanism. For example, an active region employing EA limits the device speed much less than an active region with gain or free carrier injection modulation. In the latter cases, the minority carrier lifetime of T~lns constrains the bandwidth to ri/<lG Hz. We will try to shorten the carrier lifetime by inducing defects in the resonator. By ion implantation, for example, it may be possible to increase the modulator’s speed up to 20GHz. Further investigation in this direction is necessary. 15 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1.4 Outline This dissertation is organized as follows. Chapter 2 describes the theoretical model used for simulation of the optical equalizer. Chapter 3 follows all of the design considerations. Chapter 4 presents the fabrication process of the SOA/EQ and the results are shown in Chapter 5. Chapter 6 describes all of the details on the fabrication of the miniature microdisk devices. Chapter 7 and Chapter 8 present the theory and the results from the measurements of passive and active microdisk devices respectively. Chapter 9 is the summary. The appendices include detailed information on the fabrication process of the SOA; wafer-to-wafer bonding procedure; coupling of modes in time theory for simulating the response of the coupled microdisk/waveguides system; description of the formalism behind finding the modal field distribution and the coupling coefficient of the microdisk cavity; and derives the properties of the quantum confined stark effect. 1.5 References 1 F.Coppinger, A.S Bhushan, and BJalali, “Time Magnification of Electrical Signals Using Chirped Optical Pulses”, Electronics Letters, vol. 34, no. 4, pp. 399-400, February 1998 2 O.Painter, R.K.Lee, A.Yariv, A.Scherer, J.D.O’Brien, P.D.Dapkus, I.Kim, “Two Dimentional Photonic Crystal Defect Laser”, Science, 284, pp.1819-1821 (1999) 3 T.Krauss and P.Layboum, “Monolithic integration of a semiconductor ring laser and a monitoring photodetector”, SPIE, vol.1583, pp.150-152, 1991. 4 B.E.Little, S.T.Chu, ELA.Haus, J.Foresi, and J.P.Lain, “Microring resonator channel dropping filters,”, J.Lightwave Technology, vol.l5,pp.998-1005, 1997. 16 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5 S.T.Chu, B.E.Little, W.Pan, T.Kaneko, S.Sato and Y.Kokubum, “An eight-channel add- drop filter using vertically coupled microring resonators over a cross grid”, IEEE Photon. Technol. Lett, vol.ll, pp.691-693, 1999 6 B.E.Little, H.A.Haus, J.S.Foresi, L.C.Kimerling, E.P.Ippen, and DJ.Ripin, “Wavelength switching and routing using absorption and resonance”, IEEE Phot. Technol. Lett., vol. 10, pp.994-996, 1998 7 P.P.Absil, J.V.Hryniewicz, B.E.Littlr, R.A.Wilson, L.G.Joneckis, and P.T.Ho, “Compact Microring Notch Filters”, IEEE Phot. Technol. Lett., vol. 12, no.4,pp.398-4Q0, 2000 8 J.V. Hryniewicz, P.P. Absil, B.E. Little, R.A. Wilson, and P.T. Ho, “Higher Order Filter Response in Coupled Microring Resonators”, IEEE Phot. Technol. Lett., vol. 12, no.3, March 2000, pp.320-322. 9 D.V. Tishinin, P.O. Dapkus, A.E. Bond, I. Kim, C.K. Lin, and J.O'Brien, “Vertical resonant couplers with precise coupling efficiency control fabricated by wafer bonding,” IEEE Phot. Technology Lett., vol. 11, pp. 1003 -1005, August 1999. 10 E.A.J. Marcatili, “Bends in optical dielectric guides,” Bell Syst. Tech. /., vol.48, pp.2103- 2132, 1969 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 2 Theory of pulse propagation in a SOA 2.1 Theoretical background The theory of pulse propagation in amplifiers is well known [1]~[12] and usually treats the semiconductor media as a collection of noninteracting two-level systems with transition energies extending over the whole range of conduction and valence bands. Considerable simplification occurs if the pulse width, T p, is assumed to be much larger than the intraband relaxation time, Tin = 0. I p s , that governs the dynamics of induced polarization. In the rate-equation approach, the medium response to the optical field E is described by the carrier-density rate equation [1]: current, q is the electron charge, V is the active volume, R(N) is the spontaneous emission and nonradiative recombination rates, ha> 0 is the photon energy and g(N) is the gain of the media. The propagation of the electromagnetic field inside the amplifier is governed by the wave equation: where c is the velocity of the light. The dielectric constant e is given by: £ = nl +% , where the background refractive index nb is generally a function of the transverse I - - R ( N ) - g ( N ) |Ef (2. 1) dt qV hO )0 ’ where N is the carrier density [cm-3], D is the diffusion coefficient, I is the injection (2.2) 18 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. coordinates x and y to account for the dielectric waveguiding in semiconductor laser amplifiers. The susceptibility % represents the contribution of the charged carriers inside the active region of the amplifier and is a function of the carrier density N. The exact dependence of the % on N is quite complicated as it depends on the details of the band structure. Often a simple phenomenological linear model has been used [1] to describe this dependence: where n is the effective mode index. The carrier induced index change An, responsible for self phase modulation (SPM), is accounted for through the parameter a , which is called the linewidth enhancement factor [2], [3]. We have chosen a different approach and used the k.p formalism to calculate the bandstructure of the semiconductor media and account for these processes [4], Equations (2.1) to (2.3) provide an exact description of the pulse propagation in a semiconductor optical amplifier (SOA). In practice we make a few approximations. First, we assume that the traveling wave amplifier supports only noninteracting modes. Assuming that the input light is linearly polarized and remains linearly polarized during the propagation, the electric field inside the amplifier can be written: Z {N) = - — {a+i)g{N) = - i — g{N)+An2, (2.3) (2.4) 19 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. where x is the polarization unit vector. S(x,y) is the transverse waveguide mode distribution, k0 = my0/ c , and A(z,t) is the slowly-varying envelope associated with the optical pulse. If we substitute Eqn. (2.4) into (2.2), neglect the second derivatives of A(z,t) with respect to z and t and integrate over the transverse dimensions, we obtain: 2 n S d 2S d 2S / 2 2 ) ^ 0 c _ n &2 ^ . (2.5) dA 1 dA ico0r A 1 + _ oz vg at 2nc 2 where the group velocity is defined as vg = c/ng , the group index as = n + 6 )0 (dn/dty) and the confinement factor is: r= f f > M W £ l \ s ^ y f dxdy ’ (2-6 ) The solution to the first equation in (2.5) provides the transverse mode distribution S(x,y) and the effective mode index n. The second equation in (2.5) governs the evolution of the pulse amplitude along the amplifier length. The last term takes into account the decay of the mode amplitude due to the internal loss a i D t of the semiconductor media. The group velocity dispersion, d 2 n/dco2 , is neglected hereafter, since its effect on pulse propagation for our case is negligible. Also this assumption greatly simplifies the calculations and if needed it can be included by adding a term containing d2A/d t 2 on the left side of the second equation. 20 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. We can further simplify these equations by neglecting the carrier diffusion term in the rate equation, since the carrier density is nearly uniform along the transverse directions and we can use an average value. This assumption is equivalent of neglecting the spatial hole burning process. By averaging and integrating (2.1) over the active-region transverse dimensions, we obtain: dN I ~ |a|2 = ~ - - R ( N ) - )-L L , (2.7) dt qL hd) 0 where N is the carrier density per unit length [cm4 ], R(N) is the total nonradiative rate per unit length [cm • ‘xs'1 ] , and A is normalized such as |A|2 represents the optical power. The quantity F = \A \2 /h.6 )0 is the total photon flux out of the amplifier [#/s]. Next, we make use of the transformation r = t - z / v g , where the reduced time fis measured in a reference frame moving with the pulse. Furthermore, we separate the amplitude and phase of the pulse by using A = -fp exp(i0), where P(z, r) and (f(z, r) are the power and the phase of the slowly varying envelope. Combining all of the above equations and assuming multimode propagation we can finally obtain the following set of equations (2.8). ^ L = I & - R ( N ) - r ' £ g i (N)Fi d r qL Y = (rgx{N)- am t )Fa + jRsp(N) (2.8) ^ L = r ^ A n A{N) dz A 21 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This is a set of coupled differential equations for the carrier density N (per unit length), and for the total photon flux F x of the different propagation modes with given wavelength, X. The last term in the first equation couples the carrier density with the photon flux and the flux itself is coupled to the carrier density by the gain function, gA{N), and the spontaneous emission rate, R ( N ) . The second equation describes the change of the number of photons per unit length due to the presence of gain/loss in the media, and due to the fraction, % of the spontaneous emission, which couples to the particular mode. The time variation of the carrier density leads to temporal and spatial modulation of the phase by Anx (n ) (third equation), i.e. the pulse modulates its own phase through the material gain as a result of the coupling between the equations. This process is called self-phase modulation (SPM). 2.2 Numerical solution Now we will solve Eqns.(2.8) numerically and for this purpose they were discretized in the following manner: = ■ V - R(N,) - r £ g, {N, )F / (a) Af qL Y r / ‘ = (rg ^ N j- a jF ,1 +&JN,) (b) (2.9) Az ~ — = F - ~ r s ~ {N,) (c) Az X, Figure 2.1 shows the computational grid of the discretized equations. First, the solution of the problem requires the knowledge of the boundary conditions. In our 22 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. case we have launched a light pulse into the amplifier, so we know the initial values of the photon flux Fa @ z=0, i.e. Fn, Fn, F1 3 , Fm^Fin- The other boundary conditions (BCs) require the knowledge of the carrier density before the arrival of the pulse, N @ t=-oc , i.e. Nu, N21, N3 1 , N4 1 . These can be obtained from the steady state condition, when the amplifier is biased with a certain current and the photon flux is negligible. The carrier density, Njj, is constant and can be found from: ^ = ! f - R { N ) - r Z g l(N)Fl =o (2.10) oT cjL ^ r F 13 n j V w- 1 2 / (V y K " 0 r $ 2 r n9 Figure 2.1 Computational grid used for solving the problem of pulse propagation The complete procedure for the solution of Eqns. (2.9) is the following: 1. Solve with respect to v. We know Fn and Nu from the BCs, so Nn can be calculated (2.9a). Fn is known from the BCs, so N I3 can be calculated (2.9a).... Following this procedure we can calculate all the variables in the first column in Figure 2.1. 23 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2. Solve with respect to z and T . From Fn and Nn we determine F21 (2.9b). N21 is known from the BCs, so N22 can be calculated from F21 and N2i (2.9a). From F12 and N]2 we determine F22 (2.9b), so N23 can be calculated from F22 and N22 (2.9a) and so on. Following this procedure we can calculate all the variables in the second row in Figure 2.1. 3. Repeat step 2 until the grid is covered and the end of the amplifier is reached. For this particular application we would like to solve for relatively long (~ns range) chirped pulses with a broad spectrum. Therefore, a liner gain-carrier density relationship, often used in the literature to describe the properties of the semiconductor gain media, is not applicable. Second, we would like to take into account the spectral properties of the semiconductor gain material, which is difficult to accomplish in time domain calculations. Both of these tasks require a ‘realistic’ gain model, the one obtain by using the Luttinger-Kohn k.p formalizm [4] for the conduction and valence band electron states. We used a software package by Crosslight Inc., LASTIP, to calculate the material gain spectrum and recombination rates for different carrier density concentrations and different QW thickness. This software is also capable of calculating the change of the refractive index induced by the change in the gain, which is related via the Kramers-Kroning equations. With LASTIP we produced large matrices containing all of the data we need: (i) the material gain spectrum in the wavelength range from A=1.2jim to A=1.6jjm, for carrier densities from N=1018 to N=1019 and QW thicknesses from d=30A to d=120A, Figure 2.2(a); (ii) the total recombination rate within the same limits; (iii) 24 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the spontaneous emission rate; (iv) the change of the refractive index induced by the change in the gain, Figure 2.2(b). All of these matrices were loaded by the main program calculating the properties of the optical amplifier and an effective table look-up and interpolation procedure was writen to use the data in the calculations. As a result we have the following set of continuous functions: R(N,d) - the total recombination rate g(N,l,d ) - t h e material gain R (N,d ) - spontaneous emission rate An(N, X,d)~ refractive index change (2.11) 0)2000 — 2000 0-4000 (6 0.06 0.04 0.02 0.0 0 - -0 .0 2 - 0.04 3 -0.06 -0.08 - 0. 10 -0 .1 2 - 0.14 1.3 1.4 1.5 1.6 0.75 0.80 0.85 0.90 0.95 1.00 1.05 Wavelength [pm] Photon energy [eV] (a) (b) Figure 2.2 Calculated gain spectra for different carrier density (a), and the corresponding change of the refractive index (b) for an InGaAsP QW. Furthermore, because the propagated pulses are higly chirped and relatively long, we can define an instantaneous (local) frequency dependent on the phase of the pulse envelope, Av = ■ 1 9 # 2k dr , and the total frequency is equal to v — v0 + A f . 25 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2.3 Results and discussion Before applying the theoretical model described above to the problem under investigation, namely, propagation of higly chirped pulses in a travelling wave semiconductor optical amplifier, we would like to understand the general properies of a semiconductor gain media. If the amplification is a linear process, the amplified pulse would be a replica of the input pulse as long as the amplifier bandwidth exceeds the spectral width of the input pulse. In practice, gain-saturation-induced nonlinearities lead to pulse distortion in any amplifier simply because the leading pulse edge saturates the amplifier and reduces the gain available for the trailing edge [5], [6], [7], [8], [9], [10], [11], [12]. Nonetheless, the input pulse can be amplified without significant distortion if the pulse energy is a small fraction of the saturation energy. Even when the pulse shape remains nearly unchanged during the amplification process, the pulse spectrum can be distorted considerably if the refractive index becomes nonlinear [10], This is particularly important for semiconductor laser amplifiers, where the changes in the carrier density occurring as a result of gain saturation invariably leads to relatively large changes in the refractive index. The physical mechanism behind the spectral shift and distortion is the self phase modulation (SPM) occurring as a result of index nonlinearities induced by gain saturation. In this chapter we will follow closely ref.[13] and will consider three boundary cases - CW operation, short pulse propagation and quazi-CW. 26 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2.3.1 Results: CW operation Let’s first consider CW operation of a traveling wave semiconductor optical amplifier (TWSOA). The device under investigation is a L=300jum long MQW InGaAsP device with four 9QA quantum wells. The composition of the quaternary layer is chosen in such a way, that the gain peak is centered around A.=1.55fJm. As we mentioned before, in these calculations we have included a more ‘realistic’ material gain model (as shown in Figure 2.2) which allows a more careful investigation of the spectral properties of the TWSOA. Figure 2.3(a) plots the saturation behavior of this device with an estimated input saturation power of Psat=0dBm. First, consider that three different modes are propagated separately (solid lines on the graph) with Aj=1.51jum, 2 , 2= 1 .5 O jum , 2,3=1.48/m , the third one being amplified less because it is far from the gain peak. Simultaneous propagation (dashed lines) reveals two important properties of the TWSOA. • modes at shorter wavelengths saturate faster (lower Psat). • there is cross-talk between the channels. The cross-talk can be seen from the fact that the saturation energy of a particular mode depends on the presence of the other modes and its value decreases from the one when the mode propagates independently. The cross talk is due to the fact that the amplified photons from the different modes deplete the same carrier reservoir, i.e. the differential equations for the photon flux are coupled to each other by the carrier density equation. 27 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1=100 mA, dq „=9 nm, L=300 jim d,w =9 nm> L=300 iim X = 1.51 n m sim ultaneous propagation X =1.50 in m 1=100 mA l=50 mi 1=20 mi 0 1 0 1=10 mA 1.30 1.35 1.40 1.45 1.50 1.55 1.60 -40 -30 -20 -10 20 Input Power [dBm] W avelength [nm] (a) (b) Figure 2.3 Theoretical simulation of the saturation behavior of a TWSOA (a), and modal gain spectra for different pumping currents (b). Figure 2.3(b) shows the modal (device) gain spectrum for different pumping currents, assuming 100% pumping efficiency. The increase of the pumping level (increase of the steady state carrier density) increases the separation between the quasi-Fermi levels and thus the material gain. The second peak formed around A^lAjUm comes from the second bound state, which populates at high bias. 2.3.2 Results: Pulse propagation In the case of an unchirped (transform-limited) Gaussian input pulse Figure 2.4(a) shows that the amplified output pulse becomes asymmetric, such that its leading edge is sharper compared with the trailing edge. Sharpening of the leading edge is a common feature of all amplifiers [5]-[10] and occurs because the leading edge experiences larger gain than the trailing edge. The carrier density (and gain) saturates with the pulse arrival (Figure 2.4(c)), and because the carrier relaxation time is longer then the pulse duration, N does not have time to recover for the rest of the pulse and has a value lower than its initial value. The frequency chirp, shown in 28 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 2.4(d), can be defined as: Av(t) =——- and is a result of the SPM 2K a t process. The value of Ai/0M{t) is negative across the entire pulse, i.e. the instantaneous frequency is downshifted (red shift) from the incident frequency v0 = co0/27U . The temporal variation of the chirp is almost identical to that of the output pulse shape (compare with Figure 2.4(a)). This relationship is due to the fact that the refractive index change Anx (./V), contributing to the change in the phase, is proportional to the photon density. 1.0 input output 0.8 C O 0.6 0.4 £ 0.4 0.2 0.2 0.0 0.0 5 Normalized Frequency A v*x0 Normalized Time (U % J o .o 'J? -0 .1 * > - 0.2 & -0.3 5- -0.4 8 - ° - 5 -D -0 .6 g -0.7 = -0.8 E -0.9 O -1.0 z ,1t -1 .2 g 7x10' >! 6x101 2x101 1x10' •3-2-10 1 2 3 Normalized Time (t/x0 ) 4 5 -5 -4 Normalized Time (t/t^ Figure 2.4 Effect of gain saturation and SPM on pulse shape and spectrum. The induced chirp has the shape of the output pulse. 29 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The structure in the pulse spectrum (Figure 2.4(b)) results from an interference phenomenon. Physically, the instantaneous frequency is the same at two distinct points within the pulse profile. Depending on the relative phases of the optical fields at those two points, they can interfere destructively or constructively. This interference leads to an oscillatory structure in the pulse spectrum. The asymmetry is a direct consequence of the asymmetric shape of the output pulse. A noteworthy feature of the chirp profile shown in Figure 2.4(d) is that the chirp increases almost linearly over the central part ( |r| < f 0) of the pulse. Such a linear chirp implies that the pulse can be compressed or stretched while propagating in dispersive media with normal or abnormal dispersion, respectively. 2.3.3 Results: Pulse - effect of gain recovery The results of the previous section are obtained by assuming that the input pulse is much shorter than the carrier lifetime (T 0 « T C). When the pulsewidth becomes comparable to T c, the saturated gain recovers. Such a partial gain recovery would affect both the shape and the spectrum of output pulses. Figure 2.5 shows the effect of gain recovery on the output pulse shape when the input shape is Gaussian- dotted line. For durations of t 0 = 5p s , the pulse shape is nearly identical to that obtained by assuming an infinite carrier lifetime (no gain recovery). The leading edge is sharp, and the trailing edge is long, experiencing less gain due to the gain saturation - dash- dotted line. The lack of gain recovery is shown on the left inset in the same figure, where the carrier density N vs. time is plotted. When the relative time corresponds to 30 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the arriving of the pulse, (t - T 0 )/r0 = -3 , the carrier density (and thus the gain) saturates and remains low for the remaining part of the pulse having no time to recover. By increasing the input pulse duration to r0 = 500ps (dashed line), the output pulse becomes broader as the gain recovers partially during its trailing edge. The net result of gain recovery (note that it is still saturated) is that the output pulse is less asymmetric and becomes broader than the input pulse. These features become even more pronounced for T 0 = 5ns in Figure 2.5 (solid line). N qw =4,1=100 mA, L=1 mm, d =9 nm, X=1.5 pm, EIn=5 pJ 3.0 2.5 ; S S 0.8 R elative Time « 0.6 ■u 0.4 m S 0.2 Input z 0.0 Relative time (t-T0 )/t( Figure 2.5 Effect of gain recovery on input Gaussian pulses with different durations r0 The output pulse for this value of T o is much broader than the input pulse. In all of the cases mentioned above, the gain recovery mainly affects the trailing edge of the pulse - the leading edges for all examples start approximately at the same time and have the same slope, dependent only on the value of the input intensity. The right inset shows the carrier density (gain) recovery when r0 = 5ns. 31 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In Figure 2.6(a) the pulse evolution and shaping mechanism is shown as the input Gaussian pulse propagates along the amplifier. We observe that the pulse shape (and spectrum, accordingly) becomes broader and broader due to the gain saturation. When the gain is partially recovered, the spectral shift becomes smaller and the spectrum becomes less asymmetric. In the limit of t 0 » * c, which corresponds to a quasi-CW regime, the spectrum becomes symmetric, as the gain recovery is complete, Figure 2.6(b). The output shape is also symmetric and broader than the input pulse simply because the peak experiences less amplification (the gain is saturated) then the wings. x =5 ns, X=1.5 jim, Pin=10 mW t0 =5 ns, k=1.5 Jim, Pin=10 mW i.u L = 1 m m 0.8 ~ Output “ 20 ■ ■ 5 0. 6 0.4 Input Input 5 0 0 Relative time (t~r0 )/'i( Relative Frequency Av=1.8 THz (a) (b) Figure 2.6 Pulse shaping at different points along the SOA (a), and the broadened pulse spectrum at the output of the amplifier in the quazi-CW case (b). 2.3.4 Results: Pulse - effect of the input chirp Here we will investigate the effect of the input chirp on the output pulse spectrum and chirp. Consider Gaussian input pulse described with: 1+iCf _£)2 = , (2.12) 32 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. where Pin is the peak input power and C is the linear chirp parameter. It can be seen A v ( t ) = -------- — = ----------. The local frequency v = v0 + A v{t) is linearly 2n dT I n t 0 increasing with the time. In the case of C=0, i.e. unchirped Gaussian input pulse, the output chirp is shown in Figure 2.7(b). It was already mentioned that the change of the local frequency of the output pulse is due to the gain induced index nonlinearities. This is the effect of SPM - the input photons saturate the carrier density => gain saturates => refractive index changes due to the KK relationships => the pulse phase changes the pulse local frequency changes. Therefore, A v depends on the photon flux and closely follows the pulse shape Figure 2.7(a), and the output spectrum has multiple peaks. A device working in this regime is not suitable as an optical equalizer and only deteriorates the input pulse spectrum. Increasing the input chirp to C=5 (Figure 2.7(c)), diminishes the effect of SPM and the output chirp is determined mainly from the value of C. In the limit of a very large value of C, Figure 2.7(d), the effect of SPM is negligible and does not affect the output chirp. Thus, the output spectrum is mainly determined by the input phase dependence and the output pulse shape: that in this case < p = ------ — , and the frequency chirp Ai/(r) is linear: 2 v7o 7 2 (2.13) 33 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ■ g 3 0.0 Output Pulse - ■ • - Input Pulse -0.2 - £ °'8 ’ < ® 0.6 - « 3 a o.4. T D 1 a - < a -0.4- -0.6 - - 0.8 - - 1.0 - Input chirp C=0 - 1.2 O 0.0 Z Norm alized Time (t/t0) Norm alized Tim e (t/x0 ) (a) (b) 3000 Input chirp C=5 Input chirp C=5000 2000 - !E o ■ o 4) .N E N -1000- - 2000 - -3000 N orm alized Tim e (t/x0 ) Normalized Time (t/i0) (C ) ( d ) Figure 2.7 Dependence on the output chirp on the chirp of the input pulse. The shape of the output pulse stays the same (a), but the output chirp is different, dependent on the value of the liner-chirp parameter: C=0 (b); C=5 (c); C=5000 (d). 2.4 Noise theory in SOA Noise theory [14] in semiconductor lasers has been quite successful in describing many properties including the frequency spectrum and amplitude distribution of the intensity noise. All these theories are only approximately valid for most semiconductor lasers because they assume that the laser cavity can be treated as a closed one, having a discrete set of orthogonal modes. Therefore, the electromagnetic field could be expanded in terms of these modes. In modeling a laser 34 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. cavity, the facet losses are replaced by losses uniformly distributed throughout the cavity. This approximation is still reasonable when applied to the conventional semiconductor lasers having cleaved facets with only 30% power reflectivity. The expansion of the electromagnetic field in terms of discrete resonator modes cannot be applied to traveling wave optical amplifiers, because, ideally, these devices have no modes at all. In treating traveling waves a different approach is used - the electromagnetic energy in each transverse mode is consider to be locally excited and then amplified during subsequent propagation in the device [15], The mean and variance in the photon number at the amplifier output are described by the photon master equation for a unit noise bandwidth ( mtA fts - 1) where mt is a transverse mode number, Af is an optical bandwidth given by an optical filter, T s is the sampling interval over which the number of photons are counted, A and B are coefficients representing the stimulated emission and absorption, respectively, and C is a coefficient representing other loss mechanisms, such as free carrier absorption and waveguide scattering. The mean and variance of the photon number per second at the amplifier output are: [16]: (2.14) 35 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (nx) = G(n0 ) + ( G - \)ysmtAf (n f) = G(n0) + (G - \)ysmtA f + 2 G{G ~ 1 )ys (n0) + { G - i f y fm tAf + (2.15) + G2((n0 2) - ( n 0> 2 - ( « o)) where G is a small signal gain given by exp[(A-B -C )(L /c0)), L is the amplifier length, y s = A /{A -B ), and c0 is the light velocity in the amplifier medium. Five terms on the right-hand side of the second equation in (2.15) represent amplified signal shot noise, spontaneous emission shot noise, beat noise between signal and spontaneous emission, beat noise between spontaneous emission components, and signal excess noise respectively. Factor mtAf in the second terms in (2.15) represents optical noise bandwidth. The signal excess noise disappears when the input signal is completely coherent. The above results can be expresses in terms of photo-detected mean and noise currents [17]: (Iph) = GIs+I ase r , (2.16) (i %„ ) = P ‘01, + 2elA S E + 401,1^/Af + JI^ /A f+ leG H F ,- 1 ) / , J a „ , where Is = (e/hv)Ps , with Ps the input signal power, IA S E = {e/hv)PA S E , with PA S E = h v{G -\)m tysA f the CW amplified spontaneous emission (ASE) power, and Bel « Af electrical detection bandwidth. In the numerical modeling performed in this dissertation, we accounted for the noise in the following manner. First it is necessary to calculate the optical bandwidth of the amplifier, Af, and the output ASE power. A finite number of modes with zero 36 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. input amplitude and with different wavelengths, covering the whole spectrum of the amplifier, were propagated simultaneously with the pulse. The spontaneous emission factor, /in Eqns.(2.8), describing the amount of ASE coupled into the propagation modes, is assumed constant. Thus, the only photons which appear into a given mode and consequently are amplified are contributions from the spontaneous emission in the active material. At the output of the amplifier, the spectral dependence of the amplitude of these modes describes the ASE spectrum (and thus the optical bandwidth). Integrating the modes over the wavelength gives the total output ASE power, P a s e - The output pulse power is also found with the same calculation. These values are substituted in Eqn.(2.16) and the amplifier noise power per unit bandwidth is calculated. 2.4.1 Results: Noise Figure 2.8 is an example calculation of the output noise power from a saturated optical amplifier. The noise power per unit bandwidth and the output pulse are plotted with the dashed and solid lines, respectively. During the pulse, the amplifier saturates, which leads to decrease in the carrier density, and thus decrease in the gain and the output spontaneous emission power. All of these facts contribute to decrease of the spontaneous emission power, as shown. ASE is large only during the pulse leading and trailing edges, where the amplifier is still not saturated and the effective carrier density is large. 37 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 18 I 16 § 1.0 0.8 M 0.6 3 a 3 0.4 a. S ° 0.2 0.0 Relative Time Figure 2.8 The calculated output noise power as a function of the relative time. The noise is greatly suppressed during the pulse propagation because of the saturated spontaneous emission. Figure 2.9 A tree-dimensional plot of the ASE spectrum as a function of time. The pulse arrive at T = -2, saturates the amplifier, and suppresses the output ASE. Figure 2.9 is a three-dimensional plot of the ASE spectrum as a function of time. The pulse arrives at t=-2, saturates the amplifier, and suppresses the output ASE. At t=+2 the pulse amplitude is zero, and as far as the process is in the quazi-CW regime 38 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. with pulse durations much larger than the carrier relaxation time, the carrier density recovers completely which results in an increase in the output ASE and noise. 2.5 Conclusions SPM is a dominant source of spectral broadening occurring as a result of gain saturation that is responsible for time-dependent variations in the carrier density, and hence the refractive index. The temporal and spectral changes occurring during amplification depend on the relative values of the pulsewidth and the carrier lifetime. When the pulsewidth is much shorter than the carrier lifetime, both the shape and the spectrum of the amplified pulses are asymmetric. For input pulses longer than the carrier lifetime the spectrum is broadened on both the red and blue sides. SPM induces frequency chirp, i.e. the local frequency of the pulse changes with time. This self-induced chirp has a large impact on the output spectrum (Figure 2.4(b)), when the input pulse is unchirped. With an increasing of the chirp of the input pulse the effect of this nonlinearity weakens (Figure 2.7) and in the limit of a large input chirp this effect is insignificant. After studding and understanding the general properties of a TWSOA we can underline the following design criteria for building a SOA-EQ: 1. Input pulse energy should be larger than the Esat. 2. The SOA should provide enough gain to ensure gain saturation. 3. A quasi-CW regime should be used with pulse duration of T 0 » T C. 39 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4. In the case of large input chirp the effect of the SPM induced chirp is negligible. 2.6 References 1 G. P. Agrawal and N. K. Dutta, Long-Wavelength Semiconductor Lasers. New York: Van Nostrand Reinhold. 1986, ch. 2 2 C. H. Henry, “Theory of the linewidth of semiconductor lasers.” IEEE J. Quantum Electron., vol. QE-18. pp. 259-264. 1982 3 M. Osinski and I. Buus, “Linewidth broadening factor in semiconductor lasers-An Overview,” IEEE J. Quantum Electron., vol.QE-23, pp. 9-29, 1987 4 S.L.Chuang, “Physics o f optoelectronic devices”, J.Wiley & sons, Inc., 1995 5 R. Bellman. G. Bimbaum. and W. G. Wagner, “Transmission of monochromatic radiation in a two-level system,” J. Appl. Phys. vol.34, pp. 780-782, 1963 6 L. M. Frantz and J. S. Nodvik, “ Theory of pulse propagation in a laser amplifier,” JA ppl. Phys., vol. 34. pp. 2346-2349, 1963. 7 E. O. Schulz-Dubois, “Pulse sharpening and gain saturation in traveling-wave masers,” Bell Syst.Tech, I., vol.43., pp. 625-658, 1964 8 J. P. Wittke and P. J. Warter, “Pulse propagation in a laser amplifier,” J. Appl. Phys., vol.35, pp.460-461, 1964. 9 A. Icsevgi and W. E. Lamb, Jr., “Propagation of light pulses in a laser amplifier,” Phys. Rev., vol. 185, pp. 517-545, 1969. 10 P.G.Kryukov .md V.S.Letokhov, “Propagation of a light pulse in a resonantly amplifying (absorbing) medium,” Sov. Phys. Usp., vol. 12, pp. 641-672. 1970. 11E. Fill and W.Schmid. “Amplification of short pulses in C02 laser amplifiers, ” Phys. Lett., vol.45A, pp. 145-146, 1973. 12 N. J. Frigo, “Ultrashort pulse propagation in saturable media: A simple physical model,” IEEE J. Quantum Electron., vol. QE-19, pp. 511-519, 1983. 13 G.P.Agrawal, N.Aa.Olsson, “Self-Phase Modulation and Spectral Broadening of Optical Pulses in Semiconductor Laser Amplifiers”, IEEE JQE, .vol.25, pp.2297-2306, 1989 40 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 14 L. Coldren and S. Comizine, “Diode lasers and Photonic Integrated circuits ”, Wiley Series in Microwave and Optical Engineering, 1995 15 Y . Yamamoto, “Noise and Error Rate Performance of Semiconductor Laser Amplifiers in PCM -IM Optical Transmission Systems”, IEEE J.Quantum Electronics, vol. QE-16, No.10, pp. 1073-1081, 1980. 16 K.Shimoda, H.Takahashi, and C.H. Townes, “Fluctuation in amplification of quanta with application to maser amplifiers,” J. Ph.ys.Soc.Jap, vol. 12, pp.686-700, 1957 17 S.Donati, G.Guliani, “Noise in an Optical Amplifier: Formulation of a New Semiclassical Model”, IEEE J.Quantum Electronics, vol.33, No.9, pp.1481-1488, 1997 41 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 3 Design of an optical equalizer 3.1 Design Issues In this section we will try to overlay the design criteria for an optical equalizer. First, we have to keep in mind the overall objective. The input to the device is a highly linearly-chirped Gaussian pulse. This means that at different moments different wavelengths are present, and that the local wavelength changes linearly with time. Also, the bandwidth of the pulse is compatible with the frequency bandwidth of the gain media, and the details of the gain spectrum have to be taken into consideration. We will also use the saturation properties of the amplifier, namely at high gain levels and large variation of the input power gain is largely suppressed. Figure 3.1 A three-segment spectrally and spatially inhomogeneous gain medium. We will focus our attention on spectrally and spatially inhomogeneous gain medium, shown schematically on Figure 3.1, with the idea that having independent control on the separate sections (by varying the pump currents) will give us a mechanism to shape the overall device spectrum. The requirement is that the longest- wavelength section, called 'main', has the highest/broadest gain. Also its spectrum has to overlap with the gain spectra of the shorter-wavelength sections, called 42 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 'auxiliary'. The reason for this constraint is that any semiconductor gain media is transparent below the bandgap (at the long-wavelength side of the spectrum), and is absorbing at the short-wavelength side. Thus, the purpose of the main section is to provide the gain in the device while the purpose of the auxiliary sections is to equalize the device spectrum at the short-wavelength side, and thus effectively broaden and flatten the bandwidth. d=3 nm d=9 nm 0 15 but o u t ^ 0= 1.55um 1.35 1.40 1.45 1.50 1.55 1.60 Wavelength [|x m] Figure 3.2 An example theoretical calculation of the gain spectra of two independent gain media pum ped separately. An example theoretical calculation is shown in Figure 3.2, where we have plotted the unsaturated gain of two independent gain media pumped separately. An Inx Gai_ x Asy Pi.y MQW region with different well width is assumed, with well As composition of yw =0.74, well compressive strain of ^=+0.8%. There is zero barrier strain (^=0), and barrier As composition is y*=0.378 (for a bandgap wavelength of ^=1.15|im). The long-wavelength section is L;=500pm long, has four QWs with 43 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. width d . Q W = 9mm, peak wavelength of Aqw/=\.55\im and is pumped with /;= 100mA current. The short-wavelength section is Lj= 100pm long, has four QWs with dQW=3nm width, peak wavelength of AQw=1.4pm and is pumped with 7;=12mA current. The device is assumed to be a buried heterostructure (BH) amplifier with 2pm stripe width. The large difference between the peak wavelengths of the two sections is chosen just for demonstrative purpose. The spectral responses shown in Figure 3.2 are the ones we would measure if we process independently different devices from both active regions, i.e. responses of ‘ordinary’ optical amplifiers. L=500 urn, d = 9 0 A, 1=100 mA, L 2=100 pm, d2 =30 A, l2=12 mA 3 0 F “ “B m TS 25 V 1.4pm 3 ■ # » < (0 m c 3 1.30 1.35 1.40 1.45 1.50 1.55 1.60 Wavelength [^m] Figure 3.3 The calculated CW spectrum of a two-section SOA. The short-wavelength section only equalizes the device spectrum at shorter wavelengths. Figure 3.3 shows a calculated combined CW spectral response from a device having two gain sections built from both active media described above. The short- wavelength portion of the equalizer amplifies only around /l=1.4pm, while it is transparent at longer wavelengths. Thus, by adjusting the relative strength of the 44 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. pump currents in both sections a spectral equalization can be obtained over more than zl/b=200nm. This result, of course, serves only for demonstrative purposes and in practice there would be many obstacles in realizing such large bandwidths - for example designing a broadband antireflection coating. L=0.9mm, 1=120 mA, t=5 ns, A.=1.46 pm, Pin=10 mW 1.0 . d=9 nm Q > ■ o 3 0.8 O utput C L 0.6 E < g 0.4 C O 0) 0.2 C C Input 0.0 Relative Time (t-x)/x Figure 3.4 A theoretical plot of the output pulse shape amplified by a single-section device with a long-wavelength gain region. Now we will apply the described two-section device with spatially/spectrally inhomogeneous active region to pulse equalization. The drive currents and input pulse amplitude are chosen so that the amplifier is saturated by the pulse leading edge. This limits the output dynamic range and amplitude variations, and thus provides a simple means of equalization. Remember that the input is a highly- chirped nanosecond Gaussian pulse. Therefore, effectively the amplifier operates in a quasi-CW regime and gain dynamics and carrier relaxation processes are not important. An example numerical result is plotted in Figure 3.4, where the input is 45 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. shown with a dotted line and has a central wavelength of A= 1.46pm, a bandwidth of 60nm, peak power of lOmW and duration of -Zo=5ns. The simulated amplifier is a single-section device built from the long-wavelength spectral region and is L=900pm long and pumped with /= 120mA. To understand why the output pulse has this asymmetric shape we have to consider two phenomena: (i) the device is saturated by the pulse leading edge. This explains why the output has a sharp edge at low input intensities; (ii) the input is a highly linearly chirped Gaussian pulse. This means that the instantaneous wavelength decreases linearly as the time progresses, and thus the output will have a longer tail, which follows the roll-off of the device gain at shorter wavelengths, Figure 3.2 (remember that the pulse center wavelength is 1.46pm, the bandwidth is 60nm, so that the leading edge wavelength is 1.49pm, and the trailing edge wavelength is 1.43pm). Now let’s look at the effect the short-wavelength section would have on the performance of the amplifier. An example numerical result is plotted in Figure 3.5, where the input is the same as above. The simulated amplifier is a single-section device built only from the sort-wavelength spectral region and has the same length and drive current as in the previous example. Note that the output shape is quite different. The pulse possesses a longer leading edge and shorter tail then before. This shape is a result from: (i) the chosen spectral distribution of the short-wavelength gain medium with respect to the pulse central wavelength; (ii) the assumed large input chirp. The short-wavelength section is spectrally shifted from the pulse central 46 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. wavelength, and thus the pulse leading edge has a wavelength below the bandgap of the amplifier. Therefore, the device is transparent and the leading edge experiences no amplification. This is not the case for the trailing edge. As time progresses the instantaneous wavelength of the pulse decreases, so that during its trailing edge, the wavelength is above the bandgap energy of the active material. This explains the large amplification experienced by the pulse trailing edge, observed in Figure 3.5. L=0.9mm, 1=120 mA, r=5 ns, k=1,46 pm, Pin=10 mW 1.0 d=3 n m ® 0.8 3 O u tp u t X =1.4pm U C L 0.6 £ < © 0.4 0.2 ' Input 0.0 2 1 2 3 4 -4 3 1 0 Relative Time (t-t)/t Figure 3.5 A theoretical plot of the output pulse shape amplified by a single-section device with a short-wavelength gain region. Now, following the same line of reasoning as in the CW case, let’s build a two- segment optical amplifier by combining the spectral responses of the two sections. We will design the long-wavelength section to be longer, Li=900gm, and pumped harder, Ij= 120mA, than the short-wavelength section, L2= 100pm and I2= 12mA respectively. This is because we would like to have larger gain at long wavelengths as shown in Figure 3.2. The calculated response is shown in Figure 3.6. The leading 47 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. edge of the pulse is amplified only by the long-wavelength section, while the trailing edge (and shorter wavelength part of the spectrum) is influenced by both sections. Thus, equalization can be achieved by adjusting the relative strength of the drive currents in the separate sections - the output, equalized pulse is shown with the solid line. The effective device gain is also plotted on the same graph and it exhibits an unusual, large dip during the center of the pulse due to.gain saturation [ 1], A , =1.46 urn, AA=60 nm, t =5 ns, E. =25 pJ 0 ^ ’ 0 ’ in r ■o 3 H h J Q. E < J5 @ ) cc 0.0 - i -j---- 1 -----.----J -----« -----S -----1 ___1-----i i ---- - i * \ output i ' ■ \ i i \ A i \ V * i i \ ; input ■ . / \ * , i . / \ . - s t i i % i v \ ’ ■ ■ x i / \ \ ' • N i y / ■ \ ' v 4 - ' \ i i / J 2 ) ! ' \ \ i : : • • X - \ -3 ... - 2 i > i • L i 1 - 1 0 1 2 3 16 14 1 2 m ■o 1 0 “ o (0 8 o 6 ® ° O atom 4 g Q 2 0 Relative Time tk n Figure 3.6 Pulse equalization from a two-section spectrally and spatially inhomogeneous optical amplifier. The output spectrum of the pulse in the frequency domain, calculated by taking a Fourier transform (3.1) in the time domain, is shown in Figure 3.7. s 0 M = JVI P o M e ^ e ^ d r (3.1) Due to the fact that the amplitude of the spectral components is proportional to the amplitude of the pulse in the time domain, and also because the output phase 48 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. changes quadratically with time (the frequency changes linearly), the shape of the spectrum follows closely the shape of the pulse and is effectively equalized. The horizontal axis has a unit frequency interval of Av=2.8THz, and thus spectral equalization is achieved at more than 12THz. x=5 ns, k=1.46 pm, Pin=10 mW 1.0 ® 0.8 Output o . 0.6 ® 0.4 < D 0.2 Input 0.0 Relative Frequency Av =2.8 THz Figure 3.7 The calculated output spectrum of the pulse plotted in Figure 3.6. It is interesting to investigate the sensitivity of the proposed equalizer to variations in the amplitude of the input pulse. In the following theoretical ‘experiment’ we vary the input amplitude by more than 2 0 0 % (but still assuring device saturation at all times), and the task is to find out how this would affect the output shape and amplitude. The calculation is performed and shown in Figure 3.8. We can see that the SOA/EQ is not sensitive to variation of the input signal. Change in the input amplitude by more than 200% leads to 10% variation in the output. Note that the 49 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. pulse shape and amplitude stay the same due to the very high gain of the amplifier, which is completely saturated by the pulse leading edge. t=5 ns, A,=1.46 pm • _ I 1__1 — A --- 1 — 30 25 £ 20 ' Output ^ 10 Input -4 3 1 0 1 2 3 4 2 Relative Time (t-x)k Figure 3.8 SOA/EQ is not sensitive to variation of the input signal. Change in the input amplitude by more than 200% leads to 10% variation in the output. However, in a practical situation it is difficult to build a device with very high gain. It would be beneficial to consider a configuration of two two-stage amplifiers with lower gain but working in series. The results from a simulation of these identical amplifiers connected in series, with a single device section lengths of L;=500pm and L2= 100pm and drive currents of /;= 100mA and 7 2=12mA respectively, are shown in Figure 3.9. The gain of the first equalizer is not enough to amplify the signal, and achieve complete gain saturation. Thus, the output pulse shape is broad, but not equalized. The second amplifier connected in series broadens the pulse even more and is completely saturated. Therefore complete pulse equalization can be achieved. 50 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1=2.5 ns, X,=1.5 2 nm, L1=500 nm, d=90 A, 1=100 mA Pin=10 mW L2=100 nm, d=35 A, l= 12 mA O utput Second D e v ic e "O utput First Device4 Relative Time (t-x)/x Figure 3.9 Pulse equalization from a two two-section devices. The second amplifier completely equalizes the output pulse. 3.2 Conclusions In conclusion, in order to achieve spectral equalization we have to meet the following criteria: 1. A multi-section device with the longest-wavelength section having the largest and broadest gain than all other sections is needed. The spectrum of each consecutive shorter-wavelength section has to overlap and be included into the spectrum of each preceding, longer-wavelength section. 2. Independent biasing of all sections. 3. CW equalization can be achieved at all input levels and without using gain saturation, by adjusting the relative strengths of the currents in the separate sections. 51 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4. To achieve equalization in the pulse regime, the pulse has to have large linear chirp (i.e. the frequency linearly increases with time) and a duration much larger than the carrier relaxation time (quasi-CW regime). 5. For pulse propagation, the device has to have large total gain to assure gain saturation at low input levels. 6 . If gain saturation is always achieved (high enough gain), the device is insensitive to variation in the input signal. 7. Multiple multi-section devices, arranged in a series, can be used for achieving larger combined gain and better pulse equalization. 3.3 References 1 Djordjev, K.D.; In Kim; Dapkus, P.D., “Chirped Pulse Propagation in Saturated Traveling W ave Sem iconductor Am plifiers, ” LEOS '99. IEEE Lasers and Electro-Optics Society 1999 12th, Annual Meeting, Volume: 2, 1999, pp: 786 -7 8 7 52 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 4 SOA-EQ fabrication 4.1 Introduction to SAG The fabrication of two-section semiconductor optical equalizers requires the utilization of an inhomogeneous gain medium with different sections having different spectral properties grown on the same wafer. An approach used for local bandgap engineering is the selective area growth technique (SAG) [1,2]. SAG enables monolithic integration of active regions with different spectral properties by r r T y f a) b) Figure 4.1 Selective area growth process: (a) the growth enhancement is due to gas-phase diffusion and surface migration; (b) the SAG profile after growth. means of defining dielectric stripes on the wafer surface just prior to the growth process. Figure 4.1(a) shows a schematic of a SAG wafer with the surface patterned with a SiNx mask. Because growth is not possible in the regions with the 'SAG . Figure 4.2 Orientation of a two- dielectnc mask, and due to the gas-phase diffusion section deviC e with respect to the SAG region. and surface migration processes, there is a growth rate enhancement and a composition change in the area between the dielectric stripes. The result is a red shift in the bandgap energy (towards longer wavelengths) in these regions, relative to the bandgap energy in the plain region, far from the 53 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. mask. The approach used for device fabrication is to utilize the area between the stripes (SAG region) as a gain region for the long-wavelength section, and the plain area as a gain region for the short-wavelength section. The device mesa is aligned parallel to the SAG stripes, Figure 4.2, and there is a natural adiabatic transformation in the thickness and in the bandgap energy between both sections. The resultant profile after the growth is shown schematically in Figure 4.1(b), and in Figure 4.3, where SEM pictures show the cross-section of an as-grown SAG region. SiN stripes Figure 4.3 Cross-sectional SEM photograph o f an SAG region just after the growth is performed. Pairs of SiNx stripes with different widths, w, varying from 10pm to 50pm, and 15pm open area between them, were used as a selective area growth mask pattern. The stripe patterns were spaced at 250pm center to center. The growth enhancement factor (defined as the ratio of the thickness of the grown layer in the SAG region to the thickness of the layer in the plain region without dielectric mask) is plotted in Figure 4.4 as a function of the SAG stripe width w. Varying the stripe width from 0 to 50pm, enhances the growth by 120% within the 15pm SAG region. 54 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 120 +7 1 0 0 - C 60- C 40 10 20 30 40 50 Stripe Width w [|im] Figure 4.4 Selective area growth enhancement factor as a function of the dielectric stripe width, w. Pictures of six quantum wells grown at the same time within different SAG regions are shown in Figure 4.5. The quantum well thickness increases with w, and the composition changes accordingly, which results in a red shift of the bandgap energy. w= 0pm. f J w= 15pm I w= 30pm | | w= 50pm Figure 4.5 Pictures of six quantum wells grown at the same time within different SAG regions with mask widths w. This chapter is organized as follows. First we will discuss the issues associated with the pre-growth preparation of an SAG wafer. Then, we will describe the 55 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. techniques developed within this research project for SAG device fabrication. We utilized broad-area, buried-heterostructure and ridge-waveguide structures. 4.2 Wafer Preparation for SAG The SAG technique requires patterning of the wafers just prior to the growth. All measures have to be taken to avoid surface contamination during this step. One of the crucial processes was found to be the SiNx mask formation. If CF4 plasma was used to etch away the SiNx layer, it somehow reacted with the photoresist and the heterostructure grown thereafter had very poor quality - lasers processed from those wafers did not work or had very high threshold currents. To find out the right pre growth treatment process we performed the following experiment: Five pieces from the same wafer were treated differently prior to the growth. They were loaded together into the MOCVD chamber and a standard laser structure was grown, broad area lasers were processed, and their performance was compared. Usually the SAG device pre-growth process involves SiNx deposition, photolithographic patterning of the SAG stripes, dielectric mask etching, and substrate pre-growth cleaning. In this experiment we didn’t pattern the wafer surface but only simulated a real fabrication process and then grew the laser structure on plain substrates. Samples: #1 Used for reference. It was not treated at all but just cleaved from a bare wafer and used for comparison. Referred to as ‘no treatment’. 56 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. #2 The sample was only cleaned with DI water rinsing for 2 hours. There was no surface treatment and this sample was again processed for reference to compare the effect of the cleaning on the sample performance. Referred to as ‘cleaning only’. #3 On this sample we deposited a lOOOA-thick SiNx layer. A 1.4pm AZ5214 photoresist was spun, a high energy blank UV exposure was applied and the PR was removed with concentrated AZ400K developer. The SiNx was stripped with BOE 1:10 for 3min., and the sample surface was pre-growth cleaned with a HiSO^HkOi^O (1:1:3; 65°C) dip for 20s and DI water rinsing for 2 hours. The sample is referred to as ‘BOE+DEV+clean’. #4 Same as sample #3 except that a long ashing with RD E O2 plasma (5min, Pr f= 100W, 200mTorr) was applied to clean the photoresist residue, before stripping the SiNx layer. The sample is referred to as ‘BOE+Oi+clean’. #5 Same as sample #3 except that the SiNx layer was etched with CF4 RIE plasma (lmin, PR f=10QW, lOOmTorr) instead with BOE. The sample is referred to as ‘ CF4+DEY+clean’. From the above description it can be inferred that samples #1 and #2 were used as a reference and samples #3, #4 and #5 simulate real SAG pre-growth processes. For simplicity ‘regular’ broad-area lasers were fabricated on a plain wafer. The measured LI curves from completed broad area lasers are shown in Figure 4.6. It can be seen that samples #1, #2 and #4 performed identically (within the spread in the data), while sample #3 had much higher threshold current. It seems that 57 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the cleaning of the PR with concentrated developer is not enough (cleaning with solvents were also tested without improvement in the performance) and the PR residues degrade the device performance. Extended 0 2 treatment (sample #4) cleans the surface completely and the devices perform identically with the ones fabricated from bare wafers (#1 and #2). The performance (or more precisely, the lack of performance) of device #5 is even more surprising - it did not lase at all. The CF4 plasma reacts somehow with the wafer and the PR residues, damaging the surface, and thus the QW region grown on top of it has poor quality with lots of defects. #2937 - S a m p le s S iz e L = 1 500 am - j ___________ 1 ___________ i___________ t___________ L ______________________ i___________ » ___________ i_________ 6 0 - ...........#1; no treatment ■ ---------#2; cleaning only 5 0 - —* - #3; BOE+DEV+clean # 4 ; BOE+02+clean ■ — 1 . ----------#5; CF4+DEV+clean E 1-1 3 0 - m I 2 0 - £L 1 0 - 0 - 0.0 0.2 0.4 0.6 0.8 1.0 Current [A] Figure 4.6 LI curves from samples with different pre-growth treatment. The best results are obtained with ‘BOE+02+clean’ treatment. With this experiment we found out that the pre-growth treatment is critical for wafer quality. Treatment similar to the one applied to samples #4 was employed hereafter towards the processing of high-quality SAG devices. #2937 - S am ples Size L = 1 500 urn j____i ____1 ____ 1 ____. ____ 1 ____ 1 ____ 1 ___ • #1 no treatment - - #2 cleaning only * - #3 BOE+DEV+clean - - #4 BOE+02+clean - - #5 CF4+DEV+clean 58 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 4.1 Epi-structure used for growing SAG devices La Description Material PL(nm) Strain Thickness ( A ) Doping 0 Substrate InP - - n 1 Buffer & Spacer InP - - 500 n+, 3E18 2 SCH InGaAsP 1200 0.18% tensile 900 undoped 3 Quatumn Well InGaAsP 1490 0.7% compressive 70 undoped 4 Barrier InGaAsP 1200 0.18% tensile 240 undoped 5 Quantum Well InGaAsP 1490 0.7% compressive 70 undoped 6 Barrier InGaAsP 1200 0.18% tensile 240 undoped 7 Quantum Well InGaAsP 1490 0.7% compressive 70 undoped 8 Barrier InGaAsP 1200 0.18% tensile 240 undoped 9 Quantum Well InGaAsP 1490 0.7% compressive 70 undoped 10 SCH InGaAsP 1200 0.18% tensile 900 undoped 11 Spacer InP - - 500 undoped Interruption between the first and second growth 12 Spacer InP - - 1000 undoped 13 Spacer InP - - 2000 p, 1E17 14 Spacer InP - - 3000 p, 5E17 15 Spacer InP - - 5000 p, 1E18 16 p+ contact InGaAs - unstrained 2000 p+, 1E19 The SAG pattern was aligned in the [Oil] direction on an n-type (001) InP substrate. After the patterning and cleaning of the wafer a two-step SAG growth is performed. The epi-structure consists of a 500A-thick InP buffer layer followed by a four MQW active region with a peak wavelength of h= 1.49pm, embedded in a 0.4- pm 1.2pm-InGaAsP waveguide core layer. At the end is grown a 500A-thick i-InP spacer layer. Up to this point the active region is grown and the desired wavelength shift between the SAG and the plain regions is achieved. The SAG dielectric stripes are no-longer necessary and to prevent further growth enhancement, which may result in a very non-planar surface (obstructing device processing), the sample is taken out of the chamber and the mask is etched with HF for 3min to remove the 59 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. SiNx stripes. The sample is cleaned by DI water rinsing for 2 hours and the top p~ cladding layer and p+ contact layers are grown. The epi-structure used throughout these experiments is given in Table 4.1. As already mentioned the composition of the quantum well changes in the SAG region. There are reports [3,4], that for our configuration the As1 composition increases by 5% for SAG stripe width of w=50pm, and thus the expected change in the QW strain is As=+0.4%. 4.3 Broad-area SAG devices The SAG broad area devices had a mesa width of 10pm, restricted by the spacing between the SiNx mask patterns (15pm). A two-step SAG growth was performed and the mesa was defined in the middle between the SAG dielectric patterns in order to avoid the interface regions close to the mask edge with degraded quality. A sketch of the mesa orientation with respect to the patterns is drawn in Figure 4.7(a). In Figure 4.7(b) a cross-section of the SAG interface region is shown to span around 1 to 2pm from the edge. The quantum wells thickness increases when approaching the dielectric mask, which leads to a change in the bandgap energy, increase in the strain, and degradation in the QW quality. The process starts with a spin of a 1.4pm-thick layer of AZ5214 photoresist and definition of rectangular openings 10x50pm on top of the transition region between the SAG and plain areas. This pattern will define the electrical-isolation trench between both sections. A two-step wet etch is applied: 15s of H2S 0 4 :H2 0 2 :H2 0 (1:1:3 @ 20°C (RT)) to etch the top InGaAs p+ layer, followed by 4s of HCkEfeO (3:1 @ 60 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. RT) to etch 0.2-0.4pm of the InP cladding layer. See Figure 4.8 for details on device fabrication. The photoresist is stripped by solvents (TCE + ACE + ME) and the residues are ashed by O2 R ID E plasma (150W, 150mtorr, 2min). Device direction [Oil] 1 1 1 1 1 1 n ii ii i i * Longer ^section Short section I itit-rl'iic*' region III II II - a) ; ! b) Figure 4.7 Top view of the wafer surface with the orientation of the device (a), and cross- section of the SAG region close to the mask edge. Second, the sample surface is cleaned by a dip in NH4OH/H2O - (1/6) for 30s to remove the thin oxide layer and rinsed in DI water for 2 min. A 2500A-thick SiNx layer, to be used as dielectric mask for the dry etch of the mesa, is deposited. 1 0 pm- wide photoresist stripes are defined in the middle of the SAG region by optical photolithography and CF4 RIE plasma etch (100W, lOOmtorr, 1.6min) is used to define the dielectric pattern. The sample surface is cleaned from the photoresist by solvents and O2 plasma treatment. Third, the mesas are dry etched in ECR discharge using BCE/Ar chemistry at an elevated temperature of 150°C. Deep ~3-4pm mesas are formed and the interface region with low quality (close to the SAG mask) is removed. The sample surface is 61 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. cleaned with O2 plasma and the SiNx mask is removed by a BOE dip for 5min. A new layer of lOOOA-thick SiNx is again deposited to protect the top surface of the contact layer from the following planarization process. plain region SAG region cladding layer buffer layer ] _ Longitudinal direction 2. W et etch of isolation trenches cladding la 3. Cross-sectional view t SiN n substrate bstrate substrate substrate 7. Sample lapping and 6. Second SiN deposition 5. Dry etching of 4. SiN deposition & contact deposition polyimide planarization laser mesa definition of lOpm stripe & 7|im contact opening Figure 4.8 A process flow-chart for fabricating two-section, broad area SAG devices. Now we have lOpm-wide broad area laser mesas formed. These mesas are relatively deep and narrow, so if we decide to deposit the top p-metals at this point, we might encounter two problems during the device measurement. First, direct probing on top of this narrow mesa is very difficult (width of only 1 0 pm) and the mesa could be easily broken by the probe. Second, defining a bond-pad directly on the sample surface is almost impossible, because the mesas are relatively deep and vertical, and connecting this pad to the p+ contact layer would be impossible. When we undertook this approach, in most of the devices the top p+ contact layer was not connected to the bond-pad on the sample surface due to the shadow effects during 62 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the gold deposition process. This was the reason why a polyimide (PY) planarization process was developed. HD Microsystems’ PI-2737 photosensitive polyimide was spun at 2000 rpm and soft baked at 90C for 4min in an oven. After the soft bake the thickness of the layer is more than 6 pm. A cross sectional view of a BA mesa and the spun PY layer before the curing process is shown in Figure 4.9. ____________i l J l Aw, V Maqn Del WD I : ------ 1 2 |irn IbPkVJM SOOOx S F 9 ? SOA-SAG - 7um X 4um + qold + PY Figure 4.9 A cross-section of a deep BA laser mesa processed within 15pm SAG region. A thick layer of polyimide is spun on top of it. The mesa is 4pm deep and vertical. A 7pm stripe was aligned on top of the mesa and UV exposed with 150mJ of energy at 405nm. Using negative tone lithography, the unexposed areas of the polyimide are removed by developing in DE-9040 developer for 20s. The sample was rinsed in RI-9180 rinse and dried with a N2 gun. The mesa top is thus cleared from the polymer. The polyimide is then cured in a N 2 atmosphere by ramping the temperature from 20°C to 350°C at a rate oflO°C/min and then curing at 350°C for 30min. 63 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. If the top of the mesa contains some PY residues, it could be cleaned with O2 plasma treatment (repeat 100W; 200mtorr; 0.5min; + observe, until the surface is cleaned). Top view of a SOA mesa after the polyimide planarization process is shown in Figure 4.10. There is still some PY residue on top of the contact layer. The surface of the polyimide is rough as a result from the O2 plasma treatment. iB ■ H T I ! H s y d j fa m ^Cl. V S p o t Magn i 5.0 kV 3.1 8500k 20 2 SOA m e sa + polyimide Figure 4.10 A SEM picture showing the SOA mesa top, after the polyimide curing and the application of short-time 0 2 plasma treatment. The PY layer is rough. This roughness could be relatively large. An unoptimized PY process could result in r.m.s. roughness of more than 2pm, an extreme case shown in Figure 4 .11(a). In this photograph the PY pillars have the same height as the layer thickness, and the area between them is so deep that even the sample surface is exposed. This happens, when a very thick PY was initially deposited (coating speed of less than 2 0 0 0 rpm) and long-time, low-pressure O2 plasma was used to clean the mesa surface. For comparison, in Figure 4.11(b), an optimized PY planarization process is 64 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. shown, where the mesa top was completely clean after the developing of the PY, and no O2 cleaning was applied. The polyimide is very smooth and the bond-pad could be easily defined on this surface. (a) (b) Figure 4.11 An unoptimized PY process could result in r.m.s. roughness of more than 2pm (a), and a smooth PY surface resultant from an optimized process (b). After the planarization process, the top p-metals could be deposited. For this purpose AZ-5214 photoresist is spun at 3000rpm and an image reversal photolithography is performed (spin at 3000rpm; soft bake for 30s at 120°C; align and expose the mask with 70mJ at 405nm; post-bake for lmin at 120°C; blank flood expose without mask with 240mJ at 405nm; develop with diluted 1:4 developer AZ400K for 20s). The image reversal process forms an undercut photoresist profile, which will help during the following metal lift-off step. After defining the photoresist pattern for the lift-off process, the SiNx layer protecting the mesa top is removed with CF4 plasma (100W, lOOmtorr, lmin). P-metals are deposited (300/500/2000A-Ti/Pt/Au) by e-beam evaporation at 5x1 O '7 torr base pressure. The sample is dipped in acetone to perform the lift-off process (where the metals outside the pads are removed) and then is cleaned with methanol. 65 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In order to achieve good cleaving with high quality facets, the samples have to be thinned by mechanical polishing. A 5pm alumina-containing polishing powder was used for this purpose and the backside of the sample was polished manually to 100pm thickness. The devices were cleaned by TCE+ acetone+ methanol and rinsed in DI water for 2min. The samples were attached to a microscope glass slide with a vacuum tape and n- metals (AuGe/Ni/Au-1000/300/2000A) were deposited using an e-beam evaporator. In order to form an Ohmic interface between the semiconductor and the metals, the samples have to be annealed. A rapid thermal annealing (RTA) system was used and to find the optimal annealing temperature a set of experiments were performed. In these experiments the temperature was ramped from 20 to T [°C] at 5°C/s rate, kept at the maximum temperature T for 30s, then decreased down to 100°C at the same rate of 5°C/s. The temperature T was varied and was 360, 380, 400, 420, 440, and 450 °C. Measured IVs of the different devices are plotted in Figure 4.12. It can be seen, that 420°C is the threshold temperature for an Ohmic interface that also minimizes the contact resistance. On the other hand, the annealing temperature has to be kept as low as possible in order to decrease the thermal expansion of the polyimide (the annealing is performed at higher temperature than the PY curing). In all of the following experiments it was fixed at 430°C. The samples are cleaved in bars approximately 1mm long and a two-layer AR coating is deposited on both facets. 66 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Effect of the anealing tem peratu re on the resistance,of BA laser 0 .0 8 - 0 .0 6 - 304/400 C 0 .0 4 - 304/380 0.02 - 30S/360 0.00 0.0 1.0 1.5 0.5 Voltage [V] Figure 4.12 Resistance as a function of the RTA annealing temperature. Temperatures in excess of 420°C are required for obtaining low resistance. Contact Padj Polyimid & ation rench Substrate Figure 4.13 A schematic of the processed BA device (a), and a SEM picture (top view) of the same device (b). A schematic of the processed BA device is shown in Figure 4.13(a), and a SEM picture (top view) is shown in Figure 4.13(b). Both sections of the device can be pumped independently and a 1 0 pm-wide trench is formed between them for electrical isolation. As already mentioned the SAG and the plain regions have different thickness due to the growth rate enhancement in the SAG region. This 67 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. means that the waveguide eigenmodes in these sections are different and the mode mismatch could lead to reflections at the interface. Fortunately, the transition between the SAG and the plain sections is adiabatic. The thickness of the core layer linearly decreases from t$A G to tp la in and the transformation region is approximately 50pm long, which forms a natural mode transformer. 3.24- "2 3.23 — - Effective Index > 3.22- - ® - - Reflectivity 3.21- 3.20- 3.19 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 Trench Etch Depth [pm] Figure 4.14 Calculation of the waveguide mode effective index and the reflectivity at the interface trench/waveguide as a function of the trench etch depth. Much larger reflections are expected if the isolation trench is very deep (see Figure 4.14). The thickness of the p-cladding layer is 1.5pm and if the depth of the isolation trench is 1.5pm (i.e. the trench is etched up to the waveguide core layer) the reflection coefficients would be as high as 5xl0'3[%] per interface, larger than the reflectivity of the AR coated facets (a good AR coating can have reflectivity as low as 10‘5 ). The mode effective index is calculated using a slab waveguide approximation [5,6] (ID case) and the reflection coefficient is calculated as a Fresnel 68 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. reflection [6,7] at the waveguide/trench interface: R = (n2- n, ) 2 /(n2 + «, )2, where nj and n2 are the different mode effective indices in both regions (n2 is the effective index at zero trench etch depth, i.e. the effective index of the waveguide mode far from the trench region). Thus, the trench can act as an effective mirror and form two coupled FP cavities. In order to minimize this reflection, a shallow ~ 0.3-0.4pm etch is needed, which at the same time will provide good electrical isolation between both sections. Thus, a precise control of the etching time with HChHaO chemistry is necessary. 4.4 Buried-heterostructure (BH) devices Buried-heterostructure devices were also fabricated as part of this project. The SAG, as a two-step growth approach is completely compatible with the BH process. First the wafers were patterned for the SAG growth and cleaned as discussed above. The first step of the growth is performed and the epi-structure used is the same as the one shown in Table 4.1. The samples were taken out of the MOCVD chamber and the SiNx stripes were removed by a dip in BOE for 5min. The active region was then grown and the desired bandgap wavelength shift between the SAG and the plain regions was formed. Second, a lOOOA-thick SiNx layer is deposited to be used as a dielectric mask for the ECR dry etch process forming the mesa. A 2pm wide photoresist stripe is formed in the middle and parallel to the SAG mask and the SiNx layer is patterned in a CF4 R T F , discharge. A 1-pm deep mesa is dry etched in BCb/Ar ECR plasma discharge. 69 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This plasma chemistry undercuts below the mask and the profile has a negative slope, which is favorable for formation of the current blocking layers (CBLs). The sample is first cleaned in 0 2 RIE plasma discharge (100W, 200mtorr, 2min) and then pre-growth treated with HBr/HhO (1:1) for 30s. After rinsing in DI water (18MD) for 2 hours the CBLs are grown. Table 4.2 shows their composition and thickness. Table 4.2 Epi-structure of the current blocking layers .ayer Description Material Thickness (A) Doping 5/3 ratio 1 p-layer InP 6800A 5E17 442 2 n-layer InP 2000A 5E17 136 3 p-layer InP 3400A 5E17 442 Figure 4.15 A BH mesa with grown CBLs. Figure 4.15 is an example of BH mesa with grown CBLs. The photograph is taken after a stained wet etch is applied to show the interface between the different 70 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. layers. The first p-layer has to cover the sidewalls but should not reach the top of the mesa, because it will form a current path which will increase the leakage currents. One of the specific properties of the SAG grown substrates is the different thickness of the active regions across the wafer. If the dry chemistry etch rate, used for mesa formation, depends on the material composition, then the different thickness of the active region will result in different etch depth across the wafer. Thus, it is almost impossible to grow high-quality CBL’s for all SAG regions at the same time, as the mesa depth vary and the CBL’s could be optimized only for a given depth. Therefore, it is of great importance to develop a dry process which is anisotropic and has constant etch rate, independent of the material composition. We thus used a BCl^Ar chemistry in an ECR discharge at elevated temperatures (150°C), because the etch rate shows little dependence on the material composition. After the CBL’s are grown, the dielectric mask, used for mesa formation, is removed by a dip in BOE. The samples are cleaned by rinsing in DI water for 2 hours and the top p-cladding layers are grown. The epi-structure is the same as the one shown in Table 4.1 (second growth step). The isolation trench is formed as described previously, followed by a deposition of lOOOA-thick layer of SiNx. A lOpm-wide stripe is opened in the SiNx just above the mesa top to make a widow in the dielectric for the p-contact metals. The p-metal lithography is then performed as described previously, followed by the p-metal evaporation. 71 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 4.16 A cross-sectional SEM photograph of a completed BH device. A stain etch is performed to reveal the waveguide region containing the QWs. A cross sectional view of a BH device is shown in Figure 4.16, where a selective wet etch with H2S0 4 :H2 0 2:H2 0 (1:1:3 @ RT) was applied to examine the active region. In this picture the original negative slope of the dry-etched mesa could be inferred from the shape of the active region. Figure 4.17 shows a top view of a two- section device at the interface region between both sections and the isolation trench, the top of the mesa, and the gold contact pads could are visible. Finally, the wafer is lapped to 100pm in thickness by mechanical polishing, cleaned, and n-metals deposited (AuGe/Ni/Au - 1000/500/2000A). A rapid thermal annealing is performed at 430°C as described above. The samples are cleaved in bars 1mm long and a two-layer dielectric AR coating is deposited on both facets to reduce the reflectivity. 72 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 4.17 Top view of a completed two-section BH optical equalizer. The mesa, the isolation trench and the gold bond-pads of the two sections are visible. 4.5 Ridge-waveguide devices SAG ridge-waveguide devices were also fabricated as part of this project and the fabrication process will be described hereafter. The epi-structure is almost identical with the one used for BA devices, as shown in Table 4.1. The only difference is the addition of an etch-stop layer, which is embedded in layer #12 - see Table 4.3 for details. This etch-stop layer is used during the definition of the mesa by a selective wet etchant and its purpose is to protect the area where the SAG mask was initially defined. This protection is necessary because during the first SAG growth step the SiNx mask is present, but after its removal the original InP wafer surface is exposed. Now, if we only grow the top InP p-cladding layers on this surface with the second growth step, the area will contain only InP material, and during the following 73 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. selective wet etching will be unprotected. Therefore, to avoid the excessive etch depth in the area, where the SAG mask were initially present, and to lessen the fabrication tolerances a etch-stop layer is grown as the first layer during the second growth step. Table 4.3 Modification to the epi-structure shown in Table 4.1 as used for fabrication of ridge- waveguide devices. 11 Spacer InP - - 500 undoped Interruption between the first and second growth 12’ Spacer InP . 500 undoped 12” Etch-stop layer InGaAs . unstrained 200 undoped 12’” Spacer InP . . 500 undoped 13 Spacer InP - - 2000 p, 1E17 The devices utilize a reverse mesa ridge waveguide (RM-RWG) structure [8 ]. The RM is defined by using a special acid mixture, which etches parallel to the (111)A crystal plane, so that the angle of the mesa side is 54°44” with respect to the bottom surface. This structure has two main advantages. First, we can define narrow, single-mode RWGs (at the mesa bottom), while at the same time maintaining a wide mesa top. The wide contact area decreases the contact resistance and improves the thermal properties of the device. Second, the fabrication process and the alignment become more robust and easy. One of the most difficult steps during the fabrication process is how to align the polyimide (PY) opening, used for defining the metal contact, to the mesa top. The width of the mesa top used for RM-RWG devices was 5pm, and aligning the polyimide opening is not an easy task. First, the as-spun polyimide layer is 6 -8 pm 74 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. thick, which prevents accurate alignment using our contact aligners. Second, during the curing process the PY ‘flows’ and changes its dimensions, and sometimes the resultant opening is wider than the one before curing. Thus, a new approach was developed, which allows for easy alignment of the polyimide opening with much larger tolerances. Figure 4.18 SiNx mask overhanging on both sides of the RM-RWG laser mesa. In essence the process uses the properties of the H2SO4/H2O2/H2O solution, specifically that it undercuts the top contact InGaAs layer, and thus by varying the etching time we can achieve a different mesa width. Basically we start with a wider SiNx stripe (10pm) than the desired top mesa width (5pm) as shown in Figure 4.18. The dielectric stripe is then used as an alignment mark and etch-stop layer for the polyimide planarization process. The fabrication starts with performing a two-step SAG growth and deposition of a 2000A-thick SiNx layer on top of the wafer. A 10pm-wide photoresist stripe is 75 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. patterned and the SiNx mask is etched by CF4 REE plasma. After the solvent cleaning (TCE+ACE+ME) oxygen plasma is used to clean the surface from the PR residues. The next step is to wet-etch the mesa and form the RM-RWG structure. First a H 2 S O 4 / H 2 O 2 / H 2 O (1/1/3 at RT) solution is used to remove the top InGaAs contact layer. One of the properties of this etchant is that it undercuts the InGaAs layer below the SiNx mask. Thus, by varying the etching time and monitoring the undercut with an optical microscope we can achieve different effective mesa widths (color differences between the SiNx and undercut regions allow for precise control of the undercutting using an optical microscope). The process is interrupted when a 5pm mesa width is reached. The second step is to etch with HBr/H2P0 4 (1/1 @ RT for 3min) which forms the RM-RWG structure. The mixture etches parallel to the (lll)A facet, so the angle of the bottom of the mesa with respect to the substrate is 5 4 0 4 4 ” T^s solution does not undercut, so the width of the mesa top stays equal to the width of the InGaAs contact layer and constantly tapers when approaching the etch-stop layer. The timing is not critical and is important as far as it has to be enough to completely form the mesa - after that point the etching terminates. The profiles can be monitored by cleaving small pieces from the sample and observing the mesa cross-section with an optical microscope. The formula for determining the mesa width at the sample surface is [8 ]: (4.1) where, W su rf a c e is the width of the mesa at the sample surface and Wto p is the width of the mesa at the top, h is the cladding layer height. In our case the cladding height is 76 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1.5pm, the width of the mesa is 5 pm at the top, so we could obtain 2.8pm at the bottom. An example SEM picture showing the mesa cross-section in the middle of the SAG region is shown in Figure 4.18. Note that the SiNx mask is still in place and will be used as an etch-stop layer for the following polyimide process. Next polyimide is spun (2000rpm for 30s) and soft-baked at 90°C for 4min. A 4pm stripe is used for alignment of the opening. Note that because of the wider SiNx stripe (10pm) an easy alignment process is possible with larger tolerances. The PY is developed, rinsed and cured as described above (in the BA section). If there are lots of residues left on the SiNx mask surface, an 0 2 plasma cleaning process could be used. Note that even if the opening is not very well aligned to the mesa top, the SiNx will act as a stop layer for the developing and 0 2 plasma cleaning processes. Next, photoresist is spun and two-section, 100pm-wide metal contact pads are defined. A CF4 R3E plasma is used (lOOmtorr, 100W, 1.6min) to remove the SiNx mask, and p- metals (Ti/Pt/Au - 300/500/2000A) are evaporated. A SEM cross-sectional view of the finished device is shown in Figure 4.19. The RM-RWG structure is visible as well as the polyimide layer around the mesa and the metal contact on top of it. Note that the opening is even wider than the mesa top, but due to the used technique the metals are far from the sample surface. Even part of the SiNx mask, which was buried into the PY could be distinguish in this picture. During the spinning process, the PY fills the region below the dielectric mask ‘wings’ and forms a flat area, which after the mask removal is still present and prevents the junction shortage by the deposited metals. 77 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. K-'m.iui' ln'ii, ilk- S . \ i11» 1 . — I 2 |im 3 u + c o n t 1 1 Ou/1,5u/6u/6i Ik1 ' . 1 Figure 4.19 A SEM cross-sectional view of the processed RM-RWG device. The SiN„ mask is still embedded into the polyimide layer. 100 u m 8 fiWG » Pclyim ide * c o n ta c ts * ? s e c t 15 0 kV 3 I 2 !,Ox Figure 4.20 A SEM picture showing a top-view from the final RM-RWG device with polyimide layer and gold bond-pads. At this point the isolation trench is formed as previously described. Finally, the device is lapped to 1 0 0 pm in thickness by mechanical polishing, cleaned, and n- metals deposited (AuGe/Ni/Au - 1000/500/2000A). A rapid thermal annealing is 78 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. performed at 430°C as described above. The samples are cleaved in bars 1mm long and a two-layer dielectric AR coating is deposited on both facets to reduce the reflectivity. A SEM picture showing a top-view from the final device is Figure 4.20, where the metal contact pads for both sections, deposited on top of the PY layer, are visible. 4.6 References 1 M.Aoki, H.Sano, M .Suzuki, M.Takahashi, K.Uomi, and A.Takai, Electron. Lett, vol.27, 1991, pp.2138-2139 2 E.J. Thrush, M.A. Gibbon, J.P. Stagg, C.G. Cureton, CJ. Jones, R.E. Mallard, A.G. Norman, and G.R. Booker, J. Crystal Growth, vol. 124, 1992, pp.249 3 L. Silvestre, A. Ougazzaden, D. Delpart, A. Ramadane, C Daguest, and G. Patriarche, “Study of the Growth Rate and Composition Variations in Metalorganic Vapor Phase Selective Area Epitaxy at Atmospheric Pressure and Application to the Growth of Strained Layer DBR lasers”, J. Crystal Growth, vol. 170, 1997, pp.639-644. 4 T. Tanbun, Y.K. Chen, J.A. Grenko, E.K. Byrne, J.E. Johnson, R.A. Logan, A.Tate, A.M. Sergent, K.W. Wecht, P.F. Sciortine Jr., S.N.G. Chu, “Integrated DFB-DBR Laser Modulator Grown by Selective Area Metalorganic Vapor Phase Epitaxy Growth Technique”, J. Crystal Growth, vol. 145, 1994, pp. 902-906. 5 Marcalili, E.A.J, and Hardy, A., “The azimuthal effective-index method”, IEEE J. Quantum Electron. QE-24, 1988, pp.766-774 6 Lee, D.L., “Electromagnetic Principles o f Integrated Opics”, John Wiley ans Sons, New York, 1996 7 H.A. Haus, “Waves and Fields in Optoelectronics”, Prentice Hall, Englewood C liffs, NJ, 1984. 8 Masahiro Aoki, Masaaki Komori, Tomonobu Tsuchiya, Hiroshi Sato, Kouji Nakahara, and Kazuhisa Uomi, “InP-Based Reversed-Mesa Ridge-Waveguide Structures for High- Performance Long-Wavelength Laser Diodes”, J. Selected Topics in Quant. Electronics, vol. 3, no.2, April 1997, pp.672-683 79 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 5 Semiconductor optical equalizer 5.1 Results In this chapter we will discuss the experimental results for two-section broad area selective area growth (SAG) amplifiers. Buried heterostructure devices were also processed from SAG wafers, but the performance was not quite satisfactory. One of the main difficulties with this kind of configuration is the growth of high quality current blocking layers (CBL’s) with small leakage currents. The problem stems from the fact that in our experiments we used SAG patterns with different stripe widths. The different growth rate enhancement in the adjacent SAG regions gives rise to different thickness of the MQW layers. A current blocking layer could be optimized only for given thickness of the active structure. Thus, once grown, the CBLs will show good performance only for certain devices. This is plotted in Figure 5.1(b), where the IV characteristics of BH lasers processed from different SAG regions are compared. The leakage currents could be as small as 200pA at -2V bias for 600pm-long devices built from SAG regions with 20pm SiNx stripe width and as large as 6 mA for SAG regions with 50pm SiNx stripe. The LI curves are plotted on Figure 5.1(a), and the devices show a thermal roll-off at 2mW of output power per facet. Because of this nonuniformity, further process of BH devices was discontinued. Figure 5.2 shows the measured SAG enhancement factor, which is defined as the ratio of the thickness of the material in the SAG region between the dielectric stripes 80 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. to the thickness of the material in the plain region. The width of the dielectric stripes varied from 10pm to 50pm with 15pm openings (the place where the SAG region is formed). The maximum enhancement achieved was 120% for a 50pm stripe width 0.10 - BH-SAG sa m p le #2720 Stripe w idths - 0-50pm 0.08 rO .O O O -0.002 3 * 0 .0 6 -0.004 0.04 -0.006 •0.008 -2.0 -1.5 -1.0 -0.5 0.0 0.02 0.00 - 0.02 1 2 • 2 1 0 Sample #2720 - BH laser/ 2pm stripe i . i i —t __ _ „ t i , . 2.5- r 1.5- S . 1 0 ' O) 0 .5 - □ 0.0 0.00 0.05 0.10 0.15 0 .2 0 0.25 0.30 0.35 Current [A] Voltage [V] (a) (b) Figure 5.1 Buried heterostructure SAG lasers; (a) LI curve, (b) IV curves 120 100- 4 * ) c g 80- < D O C 60- m £ 40; 20 - 15 |rm I 0 40 50 10 20 30 0 Stripe Width w [pm] Figure 5.2 Growth enhancement ratio as a function of the dielectric stripe width w. The open space between the SiNx mask patterns is 15pm. Broad-area lasers were processed from the different SAG regions on the same wafer and the results are summarized in Figure 5.3. A linear increase in the lasing 81 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. wavelength is observed with increase in w, and the wavelength span was 150nm for a 40pm span in SAG stripe width. The target wavelength in the plain region was 1.52pm. This linear dependence completely agrees with the trend observed in Figure 5.2. The threshold currents are also plotted on the same graph, Figure 5.3(b). Up to a 35pm stripe width the threshold current value is almost constant, showing that the material quality is uniform across all of the SAG regions, regardless of the fact that the compressive strain is increased in the multiple quantum well structure with increase in w. The threshold current density is estimated to be around 110A/cm2 per well. 1 E 1 1 l 1 I 1 , S 1 3 1 1 Sample #2526 Sample #2626 55- 66 50- 64 45- 60 35- 30- 25- 20 40 40 30 is -1 0 0 0 & im ® - 800 nm A - 750 nm X - 600 |im ♦ - 500 |im * - 400 (im ® -1 0 0 0 iim ® - 800 ftm A - 750 |im X - 600 nm O - 500 |i m * - 400 nm 40- SAG stripe width [pm] (a) SAG stripe width [pm] (b) Figure 5.3 Dependence of the lasing wavelength (a) and the threshold current (b) on the dielectric stripe width w. The devices are broad area laser processed from the same wafer (#2526). The data from another growth ran is shown in Figure 5.4. This sample differs in the material composition, and the bandgap of the plain region is shifted towards shorter wavelengths, as required by the design described in Chapter 3. A linear trend with varying the SAG mask width is observed in the change of the lasing wavelength Ao. The target wavelength for the plain region was Xo= 1.49pm, and a red shift of AX 82 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. =160nm is observed, when using a SiNx mask with 50[tm stripe width. The threshold current density is uniform up to 30pm SAG stripe and has a value around 150A/cm2 per well. There is a sudden increase in Jth at longer wavelengths, due to the degraded material quality (large compressive strain), and the value could be as high as 900A/cm2 per well for 50pm stripe width. Sample # 2 9 4 0 Sample #2940 1,6 4 - E 1.62 w £ 1.60 § 1- i w 15[im 58- | 1.56- Zi. 1-541 0) ■ | 1.52-1 1 .50 -j ■^T 700 S . 300 S 200 10 20 30 40 50 SAG stripe width w [pm] (a) 10 20 30 40 50 SAG stripe width w [pm] (b) Figure 5.4 Dependence of the lasing wavelength (a) and the threshold current (b) on the dielectric stripe width w. The devices are broad area laser processed from the same wafer (#2940). Two-section optical amplifiers/equalizers were fabricated from the same wafer [1,2]. The section with the longer-wavelength gain peak is defined in the SAG region while the shorter-wavelength gain section is defined in the plan region outside the dielectric mask. The devices were characterized in a CW input regime and the pulse propagation properties were not explored. To avoid the heating effects, the two sections were separately pumped with Ips/lOkHz synchronized pulse currents. The light from a tunable CW diode laser was coupled in, using a lensed SM fiber, and coupled out by free space optics. The output signal from the detector was measured with a boxcar integrator, synchronized with the pump currents. Baseline substraction 83 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. was used in order to remove the component of the light coupled into the detector directly from the input fiber. SAG s tr ip e w id th w=10 microns £ Q 8 a .£ 6 a 4 0 SAG + plain region X w SA G re g io n o n ly \ l s A G l M \ SAG section plain section ^ = 1 5 2 0 nm ^ = 1 4 9 0 nm 1480 1 500 1520 1540 1 560 Wavelength [nm] (a) SAG s tr ip e w id th w = 2 5 m ic ro n s SAG + plain region SAG region only SAG section X =1560 nm X o=1490 148 0 150 0 1520 1 5 4 0 1560 Wavelength [nm] (c) SAG s tr ip e w id th w=15 microns SAG + plain re g io n 13- 12 - a n - 1 0 - SAG region only SAG plain __ _ section *,=1540 nm Aj=149° nm 1 480 1500 15 2 0 1540 1560 Wavelength [nm] (b) SAG strip e w idth w =30 m icrons 15- 12 £ Q a 9 - i S 6- Q . JE 3 o SAG + plain region X 1 SA G region only U | y SAG section plain section Y X - 1 580 nm X 0=149O nm 1480 1500 1520 1540 1560 Wavelength [nm] (d) Figure 5.5 Spectral responses of four two-section optical equalizers with different width of the SAG region: (a) w=10pm, (b) w=15pm, (c) vt=25pm, (d) w=30pm. The spectral responses of four different devices operated under different pump conditions are plotted in Figure 5.5. Their long-wavelength sections are processed in between different SAG regions with different stripe width. The graphs (from left to right) show the measured CW gain spectra of devices processed between SAG stripes with w=10, 15, 25, and 30pm. The difference between the bandgap 84 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. wavelengths of both sections of the separate devices is AA= 30nm, 50nm, 70nm and 90nm respectively. In these graphs, the response of the amplifier with only the main SAG section pumped is labeled ‘SAG region only’. The remaining curves on the same graphs are measured when both sections are pumped. The drive current of the short-wavelength section is constantly increased, while the drive current of the long-wavelength section is kept constant. Gain ripple is present on the plots because the antireflection coating was not optimized, especially at the short-wavelength part of the spectrum. A common trend is the equalization behavior, which is not so pronounced for devices processed from narrower SAG stripe regions. In contrast, the amplifier built from 30pm SAG region exhibits a bandgap wavelength difference of zLi=90nm and the equalization behavior is strong. Increasing the pump current in the short-wavelength section boosts the gain only in the short-wavelength part of the spectrum. When the difference between the bandgap wavelengths of the two sections is small (w=1 0 pm, AA= 30nm), varying the relative strengths of the currents does not influence the shape of the overall spectral response, but only broadens the effective gain bandwidth by 17nm for the case shown in Figure 5.5(a). Increasing the difference in the bandgap wavelengths of the two sections, leads to even broader spectrum (w=15, 25pm). In the limit of very large AA, a second gain peak will appear at short wavelengths and the equalization behavior is more pronounced (w=30pm). A theoretical simulation of the same device with w=30pm using the model described in Chapter 1, is plotted in Figure 5.6. Here we assume a bandgap 85 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. difference of zU=70nm between both sections, with an SAG section drive current of 7sAG=320mA and plain section drive currents of J/=70mA, 72= 100mA, 7j=150mA and /.*=200mA of the plain section. To compare the experimental data with the theoretical model we plot in Figure 5.7 the spectral response of the device fabricated from the portion of the wafer with 30pm SAG stripes, Figure 5.5(d), superimposed with the calculation from Figure 5.6. A fairly good agreement is obtained between the theory and experiment. By varying the magnitude of the current in the short- wavelength section a spectral equalization is obtained over more than zU= 1 0 0 nm, the measurement being restricted by the range of the tunable laser. Theoretical Sim ulation _J . I i 1 i I -----1 ___ .___L SAG s e c t io n X = 1 5 6 0 nm X . = 1 4 9 0 nm ,=320m A = 0 m A 70mA y -l ,----- ,----- ,----- , k, ,----- ,----- ,----- 1 ----- , ----- ,----- ,----- 1 ----- 1 - 1460 1480 1500 1520 1540 1560 1580 W avelength [nm] Figure 5.6 A theoretical simulation of a two-section SAG device, whose experimental data is shown in Figure 5.5(d). Figure 5.8 shows the saturation behavior of the same amplifier at 1=1.55pm with only the main section pumped. A saturation output power of lOdBm is measured. 86 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. SAG w = 30 m icro n s - C o m p arison T heory / E xperim ent 15 12-j' n a 9 . l SAG= 3 2 0 m A / l 2 =200mA SAG7" I r =320m A S A G plain s e c tio n s e c tio n X =1580 nm X =1490 nm I ,. =0m A p lain ___________________________________ _________ 1480 1500 1520 1540 1560 Wavelength [nm] Figure 5.7 Comparison between theory (dashed) and experiment (solid). The equalizer is processed from SAG regions with w=30pm stripe width and exhibit an equalized bandwidth of lOOnm. SAG stripe w=30 microns, A , =1550 nm m 2 , c 7 5 (5 a 1 2 1 0 8 6 - 4 2 -Isag=120 mA -Isag=200 mA -l.._=200 mA, theoretical SAC ’ I ____I I , . =0 p la in J = S A G p la in o s e c tio n s e c tio n X =1580 nm X =1490 nm -40 -30 -20 -10 0 10 Input Power [dBm] Figure 5.8 Saturation behavior of the amplifier, shown in Figure 5.7, at ,1=1.55pm with only the main section pumped. A saturation output power of lOdBm is measured. As already stated, the demonstrated amplifier has an inhomogeneous gain medium. By increasing the drive current in the short-wavelength section, the overall 87 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. gain spectral shape changes and the gain peak moves towards shorter wavelengths (see Figure 5.5). In general, the lasing wavelength of a Fabry-Perot laser is determined by the wavelength where the gain becomes equal to the loss in the cavity. Thus, by varying the current, a tunable semiconductor Fabry-Perot laser is demonstrated, as in Figure 5.9. \Mr,^ S pectrum of w=25pm SAG 'tunable' laser > 1 < < a < 'a < < 4 < > S < 'e plain a re a current 1.52 1.54 1.56 1.58 I 4 11 ' 2 - - SAG p la in t 5 i I it -4- r-d J H k 1.5 0 1.51 1.5 2 1 .5 3 1 .5 4 1 .5 5 1 .5 6 1 .5 7 1 .5 8 W avelength [pm] Figure 5.9 Tunable two-section Fabry-Perot laser. Pumping the plain area section changes in the overall device gain spectrum and thus shifts the lasing wavelength. The lasing wavelength could be tuned from 1.56pm to 1.52pm. Because the cavity is not single mode, multiple lasing peaks could be observed, and the experiment was performed just to demonstrate the device and prove the idea of wavelength tuning by changing the spectral properties of the active material. This kind of inhomogeneous gain medium could be used as active region for a tunable D F B laser, where in addition to the standard tuning of the resonant wavelength by 88 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. shifting the grating reflectivity, one could have ‘a second degree of freedom’ in the tuning by changing the spectral shape of the modal gain. 5.2 Conclusions A two-segment semiconductor optical amplifier/equalizer with current adjustable gain spectrum is demonstrated. Selective area growth techniques are used as a main enabler for the integration of both sections. CW gain equalization is obtained over a lOOnm range and the devices show 12dB of gain and lOdBm saturation power. This equalizer may play an important role in optical analog-to-digital converting systems. Another potential application for this spectrally inhomogeneous active region is as a gain media in wide-range tunable DFB laser applications. 5.3 References 1 Kostadin Dj ordjev, Sang Jun Choi, Won-Jin Choi, Seung June Choi and P. Daniel Dapkus, “Spectral Equalizers Based on Saturable Semiconductor Optical Amplifiers,” 14th Annual LEOS’2001 Meeting Conference Proceedings, November, 2001, pp. 103-104 2 Kostadin Djordjev, Sang-Jun Choi, Won-Jin Choi, Seung-June Choi, In Kim, and P.D.Dapkus, “Two-Segment Spectrally Inhomogeneous Traveling Wave Semiconductor Optical Amplifiers Applied to Spectral Equalization”, IEEE Phot. Technology Lett., vol. 14, no.5, May 2002, pp. 603-605. 89 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 6 Fabrication and measurement of vertically- coupled microdisk resonant devices 6.1 Introduction In this chapter we will describe the fabrication process of vertically-coupled microdisk resonant devices based on the InP material system [1,2]. A schematic drawing of such a device is shown in Figure 6.1(a). The vertical geometry [3] is chosen because it offers two major advantages compared to the lateral, namely: (i) the coupling coefficient can be precisely controlled by the epitaxial growth; (ii) the material composition of the waveguides and resonator can be optimized and grown independently. The latter advantage facilitates the design of active microdisk devices - ON/OFF switches, modulators, detectors and microdisk lasers. The InP system also allows incorporation of gain or electroabsorption regions into the microdisk cavity suitable for operation in the communication window around 1.55pm. I/O W aveguides R esonan t Disk P ost Transfer Substrate InP bottom cladding, d=1jxm ^ ^ Q - W G c o r e , d = 0 .4 n .m . /= 1 .2 |im Y/ InP separation, d=0.4-0.9pm //\ Q - d is k c o r e , « m W InP top cladding, d=1,5pm Q-etch-stop l a y e r ^ ^ ___ InP Substrate a) ........ b) Figure 6.1 Vertically coupled microdisk device with a post for improving the mechanical stability and current/field uniformity, (a), and the corresponding epi-structure, (b). 90 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Another approach for fabricating vertical microresonators was proposed [4], relying on polymer wafer bonding using BCB. In this approach the polymer applies strain to the structure, and is an insulating material, which is not acceptable for the design of active devices. In contrast, our process uses a thermal wafer-to-wafer bonding technique. The resultant structure is thermally-stable and strain-free, and the interface exhibits ohmic conduction. Figure 6.2 The post below the microdisk cavity improves: (a) mechanical stability, (b) current/field uniformity, and (c) forces single mode operation. Another innovative feature in our design is the circular post below the resonant cavity, as shown in Figure 6.2. Basically this is a second disk etched between the I/O waveguides. Its purpose is threefold: (i) it improves the mechanical stability of the whole 3D structure; (ii) it improves current/field uniformity in active devices; (iii) it acts as a limiter for the higher order modes improving the transmission characteristics. A microdisk resonant cavity supports whispering gallery modes, which propagate by means of total internal reflection from the disk/air interface. The first order mode has the smallest volume and occupies the outmost region of the disk in the radial direction. All of the higher order modes have larger volume, penetrate deeper inside the disk, and thus couple/leak into the post. Effectively these modes will have much smaller g-values and will be suppressed. The separation between 91 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. this post and the bus waveguides is large enough ( 1 .2 pm) to prevent direct coupling of light into it from the I/O waveguides. 6.2 The fabrication process An example epitaxial structure for passive devices is shown in Figure 6.1(b). The samples were grown by low-pressure MOCVD on a (001) InP substrate at 655°C. The growth starts with an InP buffer layer, followed by a 0.2pm-thick InGaAs etch- stop layer. The top InP disk cladding is lpm-thick in order to eliminate the absorption losses associated with the metal contacts (important for an active device). The growth continues with a 0.4pm-thick, /l= 1.2pm InGaAsP disk core layer, followed by an InP separation/coupling layer with different thickness, a 0.4pm-thick, J h = 1.2pm InGaAsP bus waveguide core layer, and a lpm-thick InP bus waveguides cladding layer. If an active microdisk device is to be grown, the disk cladding is p-doped, with the corresponding InP setback and contact layers included. The separation, waveguide core and cladding layers are n-doped. The disk core remains intrinsic, and a gain or electroabsorption active region may be incorporated into it. Details on these epitaxial structures can be found in Chapter 8 , where active resonators are introduced. The devices are processed by optical lithography using a conventional contact aligner (Karl-Suss MJB-3) and Shipley S1813 photoresist. Linewidths of 0.8pm are obtained reproducibly. A 1000A SiNx layer (deposited at 275°C, 30W, 450mtorr, 92 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. lOmin, SiH4/NH3/N2-40/20/60) is used as a dielectric mask throughout the process and the patterns are transferred from the photoresist to the SiNx with a low pressure (lOmTorr) reactive ion etching (REE) CF4 plasma. The mesas are defined by a REE dry etch using CEL/H^/Ar chemistry [5,6]. Disk Waveguide 1 1 0 Waveguides V 1 1 0 Waveguides '< Growth Substrate a) Waveguide Formation Resonant Disk Wafer Bonding 1 Transfer Substrate Substrate Removal / Disk Formation Figure 6.3 Vertical Coupler Fabrication Process Overview: (a) as-grown structure; (b) bus waveguide formation; (c) wafer bonding to transfer substrate; (d) substrate removal and disk formation. The process flow-chart is shown in Figure 6.3. The waveguide pattern is first defined and dry etched. Next the sample is flipped over and thermally bonded to another transfer substrate. The original wafer is removed by mechanical polishing and selective wet etching. Next the disk mesa is defined and dry etched and metal contacts are formed. We are going to discuss all of the steps separately. 6.2.1 Waveguide formation In this section we will discuss the formation of the bus waveguides, the circular post, and the alignment mark, used for aligning the microdisk mesa to the bus waveguides. First a 1000A SiNx layer is deposited on top of the growth substrate and 93 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the waveguides/post pattern is transferred to the SiNx mask. The waveguide width is 0.7jim (or 0.9|im depending on the mask pattern) below the disk and adiabatically tapers to 2.5jim for better coupling efficiency to a lensed fiber. The width of 0.7|im is chosen to assure phase matching to the fundamental disk mode and, at the same time, this is the minimum resolution achievable with our contact aligner. Within this step, one of the main difficulties is the photolithography, namely the definition of submicron dimensions. A design of experiments was performed to find the optimal process parameters and the photoresist baking time, the baking temperature and the time for developing were varied. It was found that Shipley S1813 photoresist (PR) performed better than the previously used Clariant AZ5214, and thus for positive lithography throughout the fabrication we used this brand. The results from the experiments showed that the baking time and the temperature are important parameters. Using longer times and higher temperatures, makes the PR harder and the resulting patterns exhibit rougher sidewalls. Nevertheless, this regime is useful for defining very fine patterns with good resolution. When baked for shorter times and at lower temperatures, the solvent from the PR does not evaporate completely and the photoresist is still ‘soft’ and can flow to a certain extent after the developing process. This regime, although with not very good resolution, is very useful for defining very smooth sidewalls and was used in microdisk mesa formation. In these experiments we observed the PR pattern with a scanning electron microscope (SEM), thus it was very difficult to quantify the 94 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. results. We will just describe the final recipes, which are different for different processes. The sample is baked at 120°C for 2min to remove moisture. First, for better adhesion of the photoresist to the SiNx layer, Silicon Resources’ AP405 adhesion promoter is used, which is designed for dielectric materials. AP405 is spun at 5krpm for 30s, followed by spinning of photoresist S 1813 at the same speed and time. At this speed, the resultant uniform PR layer is 1.24pm thick. The photoresist is carefully removed from the sample edges with a cotton swab immersed in acetone. This is done because around the edges the PR is accumulated and the layer is much thicker than in the center of the sample. If the edges are not cleaned, they will create nonuniform contact with the mask during the alignment, and thus poor pattern transfer. The PR is soft baked at 110°C for 4min, which completely removes the solvent and forms a hard and uniform layer. The sample is aligned and exposed with UV light with 170mJ of energy, measured at 405nm. It is very important to have good contact between the PR and the quartz mask. One of the main difficulties in defining submicron features, which are, at the same time, very close to each other (the separation between the post and the bus waveguide is 1 ,2 pm) is the proximity effect. The developing rate at the point where the separation between the patterns is small is different compared to the place where they are far apart. Thus, the developed PR pattern will exhibit unwanted features (sidewall striations) around the area, where the bus waveguide approaches the post, and will be very smooth elsewhere. Moreover, if 95 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the contact between the mask and the PR is not very good, due to an interference effect the size of the waveguides will be effectively enlarged. As a consequence, the post and the bus waveguides will merge and form a single pattern. Therefore a good contact is a necessity for obtaining an optimum image transfer. The PR is developed in Microposit MF321 developer for ~20s with very gentle agitation. It is better to leave some PR residue on the surface, which could be removed by O2 plasma treatment, then to over-develop the sample. An over developed sample will exhibit rough sidewalls with large striations. The wafer is rinsed in DI water for lmin and dried with a N2 gun. REE ashing is then applied for removing the residues (O2 plasma at 60W, 200mtorr, 0.45min). The sample is then soft-baked again (120°C, 40s) to smooth the PR mesa: the PR ‘flows’ and fills the sidewall striations. The waveguide pattern is then transferred to the SiNx layer. We used a low pressure (lOmtorr) REE CF4 plasma process. This low pressure was not obtainable in our conventional RTF (Plasma Technology) system, so we utilized the ECR (Plasma Quest Model 98) machine for this purpose. The process utilizes 30sscm of CF4 at 20°C, pressure of lOmtorr, power PR F =500W for 300s. Prior to the real etching, the chamber has to be properly conditioned with a long (30min) O2 cleaning followed by 20min of CF4 plasma for reproducible results. After the pattern is transferred to the SiNx layer, the sample is first cleaned with REE O2 plasma for lmin (100W, 150mtorr, lmin), the PR is removed with solvents 96 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (ACE+ME), rinsed for 5min in DI water, and than 0 2 plasma (100W, ISOmtoir, 5min) is applied to prepare the wafer surface for the following dry etch process. The waveguides are dry etched [5-7](to be discussed below) to a 1.6-pm depth, so that the bus core is completely etched through and a rib waveguide structure is formed. The etch rate/depth are monitored in situ with laser reflectometry. Next, the sample is cleaned from the polymers formed during the dry etch and the SiNx mask, by 0 2 plasma treatment (150W, ISOmtorr, 2min), followed by a dip in BOE (1:7) for 5 min with agitation. A SEM picture of the as-formed bus waveguides and the post, which will be positioned below the disk cavity, is shown in Figure 6.4. The mesa of the waveguides is vertical having a width of 0.8pm close to the post. The wafer surface is very smooth and clean, suitable for the wafer-bonding process. w a m m Figure 6.4 Bus waveguides and the circular post, which will be positioned below the microdisk. The mesa is 1.6pm deep after the dry etch with CEp/H^Ar chemistry. 97 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 6.2.2 Wafer Bonding To join wafers at the atomic level, the requirements of the wafer surface quality, cleanliness, roughness, and flatness are very stringent. There are three types of primary interactions responsible for the initial bonding. They are Van der Walls forces, electrostatic Columbic forces, and capillary forces. Van der Waals forces originate from the atomic or molecular electric dipoles that attract each other due to their correlated orientations. Depending on the chemistry used in the surface preparation step, different chemical radicals could be adsorbed to the sample surfaces. The van der Walls forces decrease very rapidly as the distance between two molecules increases. The electrostatic forces are usually very strong and dominate the interaction between two closely contacted bodies when charging occurs. However, the electrostatic forces can be easily screened or compensated by the presence of water or water vapor from the ambient environment. Therefore, these types of attractive forces are not important in our process. The water between the flat surfaces provides an additional attractive force— a capillary force. Due to the surface tension, the water between two surfaces pulls the samples together and smoothes out the imperfection of the sample surface. This increases the contact area significantly. 6.2.2.1 Cleaning of the samples for wafer bonding The cleanliness of the sample surface is critical to bonding uniformity. To minimize particle contamination on the samples surface, it is important to prepare them in a clean room environment. Generally in our process we used two pieces of 98 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. InP, approximately lcm 2 in size, or a little bit larger. The first piece is from a blank n-doped InP wafer, which is our transfer substrate. The second piece is the microdisk sample, with the bus waveguide pattern defined and etched on the top surface. Therefore, before the bonding itself, the surface is cleaned from the polymers left after the REE dry etch process with a BOE dip, as described in the previous section. The samples are next rinsed for 5 min in running DI water and cleaned with solvents (TCE, acetone, methanol - 5min each). Subsequently, they are rinsed again in DI water for 5min and are dipped in N H 4 O H / H 2 O (1:5) to remove the surface native oxide and particles. After a final DI water rinse (lOmin minimum), the surfaces of the samples are brought into contact under the water, to prevent oxidation and contamination. The sample pair is then taken out of the beaker and the excess water is gently blown away with a nitrogen gun. There still exists a very thin layer of water between the surfaces, which helps to hold the samples together and align the crystallographic axis. It is preferred for the transfer wafer to have slightly larger dimensions (l- 2 mm on each side) for easier alignment and handling. 6.2.2.2 Loading into the fixture The samples are then loaded into a graphite fixture still with the water at the interface. This type of approach is called “wet bonding”. One of the samples is with the waveguide pattern, which provides channels for the water to evaporate during the initial heating process. Since the microdisk wafer is not perfectly flat, it is important 99 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. that the fixture can press the samples together uniformly, and thus its surface is critical for this bonding process. Figure 6.5 shows a schematic of the fixture, designed in our lab by Chao-Kun (David) Lin [8 ]. The deformable surface is made of many tiny steel ball bearings, and it is believed that the balls would conform to the contacted surface and deliver uniform pressure to the wafers. The samples are loaded between both parts of the fixture, with a Si substrate on top of them. The Si substrate is used to keep the steel balls from contaminating the samples. After bringing both pieces of the fixture together, several hundred 1-mm in diameter steel balls are filled into the cavity in the top graphite part. A graphite screw is then driven into the fixture with a torque screw driver to apply metered presses upon the ball bearings against the Si substrate. Graphite screw Threaded rod -1.5” Si sub. Samples Figure 6.5 Schematic of the bonding fixture. The small steel balls are expected to distribute the applied pressure uniformly over the samples. 6.2.2.3 Loading into the bonding chamber The fixture with the samples is then loaded into a home-made furnace with H2 and N 2 gas supplies and a heater, controlled by a closed loop temperature controller. The chamber is evacuated by a mechanical pump and the samples are heated to 100 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 100°C for 10 minutes to remove the moisture at the interface. The chamber is then back filled with H2 and the temperature is ramped up to the desired bonding temperature (505°C) in another 5 minutes. The samples remain at high temperature for 30 minutes and then finally are slowly cooled down to room temperature in 15 minutes. An example of the bonding interface (the picture is taken after substrate removal) is shown in Figure 6 .6 , where the bus waveguides and an electron-transparent InP layer remaining after the microdisk mesa formation are visible. Very good and Ohmic contact is obtainable. Figure 6.6 Bus waveguides at the bonded interface and the remaining transparent InP layer after the etch of the microdisk mesa 6.2.3 Substrate removal and edge opening The third step is the substrate removal. The samples were mechanically polished and the remaining InP from the original substrate is completely removed by selective chemical wet etch (HCkH^O - 3:1), which stops at the first InGaAs etch-stop layer (the etching continues until a shiny surface, a result from the atomically flat etch stop layer, is observed). This layer is then removed by a HaSO^EkC^FfeO (1:1:3 @ RT) 101 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. selective etchant. At this point the top disk-cladding layer is exposed for further processing. If we process active devices, then a second InP etch stop layer is introduced in the epi-structure, which is removed at this point by a selective wet etch (HC1:H2 0 - 3:1). Now the top p+ contact layer is exposed for further process. The most critical step is the disk alignment and etch of the mesa. Misalignment of the disk mesa with respect to the I/O waveguides, will lead to asymmetric coupling and significant decrease in the output contrast ratio. For this purpose a small area (~lmm wide) around the sample edge is etched down to the waveguide pattern, which makes the waveguides, the posts, and the alignment marks defined with the first dry etch visible. Thus, the microdisk mesa could be precisely aligned to the bus waveguides using these alignment marks. To open the edge we first deposit a SiNx layer and spin a PR at 5krpm. The PR from the sample edges is cleaned with a cotton swab immersed in acetone. The SiNx is removed from the edge with a dip in BOE (1:7 for 3min) and the PR is stripped from the dielectric surface with solvents and RTE 0 2 plasma (100W, 150mtor, 5min). Now, an RIE CH4 plasma (to be described latter) is used to etch down the sample edges up to the waveguide pattern, so that the alignment marks become visible. The sample is cleaned from the polymers with 0 2 plasma (100W, ISOmtorr, 2min) and the SiNx is removed with a dip in BOE for 5min. At this point the top surface is ready for microdisk pattern formation. A new layer of 1000A SiNx, which will serve as a dielectric mask for the disk mesa formation, is deposited. 102 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 6.2.4 Microdisk mesa formation The next step is the disk mesa formation. The photolithography is performed again using S 1813 photoresist. AP405 adhesion promoter is spun at 5krpm for 30s followed by application of S 1813 PR at the same speed and time. The sample edges are cleaned with a cotton swab, so the waveguides and the alignment marks are visible for better alignment. The photoresist is baked for lmin at 105°C, and as described above, is left in a ‘soft’ state for obtaining smoother sidewalls. The disk mesa is aligned to the alignment marks and exposed with 150mJ at 405nm. Next, we follow the same process as the one used for the waveguide pattern formation: developing, ashing, second soft-bake, and etching of the SiNx layer by low-pressure R T F . CF4 plasma. The PR is removed by 0 2 RUB plasma (100W, 150mtorr, lmin), solvents (ACE+ME) and rinsing in DI water for 2min. The surface is prepared for the dry-etching of the mesa by cleaning with 0 2 plasma (100W, 200mtorr, 5min). 6.2.4.1 Dry etching process The most critical step in microdisk mesa formation is the dry etch process. The device is very sensitive to any kind of loss, with the losses due to the scattering from sidewall roughness being the main contribution. So any measures for decreasing the roughness have to be taken. Also the verticality of the sidewalls is very important. A tilted wall will introduce coupling between modes with different polarization and degrade the performance. 103 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. a) BCl3 /Ar chemistry b) Cl2 /CH4 /H2/Ar chemistry Figure 6.7 Etching experiments for obtaining vertical profiles. We utilized a design of experiments for finding the best chemistry suitable for our task, namely etching of vertical and very smooth mesas. First, we compared the dry etching chemistries available in our ECR system. We utilized BCF/Ar, C^/Ar, CI2/CH4/H2/AT and CH4/H2MX chemistries with different compositions. The first three require elevated temperatures of 150°C. The highest etch rate was obtained with BCl3 /Ar - Ifim/min. This chemistry results in a very smooth surface, but unfortunately we were not able to obtain vertical sidewalls with our ECR machine. The resultant profile was always undercut (Figure 6.7(a)) because of the high chemical etching component in the chemistry. The C^/Ar and CVCFLt/F^/Ar were also investigated but were not reproducible. The etched profile varied from run to run without any pattern. The selectivity between the InP and the SiNx mask was poor - the presence of chlorine in the plasma mixture causes fast erosion of the mask. The sidewalls were somehow rough, as shown in Figure 6.7(b). 104 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 6.8 SEM photograph of a microdisk mesa immediately after the dry etch. The mesa top and sidewalls are still covered with polymers. The best results were achieved by using CHU/H^Ar RBE plasma at room temperature. Because of the carbon related nature of this chemistry there is polymer formation over the mesa top and sidewalls, protecting them from further etching and undercut (Figure 6 .8 ). Thus, a smooth and vertical profile could be easily achieved and the mask erosion is practically zero. Afterwards the polymer can be removed by O2 plasma ashing followed by BOE etching. One of the disadvantages of this process is that the etch rate is very slow - 200-500A/min depending on the pressure, which is 50 times smaller then the one achieved with the BCVAr. The dry etching process was optimized leading to negligible mask erosion and extremely smooth mesa sidewalls. The polymer formation during the etching was controlled by carefully adjusting the CH4/H2 ratio in the RJE chamber [5-7], InP etching in the capacitively coupled RIE regime was investigated by applying high RF powers but without microwave and magnetic currents in the ECR system. The RF power is fixed at 495W throughout our experiments. It is found that the etch rate is strongly dependent on the chamber pressure but not significantly influenced by the 105 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. chemistry ratios or the absolute amount of CH4. For our etching tests we use H2 /Ar (30/6sccm) and the chamber pressure is fixed at 15mtorr. A transition of etching behavior occurs at the CH4 flow rate of 8 sccm in the given circumstances. When the CH4 flow rate is less than 8 seem, etch rates drop steeply. In this regime there is lack of CH4 radicals and thus polymer deposition. The lack of sidewall protection, that is supposed to be provided by polymers, results in severely undercut profiles. On the other hand, CH4 flow rates exceeding 8 seem generate redundant polymers without changing the etch rates as mentioned. Excessive polymers accumulated on dielectric masks or disk patterns making the effective masks rough and enlarged. The C H 4 flow rate of 8 seem, provides the steepest profile and is chosen for our actual etching process. Thus our RIE process, nick-named Low Pressure - ‘LP’, is as follows: pressure-15mtorr, RF power-495W, C H 4 / H 2 / A J - 8/30/6 seem, RT, etch rate- 260A/min. To avoid the thick polymer accumulation in the early stages of etching and reduce the sensitivity to surface contaminations, a high-pressure RIE using lower flow of CHUis introduced: CHf/H^Ar - 4/20/10 seem, pressure 75mTorr, RF power 495W at RT. However, the higher pressure results in undercutting, if used for a long time as the only process. Therefore, we developed a multistep REE etching technique involving high-pressure ~75mTorr RIE conditions for 10 min, followed by a lower pressure ~15mTorr etch to the completion of the structure - ‘HP + LP’ process. The overall etching rate is approximately 0.03 mm/min. 106 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ■ cc.V S p o t Magn WD I ---------------------------1 I pm 15,0 kV 2 1 25000k B.7 07um p assive. H H600)+RIE8(440G) Figure 6.9 SEM photos of smooth and vertical 2.3pm deep microdisk and microring mesas, etched with the optimized conditions. Example pictures microdisk and microring mesas are shown in Figure 6.9. The microring is lpm in width and 2.3pm deep. Very smooth and vertical mesas are fabricated repeatedly. The SEM we have did not have enough resolution to capture the sidewall roughness. Figure 6.10 A photograph of a microring mesa taken with a high resolution FE-SEM. The sidewalls are very smooth with r.m.s. of less than lOnm. 107 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Therefore, we used a high resolution FE-SEM and a photograph of a submicron microring mesa is shown in Figure 6.10. This picture reveals the very smooth mesa sidewall and substrate surface. R.m.s. roughness of less than lOnm is observed on the sidewalls, which is a result of the optimized RIE etching process. HHHHH Figure 6.11 A SEM picture showing the completed microdisk device. The bus waveguides are visible bellow the thin InP membrane left after the disk mesa formation. Figure 6.12 A SEM picture showing the completed microdisk device. The coupling layer is completely etched, so there is no membrane left. The alignment mark is also visible to the right. 108 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. At this point the fabrication of passive devices is accomplished. The structure is shown in Figure 6.11 and Figure 6.12, where the microdisk is vertically coupled to the bus waveguides. The top and bottom alignment marks are visible on the right. In Figure 6.11 a thin, transparent InP membrane is left after the fabrication process. This thin membrane helps improve the alignment tolerances, as the coupling coefficient is not so sensitive to a slight misalignment [4] of the microdisk with respect to the I/O waveguides with the membrane present. We will show that the thickness of the membrane is important for decreasing the cavity loss. If it is more than 0.3pm, then an effective coupling of energy from the resonant cavity to the membrane occurs and the loss increases. Thus, sometimes, it is necessary to remove it completely. This reveals the bus waveguides, bonded to the transfer substrate. Figure 6.12 shows that the dropped waveguide is designed to be bent, so we can measure both the transmitted and dropped ports at the sample output. 6.2.5 Polyimide planarization and metal contacts formation At this point the fabrication of passive devices is completed. If the task is to process active microdisk resonators, then we have to define metal contacts and bond- pads. The requirement is that the bond-pads have to be far from the disk mesa, and thus suitable for wire bonding. This configuration would also exhibit low optical loss, as the optical field will not interact with the metal. Moreover, the parasitic capacitance has to be small enough and should not limit the frequency response of the device. 109 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. For this purpose we used a layer of polyimide material to elevate the bond-pad to 3pm above the substrate surface. The polyimide process is as follows: First we deposit 2500A of SiNx to improve the mechanical stability and to protect the top surface from contaminants. An image reversal photolithography is used to remove the SiNx layer from the mesa top. AZ5214 is spun three times at 3krpm and then baked for 30s at 120°C. Disk patterns with radii 1.5pm smaller than the radii of the microdisk mesas are aligned and exposed with 70mJ at 405nm. The sample is further post-baked to cross-link the PR at 120°C for lmin. Then, a flood exposure is applied without any mask with energy of 250mJ at 405nm. The PR is developed in AZ400K developer (1:4) for 7s or more and the pattern is monitored with an optical microscope. We have a certain control on the size of the opening by varying the development time. After the p-contacts pattern is formed the SiNx from the mesa top is removed by a dip in BOE for 5min and DI water rinse. Next, the devices are loaded into an e- beam metal evaporator (Edwards BJD-1800) and the first p-metals are deposited- Ti/Pt/Au (300A/500A/1500A). A lift-off process is performed by dipping the samples into acetone followed by cleaning with methanol. Another layer of 500A SiNx is deposited to protect the metal contacts from the following polyimide process. HD Microsystems PI-2737 photosensitive polyimide is spun at 1800 rpm and soft baked at 90C for 4min in an oven. After the soft bake the thickness of the layer is more than 6 pm. The same disk pattern, which was used for p-metal definition, is aligned again on top of the mesa and UV exposed with 150mJ of energy at 405nm. 110 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This is negative tone lithography, so the unexposed areas of the polyimide could be removed by developing in DE-9040 polyimide developer for 20s. The sample is rinsed in RI-9180 polyimide rinse and dried with a N 2 gun. The polyimide is then cured in a N 2 atmosphere by ramping the temperature from 20°C to 350C at 10°C/min rate and then curing at 350°C for 30min. If the top of the mesa has some PY residues after the curing process, they can be cleaned with 0 2 plasma treatment (repeat 100W; 200mtorr; lmin; + observe, until the surface is cleaned). An example photograph of a polyimide layer covering the sample surface and the microdisk top contact layer completely opened and ready for bond-pad definition is shown in Figure 6.13. Figure 6.13 An example photograph of a polyimide layer covering the sample surface and the microdisk top contact layer is completely open and ready for bond-pad definition. After the planarization process, the top bond-pad metal could be deposited. For this purpose AZ-5214 photoresist is used again, with the same image reversal 111 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. procedure described above, to define the bond-pad pattern. The image reversal process forms an undercut profile, which will help during the metal lift-off process. After defining the PR pattern, the SiNx layer protecting the mesa top is removed with CF4 plasma (100W, lOOmtorr, O.Smin). Bond-pad metals are deposited (Ti/Au 300A/3000A) by e-beam evaporation at 5xl0"7 torr base pressure. The sample is dipped in ACE to perform the lift-off process and then cleaned with ME. Photographs of samples at this stage of the process are shown in Figure 6.14. These are completed active microdisk devices. The bond-pad is far from the disk mesa improving the mechanical stability and decreasing the capacitance. On the right, a cross-sectional view of a completed device is shown. The post and the bus waveguide are also visible. G o 'c C c r .ta c - . * bond pad Figure 6.14 Photographs of completed active microdisk devices. The bond-pad is far from the disk mesa improving the mechanical stability and decreasing the capacitance. On the right a cross- sectional view through the disk mesa is shown. The post and the bus waveguide are visible 6.2.6 Substrate thinning, n-contact definition, and mounting. The next step is cleaving the devices into individual bars, which could be mounted and measured. In order to achieve good cleaving with high quality facets, the samples have to be thinned by mechanical polishing. A 5pm alumina polishing 112 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. powder is used for this purpose and the backside of the sample is polished manually to 100pm thickness. The devices are cleaned by TCE+ME+ACE and rinsed in DI water for 2 min. Next, they are attached to a microscope glass slide with vacuum tape and n- metals (AuGe/Ni/Au - 1000A/300A/2000A) are deposited by e-beam evaporator. In order to form an Ohmic interface between the semiconductor and the metals, the samples have to be annealed with an RTA system. The temperature is ramped from 20 to 430°C at 5°C/s rate, kept at this temperature for 30s, then decreased down to 100°C at the same rate of 5°C/s. Reflectivity 1.86 o 1.84 1.82 V- 1.8 O > 1.78 \ O o £ 1.76 o o o to 1.74 — 1.72 \ 1.68 o o U 1 1.66 0.15 0.175 0.2 0.225 0.25 0.275 0.3 AR layer thickness [p m] Figure 6.15 Reflectivity as a function of the refractive index and the thickness of a single layer coating. At the end the samples are cleaved into bars 600pm wide. They are mounted on copper mounts, 1x1 O x 20mm in size, with conducting epoxy for device characterization. A single layer AR coating (SiOx, n=1.76, t=0.22pm) is deposited on 113 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. both facets. Figure 6.15 shows a calculation (performed under plane-wave approximation) of the modal reflectivity as a function of the dielectric layer index and thickness. The waveguide modal index is assumed to be 3.1 and the wavelength is A,=1.55pm. It can be seen that in order to achieve 10"4 reflectivity, the refractive index of the dielectric layer has to be n=1.76±0.1 and the thickness has to be t=0.22±Q.003pm. The SiOx layer is deposited with an e-beam dielectric deposition system, Sloan n SL1800, at 8x10' torr background pressure. The material is a non-stoichiometric oxide. Due to partial oxidation as a result of reaction with oxygen in the deposition chamber, the refractive index of the SiOx layer is strongly dependent on the background pressure, the deposition rate (2A/s in our case) and the presence of O2 in the chamber. Usually during the deposition process, the background pressure was changing. Therefore, to increase the run-to-run reproducibility of the refractive index, a small amount of background 0 2 is introduced into the chamber, during the deposition process, with the pressure gauge reading fixed at 2xl0'5torr. The thickness of the layer was monitored in situ with reflectometry [9]. The light from a broadband source, filtered at 1.55pm, is reflected from the surface of a reference sample (a Si piece introduced into the chamber only for reflectivity measurements) and detected by a detector. The deposition is stopped when the reflectivity (the signal from the detector) is at a minimum, which corresponds to a quarter wavelength thickness of the deposited material at the filtered wavelength. 114 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. If the processed sample is to be used for high-speed measurements, then we must mount it on an RF sub-mount, suitable for high-speed measurements and probing. We used a AI2Q3 substrate, shown schematically in Figure 6.16. The mount has a microwave strip-line (G-S-G) at one end and a terminal, 5 0 0 integrated load at the other end. The bar with microdisk resonators is cleaved into separate devices. They are mounted with conducting epoxy on the area in the middle between the strip-line and the load, as shown in the same figure. microwave strip line microdisk 5 0 0 GJ i G ..... / / / / / / * f / c / * /if v . r Figure 6.16 A schematic drawing of an active microdisk device mounted on a high-speed test mount. The mount has a 5 0 0 microwave strip-line and a 5 0 0 integrated resistor. m \ Vend :-a :-r 'fcts t f d'oope? Figure 6.17 A picture of an active microdisk device mounted on a high-speed moun ' e microwave strip-line is visible on the left and the terminal load is on the right. 115 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Next, the resonator is simultaneously wire-bonded to the end of the strip-line and the load. A picture of a mounted microdisk modulator is shown in Figure 6.17. The strip-line, the load and the wire-bond are visible. The high-speed measurements are performed with a S-G-S RF probe (0-40GHz) with 200pm pitch (the distance between the signal, S, and ground, G, tips). 6.3 Measurement The test setup for measurement of the transmission properties of the microdisk is shown in Figure 6.18. We use a New Focus 63xx Velocity tunable CW laser diode in the wavelength range from 1540nm to 1610nm as a source. The diode has an internal feedback loop which assures constant power mode throughout the entire wavelength interval. The sample mounted on the copper bar is attached to a Melles Griot stage with a temperature-controlled heat sink. Two additional stages (piezo-controlled for fine adjustments) are used for coupling light in, by using lensed fiber, and out by using a microscope objective and free space alignment. The detector is an HP Lightwave multimeter HP8153A and a computer, utilizing the LabView interface, is used for automatically tuning the laser and collecting the data. Microscope T ID er Objective Figure 6.18 Test setup for microdisk characterization. 116 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. If an active device is measured, then a HP4145B semiconductor parameter analyzer is used for device biasing with voltage or current. For high-speed measurements, we used a HP8510 network analyzer (0-50GHz) with PicoProbe high-speed probes (0-40GHz), mounted on an additional stage. 6.4 References 1 Kostadin Djordjev, Seung-June Choi, Sang-Jun Choi and P.D. Dapkus, “High-Q Vertically-Coupled InP Microdisk Resonators”, IEEE Phot. Technology Lett., vol. 14, no.3, pp.331-333, March 2002. 2 Kostadin Djordjev, Seung June Choi, Sang Jun Choi, and P. Daniel Dapkus, ‘High-Q, Vertically-Coupled Microresonators Built by Wafer-Bonding Technique’, 14th Annual LEOS’2001 Meeting Conference Proceedings, November, 2001., pp.509-510 3 D.V. Tishinin, P.D. Dapkus, A.E. Bond, I. Kim, C.K. Lin, and J.O'Brien, “Vertical resonant couplers with precise coupling efficiency control fabricated by wafer bonding,” IEEE Phot. Technology Lett., vol. 11, pp. 1003 -1005, August 1999 4 P.P.Absil, J.V.Hryniewicz, B.E.Little, F.G.Johnson, K.J.Ritter, and P.T.Ho, “Vertically coupled microring resonators using polymer wafer bonding,” IEEE Phot. Techn. Letters, vol. 13, no. 1, pp. 49-51, January 2001. 5 Seung June Choi, Kostadin Djordjev, Sang Jun Choi, and P. Daniel Dapkus, “CH4 -Based Dry Etching of High Q InP Microdisks’, J. Vac. Sci. Technol. B, vol.20, no.l, pp.301-305, Jan/Feb 2002. 6 Seung June Choi, Kostadin Djordjev, Sang Jun Choi, and P. Daniel Dapkus, 'Smooth, CHrBased Dry Etching of InP Microdisks', SCCVAS Leading Edge Student Symposium, Anaheim (CA), October, 2001 7 Seung June Choi, Kostadin Djordjev, Sang Jun Choi and P. Daniel Dapkus, ‘Fabrication of Vertically Coupled InP Microdisk Resonators by using Smooth, CH4-based Reactive Ion Etching Methods’, 14th Annual LEOS’2001 M eeting Conference Proceedings, November, 2001., pp.628-629 8 Chao-Kun Lin, “Wafer-Bonded Bottom-Emitting 850nm VCSELS for Short Distance Free-Space Optical Interconnects”, PhD. Dissertation, USC, December 1999 117 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 9 Fan, Riantm Verdiell and Dagenias: “Real-time insuty monitoring of Ar coating for SOA by ellipsomtry”, IEEE Phot. Technology Lett., vol.4, no9, September 1992 118 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 7 Passive microdisk devices: design and performance parameters. 7.1 Introduction In this chapter we will study high-Q vertically coupled microdisk resonators based on the InP material system (Figure 7.1) and the parameters influencing their performance. The vertical coupling geometry [1] is chosen because it offers two major advantages compared to the lateral coupling scheme, namely: (i) the coupling coefficient can be precisely controlled by the epitaxial growth; (ii) the material composition of the waveguides and resonator can be optimized and grown independently. I/O Waveguides Disk 'we M embrane Post Transfer Substrate a) b) Figure 7.1 Schematic diagram of a vertically coupled microdisk resonator with a post a), and definition of the parameters of interest in this study: disk radius R\ thickness of the coupling layer dc; waveguide etch depth dW c\ thickness of the remaining thin membrane t. In order to design high-Q, high-contrast-ratio resonators, different phenomena influencing their performance have to be understood. In this study we investigate how different dimensions of the structure affect the performance of the microresonators. We vary the coupling separation dc, the disk radius R, the etch 119 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. depth of the bus waveguides dwG, and the thickness of the thin InP membrane between the waveguides and resonators that remains after fabrication, t, Figure 7.1(b). The experimental results are compared with numerical simulations obtained from a full-vectorial mode solver. 7.2 Structural parameters of interest The epitaxial structure is comprised of two vertically stacked waveguides with a common coupling/confinement layer, as given in Table 7.1. Table 7.1 An example epi-structure of a passive microdisk. Layer Description Material X /Strain d[A] 1 0 Bus cladding InP - 1 0 , 0 0 0 9 Bus core InGaAsP 1 .2 / 0 % 4,000 8 Coupling Layer InP - dc 7 Disk core InGaAsP 1 .2 / 0 % 4,000 6 Disk top cladding InP - 1 0 , 0 0 0 5 Etch stop 3 InGaAs - 2 , 0 0 0 4 Etch stop 2 InP - 2 , 0 0 0 3 Etch stop 1 InGaAs - 2 , 0 0 0 2 Buffer InP - 1 0 0 0 1 Substrate InP - - The first-grown waveguide structure, which will eventually be used as the resonator, consists of an InP buffer layer, followed by etch-stop layers, a 1-|im-thick InP cladding layer, a 0.4-jim ^=1.2jUlm InGaAsP disk core layer, followed by an InP separation/coupling layer with variable thickness, dc. The second waveguide 120 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. structure consists of the common separation/coupling layer, a 0.4-pm .4= 1.2pm InGaAsP waveguide core layer, and a 1-pm InP cladding layer. The waveguides are 1-pm wide in the region below the disk and are adiabatically tapered to 2.5pm near the edge of the device for better coupling efficiency to a lensed fiber. The waveguides were dry etched to variable depth, dwG, which was monitored in situ with laser reflectometry. See Figure 7.1(b) for details of the device geometry. Figure 7.2 An example, TE, transmission characteristic of a microdisk with radius 12p.m. Sharp and deep dips are observed, with a quality factor in excess of 7,000. In the insert a magnified resonance around X= 1.55pm is shown with FWHM Al=0.22nm. An example TE transmission characteristic of a microdisk with radius i?=12pm is plotted in Figure 7.2. Sharp and deep resonance peaks are observed with linewidths as narrow as AX = 0.22 nm, indicative of a quality factor in excess of 7,000 [2, 3], The insert shows a magnified dip around >1=1.5 5 pm to demonstrate the linewidth. 0.22nm Disk R=12 pm, Separation dc =0.8pm, Q>7000 < C ' A E 0.2 0 - If f v. \ ■ 0.00 1555.01555.21555.41555.6 1550 1560 1570 1580 1590 Wavelength [nm] 121 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. The FSR of the same device is lOnm and the finesse is 50. The TE and TM resonance peaks are shifted by zW=5nm in the vicinity of /£=1.55fim for devices with R=12\im, see Figure 7.3. This shift is expected, since the effective index of the disk mode is different for the two polarizations. This difference becomes larger at smaller radii (see Figure 7.6). Disk R=12 pm, Separation d =0.7 pm. AX=0.58nm Q =2600 * 1 .10 - * . . w . °-9° i i w a ! W 0.80 £ 0-70 t o 0 .5 0 - 5 0.40- E 0.30 2 0.20 1550 1560 1570 1580 1590 1600 1610 Wavelength [nm] Figure 7.3 The TE and TM resonance wavelengths are different by d2=5nm in the vicinity of ,4=1.55pm, because of the different mode effective index. In this study [4] we vary the coupling strength by growing samples with different separation between the I/O waveguides and the disk core layers, dc. Samples with dc=0A, 0.5, 0.7, 0.8, 0.9Jim were fabricated. The second parameter of interest is the disk radius. The devices tested are microdisks with radii of R=4, 6 , 8 , 10 and 12|im. Finally, we also varied the etch depth of the I/O waveguides, dwo, which represents the distance between the core layer of the waveguides and the thin membrane. On the following graphs the origin of the coordinate system is chosen to be at the end of the 122 R eproduced with perm ission of the copyright owner. Further reproduction prohibited without perm ission. bus waveguide core layer. Thus, a positive dwG means that the core layer is completely etched through, and a negative dwG means that the etching stops somewhere inside the waveguide core layer. The fourth parameter in this study is the thickness of the remaining InP membrane, t. 7.3 Properties of a microdisk resonant cavity A microdisk resonant cavity supports whispering gallery modes, which are traveling waves propagating by means of total internal reflection from the disk/air interface. The first order mode has the smallest volume and occupies the outmost region of the disk in the radial direction. All of the higher order modes have larger volume and penetrate deeper inside the cavity. When a bus waveguide is coupled to the disk cavity, it alters its performance, changes the resonant frequency and decreases the quality factor. A simple description of the coupled disk/waveguides system can be performed using coupling of modes in time theory (CMT), see Appendix 4 for details. CMT is a perturbational approach appropriate in the limit of small losses. Thus, all possible loss mechanisms could be simply added as linear terms to the rate equation. The main equation derived in the appendix is the one describing the dependence of the transmission on the cavity loss: t (7.1) 123 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Eqn. (7.1) is a simple and useful tool for understanding the resonant behavior of a microdisk coupled to I/O waveguides. At resonance, in order to achieve 100% (7=0) power extraction from the input waveguide the resonator has to be imbalanced, i.e. < 2 / _1 + Q~l = Q~x, and thus even in the presence of loss the device can achieve a large extinction ratio. It is useful to remember that any loss in the cavity will diminish the total Q of the resonator. It will also decrease the amount of light that is resonantly transferred to the dropped port, even if the input is critically coupled. Eqn.(7.1) is derived using two major assumptions, namely that the waveguides are single mode and that the phase matching between the disk and the bus waveguides is maintained at all times. In our experiments we vary different structural parameters, which alter the phase matching conditions, and thus a decrease of the power extraction efficiency is expected from the one derived from Eqn.(7.1), as some of these parameters are varied. The aim of this chapter, as already mentioned, is to investigate how the main properties of the coupled microdisk/waveguides system, namely the quality factor and the power extraction efficiency, are affected by varying different structural parameters. It will be beneficial to consider, in advance, the trends we might expect from these quantities. The disk resonator, as with any kind of resonant cavity, has a quality factor proportional to the photon lifetime, Q = O)0T h = cd0t rt / LR T . The photon lifetime T P h , on the other hand, is proportional to the round trip time, Trt , (or the cavity length 2 nR/vg , where vg is the group velocity) and inversely proportional 124 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. to the round trip loss, Lrt- Thus, as far as the interest is only in the functional dependence, the quality factor can be written as, Q~R/Lr t. We can categorize the loss into two major groups: distributed loss and localized loss. 7.3.1 Distributed loss The loss can be characterized by a distributed absorption coefficient, a [cm1]. In the limit of small a 2 nR , the round trip loss LR t can be expressed as Lrt = 1 - ex p (-a27[R) ~ ccR, and thus the quality factor is Q ~ RlLR T ~ H a. When the loss is only due to the material absorption, a = a h is a constant, and the quality factor is thus expected to be a constant, Qb = const(R) , independent of the resonator radius (neglecting the week dependence of the mode effective index on R). However, this is not true if a is caused by scattering from sidewall roughness into the radiation modes. It can be shown that o th c scales inversely with R, a s c - H R . This is shown in [5], where a perturbational ‘Fermi’s golden rule’ approach is undertaken, in [6 ], where a volume current method is used, and in [7], where the Rayleigh scattering in dielectric microspheres is considered. Therefore, the quality factor limited by the scattering from sidewall roughness is expected to increase linearly with the disk radius, Qsc ~ R . The surface roughness may also cause coupling between the forward and backward propagating waves, if the correlation length of the surface perturbation phase matches both modes. This leads to splitting of the degeneracy existing in the 125 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. resonant frequencies of both modes by 8 ( 0, proportional to the strength of the coupling, and the measured transfer characteristics will exhibit double peaks. This will effectively lower the Q value, but will depend on the details of the surface roughness and will not be considered hereafter. 7.3.2 Localized loss Let’s assume that the source of the energy loss is localized only at a given place inside the cavity. Then the mode loses a certain amount of energy only once per round trip, and thus Lrt is constant with respect to the cavity dimensions, Lrt = const(R) . The quality factor of this kind of system scales linearly with the disk radius, Q ~ R/const ~ R . A microdisk resonant cavity views the coupling to the bus waveguides as a localized loss, which is equivalent to the mirror reflectivity in a Fabry-Perot cavity, r = VI—k , where K is the power coupling coefficient. The resonant mode loses energy every time it passes close to the waveguides. Thus, in the ideal case when the coupling strength is kept constant when changing R, the coupling limited Q should depend linearly on the radius, Qt ~ R. However, in our experiments the separation distance is fixed by the epitaxial growth, and K has a square root dependence on R [8 ]. This relationship is calculated in the slab waveguide approximation and accounts for the effect of increased interaction length between the I/O waveguides and the microdisk resonator as the radius of the disk increases. Thus, the round trip loss can 126 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. be expressed as LR T ~ k ~ s [ r and the coupling limited Q dependence is easily obtained as: Qt ~ R I^R ~ 4 r . 7.4 Dependence on the disk radius, R In this section we describe experiments that examine the relationship between Q and R. Disks with different radii, R=4, 6 , 8 , 10, 12pm, were fabricated on substrates with different coupling distances, dc. The results from the measurement of devices with dc =0.1 and 0.8pm are shown in Figure 7.4(a). In this and in all of the following experimental graphs we choose to plot the maximum values of the measured quantity and with vertical lines we show the spread in the data. Our belief is that the maximum data values represent “optimized” results, i.e. results from “ideal” devices utilizing the specific configuration, which are suitable for theoretically studying the trends. Several possible random variations in processing can result in lower values of Q. These may include defects in the resonator and variations in the width of the waveguides that will influence the coupling coefficient and phase matching. The higher the Q of the resonator the more sensitive it will be to these variations. Thus, the lower values of Q or the extraction efficiency that result from these random defects in fabrication will obscure the understanding of the device performance. Q increases with the disk radius. From the discussion above it can be inferred that it may be limited either by the coupling to the bus waveguides, which is expected to have a square root dependence on R, or by the scattering from the surface roughness, which is expected to have a linear relationship. The strong 127 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. dependence of Q on the separation, which will be discussed in the next section, favors the first choice. * - d =0.7 nm ; d =0.8 |xm ; TE -8 0 *= 4 6 8 10 12 4 6 8 10 12 Disk radius R [pm] Disk Radius R [pm] Figure 7.4 Dependence of the quality factor, Q (a), and the power extraction efficiency (b) on the disk radius, R. The coupling separation between the disk and bus waveguides core layers is cf=0.7pm and 0.8pm. To understand the behavior of the power extraction efficiency, shown in Figure 7.4(b), a numerical solution of the effective index of the first disk mode is performed. Often the conformal transformation approach is employed [9] for this purpose and the method is exact for one dimension. If used for 2D structures, this technique has to be combined with the effective index method, leading to approximate results. In this thesis we apply a full-vectorial mode solver for curved waveguides, which solves Maxwell’s equations exactly in cylindrical coordinates. Figure 7.5 shows the calculated distribution of the first two TE disk modes. In these graphs the contour line represents the cross-section of the disk cavity/core and the bus waveguide. The field is positioned very close to the semiconductor/air interface. 128 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 7.5 Distribution of the first two modes in /?=12pm disk. The contour line represents the cross- section of the disk cavity/core and the bus waveguide. The field is positioned very close to the semiconductor/air interface. The calculated effective indices of the first quasi-TE and TM modes in a microdisk cavity with different radii are shown Figure 7.6. With a decrease of the radius, the difference between the effective indices of the TE and TM modes increases, owing to the different portion of the fields that extend into air. For this experiment the calculated effective index of the bus waveguide mode is nw g=3.10. d=0.8pm; t=0.2|im; X=1.55pm 3.17-1 ..................................... , 1 , ! , 1 , 1 . .1 .--- X d) TJ 3.12- c I 3.07- «3 j j 3.02- 1 / - UJ / af ® 2.97- / / O 1 2 2.92- jg 1 — • — ne ( f TM polarization m i - - n „ TE polarization Q 2.87- 1 e f T ■ 2.82- T » l"'«.......‘"’ I ..------- 1 -- 1 -- r-T -r'*T -1| * 0 2 4 6 8 10 12 14 16 18 20 22 Disk radius R [pm] Figure 7.6 Effective indices of the first TE and TM disk modes as a function of disk radius, R. At small radii the difference in the effective indices between both polarizations becomes larger. An increase of the power extraction efficiency is observed with the increase of the disk radius with a maximum value of 97% at R= 12|im. This dependence may 129 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. result from one of two different phenomena. The first is the phase matching of the disk mode to the I/O waveguide mode, which is critical for obtaining large contrast ratios. From Figure 7.6 it can be seen that the disk mode has an effective index of 3.10 at radius of i?=13pm. Thus, absolute phase synchronism could be achieved only for this radius, and a decrease of R should decrease the power extraction efficiency. The second phenomenon is the power balance in the coupled |JL-disk/waveguides system. As already stated in the previous section, zero transmission (100% power extraction efficiency) could be obtained only when Q [x + QJ 4 = Q~l , i.e. when the power coupled into the resonator is completely balanced by the power coupled out of it, and the power lost due to the internal absorption. Change in the disk radius changes the coupling coefficient (by means of increasing the interaction length) and the losses due to the scattering from sidewall roughness by a different amount, thereby altering the power balance. In our experiments it is difficult to distinguish between both phenomena and a set of devices with larger span of R will help in finding the exact phase matching conditions. 7.5 Dependence on the separation distance dc The coupling coefficient between the waveguides and the disk cavity is exponentially dependent on the separation, shown in Figure 7.7, as calculated in the slab waveguide approximation. The other losses are not expected to change for disks of the same radius. Increasing the separation decreases K exponentially, and if the quality factor is coupling limited, this should lead to an exponential increase of Q - 130 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the trend observed in Figure 7.8(a). In this figure the quality factors and the power extraction efficiencies for microdisks with radii R= 8 and 12pm are plotted as a function of the coupling distance, dc. Note that, as discussed before, the disks with smaller radius have lower Q value, but have the same dependence on dc. < r» . y C o o st s o OS c Q, 3 o o S a r a © £ o 0 . 8 ■ - R=4pm - - R= 6 pm - - R=8 pm • - R=10 pm - - R=12 pm 7 6 5 4 3 2 1 0 0.7 0.8 0.9 0.4 0.5 0.6 Separation d [pm] Figure 7.7 The coupling coefficient k, as a function of the separation distance dc. The calculation is performed using a slab waveguide approximation and is discussed in [8] The devices with 0.9pm coupling separation exhibit a somewhat strange behavior, as the quality factor suddenly drops. One would expect the total Q of the system to be determined by the highest loss mechanism, 1 !Qm = H Q i + l l Q t ^ l Q d. Thus, in the limit of large separation (large Qt, Qd) the total Qm should saturate at the value determined by the internal losses of the cavity, Qi. The observed sudden drop in the Q value is an indication that some other mechanisms are playing a role. We believe that the difference in sample quality is 131 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the main contribution to this behavior, because the sample with Q.9|im coupling separation was grown at a much earlier time then the other samples. x103 * - R=12pm 8 - ! I i I i I i I i I i L . 7-| 6 0 5 ts « . U L 4 | 3 . < 0 3 0 2 1-1 0 I 1 T t a) * - R=8pm ; TE j i i i i i i i i i i 1-1 0 0 ? I b) -90 -80 £2, C -70.2 0 60 2 1 1-40 § a, -30 -20 0.4 0.5 0.6 0.7 0.8 0.9 0.4 0.5 0.6 0.7 0.8 0.9 Separation dc [ pm ] Separation dc [pm] Figure 7.8 Dependence of the quality factor, Q (a), and the power extraction efficiency (b) on the coupling distance, dc, between the disk and bus waveguides. The tested devices have radii of R=8 and 12pm. The power extraction efficiency is plotted in Figure 7.8(b). Disks with the same radius should have the same mode effective index, and the phase matching conditions should not change with dc, assuming that the bus waveguides have constant effective index during the different process runs (see next section for details). The variable parameter is the coupling coefficient, which changes the power balance of the cavity. Note that the disks with larger radius have larger power extraction efficiency because of the better phase synchronism. The data is rather scattered due to sensitivity of the bus waveguide effective index on the etch depth, but one can note the clear maximum in the power extraction efficiency at dc= 0 .8 pm. 132 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This is the point at which it is believed that a power balance in the coupled bus waveguides/cavity system is achieved. 7.6 Dependence on the waveguide etch depth dW G In this section we describe experiments in which the bus waveguide etch depth, dwG, is varied. In varying this parameter, we are in essence changing the effective index of the bus mode. This is shown in Figure 7.9, where the effective indices of the first bus TE and TM modes are plotted. The etch depth, dwc, is defined with respect to the end of the waveguide core layer. Thus, a negative dwG means that the waveguide is etched up to somewhere in the middle of the core layer, and a positive dwG means that the core layer is completely removed. In all of these experiments we keep the thickness of the top membrane the same, n= 0 .2 fim. X=1.55jxm; d =1pm; w=1pm; t=0.2gm; X 3.15 % 3.13- | 3.12- 600 i f B 0 .0 am - - U J 3.11- Substrate 3.09- m 3.08- -0.2 0.0 0.2 0.4 Waveguide etch depth dW G [p,m ] Figure 7.9 Calculation of the effective indices and losses for the first TE and TM modes of the bus waveguide as a function of the waveguide etch depth, dW G . 133 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The TM waveguide mode has a larger effective index and larger leakage loss into the substrate because it spreads more in the vertical direction than does the TE mode. By decreasing the etch depth, dwG, the effective index of the top cladding layer is effectively decreased (the thickness of the membrane is kept constant, t= 0 .2 |L im ), and thus the effective index of the bus mode is decreased. The height of the lower cladding layer dciad is fixed at Ifim in order to lessen the leakage loss of the mode into the substrate, which increases as the mode effective index becomes progressively smaller than the index of InP (see Figure 7.9). Figure 7.10 Bus waveguide mode distribution. The bus cladding is lpm thick and the top InP membrane is 0.2pm. d WG- +0.5pm (left), and d WG - -0.2pm(right). Figure 7.10 shows the field distribution of the first bus waveguide mode for the two boundary points in Figure 7.9, i.e. when dwa= +0.5pm(left) and dwG= - 0.2pm(right). When dwG is a positive number the mode is very well confined and the leakage loss is negligible. This is not the case when dwG is negative. Then the field is pushed down closer to the cladding layer and penetrates deeper into the substrate, which results in larger energy leakage (the trend plotted in Figure 7.9 for negative dwd)- The height of the lower cladding layer dciad is also important for designing low loss bus waveguides. Figure 7.11 shows a calculation of the bus mode effective 134 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. index and loss as a function of the waveguide cladding layer thickness. The leakage of the power into the substrate increases as the field penetrates deeper into it at small cladding thicknesses. For example, the bus waveguide mode distribution is plotted in Figure 7.12, where the thickness of the cladding, dciad, is only 0.2pm. X=1.55pm; w=1pm; t=0.2gm; dwa=0 2000 X -8 c 3= UJ 3 0 .4 u m Substrate n , i ! ---1 ---1 ---1 ---1 ---1 ---1 ---------- ? ---r i.O 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 Cladding layer dc |a d [pm] Figure 7.11 Calculation of the effective indices and losses for the first TE and TM modes of the bus waveguide as a function of the cladding layer thickness, dd^ HHPPiSllU ■ m m # * * 4Sf£ - ‘ f iJ' — Figure 7.12 Bus waveguide mode distribution. The bus cladding, ddaii, and the top InP membrane are 0.2pm thick. dW G = +0pm. The mode leaks into the substrate. The experimental results for microdisks with radius of i?=12pm and coupling separation of tlc=0.8pm are shown in Figure 7.13. There is a clear increase of the quality factor and the power extraction efficiency with the decrease of dwc- A 135 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. comparison between the effective index of the disk mode (Figure 7.6) and the effective index of the waveguide mode (Figure 7.9) suggests that phase matching causes the observed higher power extraction efficiency at negative dwG. It is noteworthy that the calculated effective index strongly depends on both the etch depth (dwG.) and the thickness of the remaining top InP layer (t), as both of them alter the effective index of the top cladding layer. Thus, nonuniformity in the etch depth resulting from the fabrication process is expected to cause a deviation of the mode effective index from the values shown in Figure 7.13, and the graph gives an estimate of the sensitivity to fabrication errors. x10 7- 6 - 5 5 - - R=12 pm ; TE polarization g < Q 3 o 4 -| 3 2 1 a) t A - d=0.8 am; ★ - d =0.7 am; 100 -90 S ' -80 0 1 C 1-70.2 4-4 O -60 2 ■ R UJ - 5 0 " 0) -40 | cl 1 -3 0 b) 20 -0.2 -0.1 0.0 0.1 0.2 0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 Waveguide Etch Depth dW G [ urn ] Figure 7.13 Dependence of the quality factor, Q (a), and the power extraction efficiency (b) on the waveguide etch depth, dy?c- Tested devices have 7?= 12pm and <4=0.7 and 0.8am. The dependence of the quality factor is plotted in Figure 7.13(a) for disks with R=\2\ixa and coupling separation t4=0.8fxm. One would expect a decrease of Q with the decrease of dwG, as the coupling coefficient effectively increases at phase 136 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. matching conditions. Exactly the opposite behavior is observed on the plot - Q increases at small dwG■ A high Q cavity is very sensitive to any kind of external perturbations and defects, and it can be used as a detector with amplification factor proportional to g 2 [10]. The cavity views the bus waveguides as localized defects, as mentioned above, not only because it loses energy due to the coupling, but it also loses energy due to the scattering of light from the interface region between the bus waveguides at the disk edge, and the air cladding at the disk center region. To estimate the amount of this scattering, which is a result of the modal mismatch between these two regions, we will calculate the difference, Sn, between the modal effective indices of the disk cavity above the bus waveguides, and above the central air region. Using a slab waveguide approximation we can estimate the effective index of the resonator with different thickness of the bottom cladding layer. The index difference &i is then found to be Sn=10~4 when dwa= -0.3pm, and the index difference increases by more than an order of magnitude, A=3xl0"3, when dwa= +0.3pm. The wave traveling close to the disk interface sees this index change four times per round trip, which results in scattering and a decrease in Q. Furthermore, when the waveguides are etched closer to the disk core layer, the field in the resonant cavity overlaps to a greater extent with the surface roughness on the bottom disk surface, created by the dry etching process, and thus experiences more loss and a lower Q value. It is interesting to note that the Q values for devices with coupling distance of Jc=0.7pm stay almost the same and are not dependent on dwc■ This shows that Q is 137 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. coupling limited, as discussed in the previous sections. The loss associated with the scattering from the waveguides is small (see the numbers above) and only high Q devices will be affected. 7 .7 Dependence on the thickness of the membrane t In this section we examine the effect of changing the thickness of the InP membrane. In Figure 7.14 a calculation of the effective index and the loss of the disk mode is shown as a function of t. From the graph it can be inferred that the mode effective index has a very weak dependence on t and the main factor affected is the loss. If the membrane thickness is more than 0.4|mm, the light from the disk resonant cavity couples into the slab waveguide formed by the membrane with air cladding layers. 1=12pm; w=1pm; TE; k=1,55pm; ■? 3.103 g 3.103 u3 3.103 4 0 TS O "g 3.103 O r> 3.102 3.102- 0.1 0.2 0.3 0.4 0.5 Thickness of InP membrane t [pm] Figure 7.14 Calculation of the disk effective index and the loss as a function of the thickness of the remaining InP membrane, t. An increase of the loss is observed at large t, due to the more effective coupling of energy from the disk into the slab waveguide formed by the membrane. 138 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A full vectorial calculation is plotted in Figure 7.15, which shows how the mode energy leaks into the membrane. At this thickness the effective index of the slab waveguide mode approaches the effective index of the resonant cavity and an effective coupling is possible. Figure 7.15 Coupling of light from the microdisk cavity to the slab waveguide formed from the remaining InP membrane. The thickness of the membrane is f=0.5pm. x10 4.5 4.0 3.5 0 3.0 T > £ 2 -5 £* 2 .0 1 1.5 a 1.0. 0.5-| 0.0 - dc=0.8 pm m - InP membrane thickness: • - 1=0.15 pm /• A, - 1=0.3 pm ^ " # ★ - 1=0.4 pm s y ■M s * * W y y s A''' 3 4 5 6 7 8 9 10 11 12 13 Disk radius R [pm] Figure 7.16 Quality factor as a function of the disk radius, R, for different thickness of the remaining InP membrane, t. A decrease of Q is observed at large values of t due to the leakage of energy from the cavity into th e membrane. The loss increases exponentially leading to lower Q values as shown in Figure 7.16. In this figure the experimental quality factor is plotted as a function of the 139 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. radius, and the thickness of the membrane is used as a parameter. The expected decrease of Q at thicker membranes is observed. 7.8 Summary High-Q, InP vertically coupled microdisk resonator/waveguide couplers were fabricated using a thermal wafer bonding technique to pattern both sides of an epitaxial structure containing two stacked waveguides. The devices exhibit single mode operation and large free spectral range. The vertical coupling geometry provides the freedom to design the disk and bus waveguides separately and to incorporate active regions into the disk. The quality factor of the resonators is shown to be coupling limited in the space of parameter values investigated, which is confirmed by the linear (or square root) dependence of Q on R at constant dc, and by the exponential dependence of Q on the separation distance dc at constant R. The phase matching of the disk and waveguide modes is found to affect the measured power extraction efficiency and by varying the waveguide etch depth, dwG, a better phase synchronism and better contrast ratios were achieved. High Q cavities are very sensitive to any external influences and even a small perturbation caused by the waveguide etch depth spoils the Q. Increasing the thickness of the InP membrane remaining after the resonator fabrication also decreases the quality factor by coupling energy from the disk cavity to the membrane slab waveguide. 140 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7.9 References 1 D.V. Tishinin, P.D. Dapkus, A.E. Bond, I. Kim, C.K. Lin, and J.O'Brien, “Vertical resonant couplers with precise coupling efficiency control fabricated by wafer bonding,” IEEE Phot. Technology Lett., vol. 11, pp. 1003 -1005, August 1999. 2 Kostadin Djordjev, Seung June Choi, Sang Jun Choi, and P. Daniel Dapkus; “High-Q, Vertically-Coupled Microresonators Built by Wafer-Bonding Technique”, 14th Annual LEOS’ 2001 Meeting Conference Proceedings, November, 2001., pp.509-510. 3 Kostadin Djordjev, Seung-June Choi, Sang-Jun Choi, P.D.Dapkus; “High-Q Vertically- Coupled InP Microdisk Resonators”, IEEE Photonic Technology Letters, vol. 14, no.3, March 2002, pp.331-333. 4 Djordjev, Kostadin; Choi, Seung-June; Choi, Sang June.; Dapkus, P.D; ‘Study of the Effects of the Geometry on the Performance of Vertically-Coupled InP Microdisk Resonators’, IEEE Journal of Lightwave Technology, vol. 14, no. 8 , August 2002 to be published 5 I. Ury, PhD Thesis, CalTech, 1980 6 B. E. Little and S.T.Chu, “Estimating surface-roughness loss and output coupling in microdisk resonators,” Optics Letters, vol. 21, No.17, pp. 1390-1392, September 1996 7 M. L. Gorodetsky, A. D. Pryamikov and V. S. Ilchenko, “Rayleigh Scattering in high-Q microspheres,” J. Opt. Soc. Am. B, vol. 17, No.6 , pp. 1051-1057, June 2000 8 B.E.Little, S.T.Chu, H.A.Haus, J.Foresi, and J.P.Lain, “Microring resonator channel dropping filters,”, J .Lightwave Technology, vol,15,pp.998-1005, 1997. 9 M. K. Chin and S. T. Ho, “Design and modeling of waveguide-coupled single-mode microring resonators,” Journal of Lightwave Technology, vol. 16, pp. 1433-1446, August 1998. 10 B. Little, Sai T. Chu, H. Haus, “Second-order filtering and sensing with partially coupled traveling waves in a single resonator”, Optics Letters, vol.23, pp.1570-1572, October 1998 141 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 8 Active Microdisk Devices 8.1 Devices with an Eiectroabsorptive Active Region 8.1.1 Theoretical The electroabsorption effect (EA) is widely used in practice for designing modulators [1 ,2 ] but if applied to a microdisk device, the absorptive mechanism can be used to suppress the resonant power transfer rather than to promote loss. Thus, the input signal suffers only small attenuation. The design methodology we have employed here minimizes the required applied voltage to obtain a certain contrast ratio, CR. This approach may result in some residual loss in the high transmission state that would be viewed as insertion loss. If the applied voltage is not a constraint, choosing another design can reduce this insertion loss. To investigate the influence of the loss on resonator performance, contours of equal power transmission T=\t\2 and quality factor Q derived from CMT are plotted at resonance for a R=5pim disk in Figure 8.1 against two independent variables. The first independent variable is the coupling coefficient, jc=jq=iq, which can be controlled by choosing the appropriate vertical separation between the waveguide and disk, and the second variable is the loss in the disk, a. The intersection values for the Q and T contours on the x-axis (when j c =0) are for a stand-alone disk, i.e. when the disk is not coupled to the bus waveguides. Increasing the losses inside the device leads to decrease of the quality factor and the 142 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Q is loss-limited. The values of Q and T on the y-axis are for an 'ideal' disk without internal losses a=0. Q is coupling-limited and the input energy is completely transferred (T=0) to the output waveguide independent of K (assuming symmetric coupling). Any practical configuration {tt£0, K^O) can be found somewhere in between. T ransm ission a n d Quality F actor a t R=5|x m 0.055. 0.05 0.045 * °'04 O ) C 0.035 Q. 3 O O 0.03 5 o CL 0.025 0.015 0.01 -oja- •0.9 0.005 40 35 Absorption a [cm 1] i Figure 8.1 Contours of equal power transmission T=\t\2 and quality factor Q derived with the CMT theory are plotted at resonance for R=5jum as a function of the power coupling coefficient and cavity loss. Note that the disk can tolerate moderately large losses and still provide useful performance as an add-drop filter. For example, to obtain power transmission of T=0.1 (90% of the input power is extracted from the bus waveguide) with loss coefficient a=5cm l, we can design the power coupling coefficient to be k=3.5% and the resulting Q=4000. To obtain the same transmission at higher losses, a=10cm'1, 143 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the coupling efficiency m eeds to be increased, k=6%, to compensate for the power lost. This, of course, will decrease the quality factor of the cavity to Q=2000. A larger disk is less sensitive to the coupling, meaning that at the same k and a the larger disk has larger quality factor but also larger transmission (we want small, possibly zero transmission at resonance). Thus by going to bigger disks we trade CR for quality factor (or bandwidth, AX). Figure 8.1 also shows the possibility of designing an active device by using absorption as a trimming mechanism. For acceptable, practical values of the contrast ratio of CR = lOlog^™/^" 10dB , and AT=0.7, this graph can be used for design considerations. Let's assume that the coupling coefficient is ic=2%, and we would like to vary the losses in order to achieve a change of transmission AT=0.7. For To n =0.1 the loss has to be ob=3cm1, and the resulting Q=6000, for To jf=0.8 we need a=50cni1 and the resulting Q=1200. Thus, to achieve the desired contrast ratio the electroabsorption parameter has to be m=A(rfab=16. In bulk materials, EA is achieved by the Franz-Keldysh (FK) effect and in 2D quantum well materials EA is achieved by the quantum confined stark effect (QCSE). The QCSE is more sensitive to the applied voltage and the absorption change is much larger than in the case of the FK effect. Thus, the QCSE is the preferred choice [2]. We follow the method outlined in [3], where the QW states under applied field are calculated via the transfer matrix approach for solving the effective mass equation, and the exciton discrete states are found by the variational 144 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. method. The complete absorption spectrum can be calculated as a sum of bound-state excitonic transitions and continuum-state quantum well contributions (see Appendix 5 for details). The purpose of this section is to investigate the applicability of the QCSE in active microdisk devices. It was shown in Figure 8.1 that a microdisk resonator is very sensitive to any kind of loss. Thus, the challenge is to optimize the active region in such a way as to meet all of the design targets. Positioning the mode frequency closer to the bandgap increases the absorption change, and thus, decreases the required switching voltage. The closer position also increases the background absorption at zero voltage and decreases the CR. Thus, the trade-off is between the drive voltage and the CR, and the challenge is to find the appropriate QW composition and size, which will give maximum transmission change and CR at Ao=l.55/un. To accomplish this task we developed a procedure for QW optimization, which is similar to the one outlined in [1] but differs in the fact that here the task is to find the optimal conditions for electroabsorption microdisk devices [4]. The results are shown in Figure 8.2. The procedure for the lnxGai^AsyPj-/InxGai.x AsyP]-y system is the following: • Assume barrier strain and wavelength, for example Sb=0% and /k=1.15/im. The composition parameters X b, yb are then found. 145 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 140- \ Absorption alfaon [cr Q6S Q67 068 Q6B 07 Q7I Q72 Q73 Q74 Q75 Q el^lrtegcletti Q67 068 Qffl 07 Q7t Q72 Q73 Q74 Q75 Q/Vcmpcsitiariywfarn (a) Absorption changa A a =2329.1CE [cm 0.67 0.68 0.69 0.7 0.71 0.72 0.73 0.74 0.75 Refractive index change A nffl8X = 0.00(1)18 0.67 0.68 0.89 0.7 0.71 0.72 0.73 0.74 0.75 QW composition yw from (c) 0.66 0.67 0.68 0.69 0.7 0.71 0.72 0.73 0.74 0.75 Absorption alfa0)) [cm'1] 0.67 0.68 0.69 0.7 0.71 0.72 0.73 0.74 0.75 QW composition yw from (b) TrBnsrrisicninCFFsMe 140 12) w100 Q03 Q67 Q6B Q69 Q7 Q71 Q72 Q73 Q74 Q75 O H ^inth8ti3HrisaaiATB[=QS54.1 140 ®1(D Q08 Q67 068 Q69 07 Q71 Q72 Q73 Q74 Q75 GWrxnpositicnywScm (d) Figure 8.2 Example QCSE optimization of the QW structure for an active microdisk. • Assume QW strain (s=0. 8 %), and calculate the QCSE at wavelength A=1.55iim for different As compositions yw=(0.66 to 0.77) and 146 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. different QW sizes Lw =(70 to 150 A). The applied field is E=65kV/cm, which corresponds to V=2.6V for an intrinsic layer thickness di=0.4^im. The bandgap energy and the overlap integral (@65kV/cm) between the CB and the first VB bound state are given in Figure 8.2(a). The overlap integral, which determines the oscillator strength of the absorption, depends mainly on the QW size and is weakly dependent on the composition. The smaller the QW size, the stronger the particle confinement. Under an applied electric field the e and hh wave functions shift towards the opposite ends of the QW. Thus, the stronger confinement leads to larger overlap integral, but the device is less sensitive to the applied voltage. • Calculate the absorption coefficient at zero bias and at E=65kV/cm, Figure 8.2(b). • Use the results from the previous step, and calculate the change in absorption A a (max. value is /da^a x=2895cnf1 ) and the change in refractive index An (max. change Anm aX =0.00056) at A=1.55fim due to the applied field, Figure 8.2(c). • Assuming an average optical confinement of r= 2% , 2QW's, Qe= 10,000, and using eqn.(A4.9) the calculated transmission in the OFF state, To ff, and AT are given in Figure 8.2(d). The best values are achieved at LW =100A and yw=0.6803, which corresponds to a bandgap energy of Eg=0.825 eV (Aw=1.5jim). This composition does not coincide 147 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. with the one where A a is maximum (1^=100, yw=0.705), because the higher background absorption decreases the achievable contrast ratio. Thus, the objective is to find a composition with a high absorption change, and at the same time with small background absorption. The transmission achieved in the 'off state is TO ff=10% and AT=70%, which corresponds to CR=10dB. From these results it can be concluded, that the design of an EA active microdisk will require precise tuning of all the parameters, disk dimensions and active region composition. The performance of single-disk EA devices is limited by the minimum achievable loss in the ON state, which determines the quality factor (the bandwidth) of the unbiased device, and the available dynamic range A a of the active EA region. A reasonably high on-resonance extinction could be achieved and the device could be useful for modulator and ON/OFF switch applications. Further improvement of the performance could be realized by designing higher order filters [5]. 8.1.2 Experimental In this part we demonstrate, experimentally, microdisk devices with electroabsorptive (EA) active regions. To our knowledge, these are the first active microdisk resonators utilizing the quantum confined Stark effect (QCSE) as a way of loss trimming the frequency response of the cavity [6 , 7]. These devices might be useful as active switching/routing elements in a photonic circuit or as small and fast resonant modulators. 148 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 8.1 Epi-structure of an active microdisk resonator utilizing the Quantum Confined Stark Effect Layer Description Material X /Strain d [A] Doping 17 N+ - cladding InP - 3,000 n+, l e l 8 16 n - cladding InP - 7,000 n, 5el7 15 Bus core InGaAsP 1 .1 / 0 % 4,000 n, 3el7 14 Coupling Layer InP - 8 , 0 0 0 n, 3el7 13 SCH InGaAsP 1.25/0% 1,700 1 2 4 x QWs InGaAsP 1.55/1% 80 1 1 3 x Barriers InGaAsP 1.25/0% 1 0 0 1 0 SCH InGaAsP 1.25/0% 1,700 9 Zn - setback InP - 1 , 0 0 0 i 8 p ~ - cladding InP - 1 , 0 0 0 p, 3el7 7 Low p - cladding InP - 1 , 0 0 0 p, 5el7 6 High p - cladding InP - 6 , 0 0 0 p, l e l 8 5 p+ contact InGaAs - 2 , 0 0 0 p+, le l9 4 Etch stop 2 InP - 1 , 0 0 0 i 3 Etch stop 1 InGaAs - 2 , 0 0 0 i 2 Buffer InP - 1 0 0 0 n , 3el8 1 Substrate InP - - n+ The vertically coupled microdisk switches to be discussed are fabricated using the epitaxial structure shown in Table 8.1. The growth starts with an InP buffer layer, two etch stop layers and disk p-cladding layers which have decreasing doping to reduce free carrier absorption in the disk resonator. The disk core, with total thickness of 0.4pm, consists of two separate confinement layers (SCH) with Xscn= 1.25pm and 4 quantum wells (QWs) with emission wavelength at /? w =1.55pm 149 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. and 3 barriers with ,3*= 1.25pm. Next, the n-doped separation/coupling layer was grown with a thickness of 0 .8 pm, which, as discussed in the previous chapter, leads to optimal device performance with our geometry. Finally, the bus waveguide core layer with AwG=l-lpm was grown followed by the top n-cladding InP layers. For details on layer thickness and doping see Table 8.1. R=10pm; TE; EA Modulation; 4 QWs C 1.0 a 0.6 ^ v= DV;— f-V=-1V j V = -2V ; i-V = -3 fkO O 0.1 1550 1560 1570 1580 1590 1600 1610 Wavelength [nm] Figure 8.3 Loss trimming of the transmission response of a microdisk/waveguide coupler with radius i?=10pm. By applying -3V reverse bias the losses in the cavity increase by Aa^ldcm'1 , which leads to change in the quality factor by AQ=3200 and in the transmission by AT— QA5. The device process is described in detail in Chapter 6 . Here we will only discuss the experimental results. Figure 8.3 shows the measured transmission properties of a microdisk with a radius of i?=10pm with TE input polarization. Sharp and deep peaks are observed at the resonant wavelength. The device has a free spectral range, FSR, of 10.5nm, quality factor, Q, of 5700, ftness, F, of 40, and transmission at resonance, T, of 0.1 around 1.584pm when there is no applied bias. Note that T 150 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. increases and Q decreases at shorter wavelengths, where the active region is absorbing and the internal loss is larger. Due to the QCSE, applying a reverse bias shifts the absorption edge towards longer wavelengths and introduces additional loss into the cavity. This leads to an increase of the transmission and decrease of the Q. The measured values of T and Q, and the calculated values of the absorption coefficient O f are shown in Table 8.2 for different bias at /k= 1.584p.m. From the insert in Figure 8.3 it can be seen that due to the electrorefraction effect, the resonant frequency is red shifted by 0 .2 nm, corresponding to a modal refractive index change of about zk=4xl0'4. Table 8.2 Transmission at resonance T, quality factor Q, and the calculated modal absorption coefficient a, as a function of the applied reverse bias at k=1584nm. The coupling coefficient, k , is assumed to be 4.5%. Applied Bias [V] T Q a [cm'1 ] 0 0 .1 5,700 3 -1 0.17 5,000 5 - 2 0.35 3,500 1 0 -3 0.55 2,500 19 An EA microdisk switch/modulator is a four-port device and we can change both its transmission and dropped coefficients by introducing additional loss into the cavity at a particular wavelength. It is interesting to compare the expected performance with the measured results from this device. Figure 8.4 is a calculation of the transmission and dropped coefficients as a function of the cavity loss coefficient a.The measured values of T at d= 1.584pm are also shown on the same plot with triangles. The coupling coefficient is assumed to be 4.5%, which is estimated to be the actual value for our devices. At a=0, i.e. the case of an ‘ideal’ resonator, the 151 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. input power is completely transferred to the dropped port. The increase of the cavity loss increases the transmission, T, and decreases the dropped coefficient, D. Note that the rate of change of D, dD/da, is much larger than the rate of change in T, dT/da, because part of the dropped power is absorbed, leading to a smaller D for a given a. Thus, one can obtain a larger contrast ratio, CR, from the dropped port at the expense of larger insertion loss. For the particular wavelength of interest (/L=l.584pm), the cavity absorption coefficient is Cfc=3cm ' 1 at zero bias, which will result in 7=0.1 and £>=0.5 (additional IL of 3dB in the dropped port). R=10pm; EA Trimming; Coupling k =4.5% Transmitted Power - Dropped Power O 0.8 Q 0.7 C 0.6 m 0.3 (0 0.2 20 25 30 Modal Loss a [cm'1 ] Figure 8.4 Dependence of the transmission and dropped coefficients on the loss in the microdisk cavity. The measured values of T at different bias are also plotted. If we assume for the sake of argument that one can change the modal absorption to oc= 30cm'1, then the attainable contrast ratio at the transmitted port CRt is 8.7dB and at the dropped port C Rd is 11.3dB. Therefore, the dropped port will exhibit 152 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2.6dB larger CR than the transmitted port, the difference coming from the additional power absorbed in the cavity. Note that, in contrast to the conventional waveguide EA modulators/switches, where the light is completely absorbed into the active region, a microdisk device absorbs only a small part of the power, and the change in the output intensity is achieved mainly by rerouting of the light from one port to the other. Thus, heating and absorption-saturation effects can be avoided. A small initial cavity loss, < % , will be more favorable for ‘D modulation’, rather than T modulation’, because the insertion loss decreases and the rate of change of D increases rapidly. R=10pm; TE; EA Modulation; 4 QWs; C o "5 m amm E m e is ■ o 0 N 7 5 E o z k=1552.4 nm 562.6 nm A,=1573.0 nm k=1584.2 nm X=1595.3 nm V 'X j .0 -3.5 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 Applied Voltage [V] Figure 8.5 Transmission coefficient as a function of the applied bias measured at different resonant wavelengths. The maximum change is achieved at A^29nm from the bandgap. In general, a larger absorption change, Aa, and small initial absorption, O o, are needed for larger contrast ratios. At wavelengths closer to the bandgap, the 153 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. absorption change A a is larger, but so is O q. At longer wavelengths both of them decrease. Therefore, to find the optimal resonant wavelength, which will give the maximum ratio Aa/ao, and thus the largest CR, the transmission is measured and plotted as a function of the reverse bias (Figure 8.5) for all resonant peaks of the same device. The normalization is performed with respect to the transmission at resonance with zero bias. At a wavelength close to the bandgap, .4=1552.4nm, the maximum AT is small. With an increase of X, the CR increases and reaches its maximum at /l=1584.2nm, i.e. A4=29nm from the bandgap. Further increase of the wavelength leads to decrease of AT. R=12pm; TE; E lectroabsorption 4QW s; Q ~5,000; 1.0 0) 0.8 © 0.4 L L k V=-1V V= -2V V= -3V 1550 1 5 6 0 1570 1580 1590 1600 W avelength [nm] Figure 8.6 Loss trimming of the dropped response of a microdisk with radius f?= 12p.m. Larger contrast ratios C7?>10dB could be achieved at the expense of larger insertion loss. Figure 8 . 6 is an example of larger contrast ratio, achieved by loss trimming of the dropped characteristic of different device with radius i?=12pm. In the insert to the same graph Ci?= lOdB is measured at V= -3V form the resonance around .4=1.55pm. 154 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 8.1.3 Summary InP vertically coupled microdisk resonator/waveguide switches with active region utilizing the QCSE are demonstrated for the first time. The devices exhibit single mode operation, large free spectral range (i75i?=10.5nm) and a high Q of 5700. The QCSE provides a way of loss-trimming of the resonant characteristics: by increasing the reverse bias, an increase in the transmission and decrease in the dropped power is observed at the resonant wavelengths. These devices are viewed as building blocks for future photonic integrated circuits. Miniature active switches, routers and fast modulators amenable to large-scale integration are envisioned as part of a WDM system. Their small size, the use of polyimide to reduce the parasitic capacitance, and the QCSE itself, assure a fast frequency response. The resonant nature of these devices guarantees their excellent sensitivity. Further improvements in the design, such as, an increase in the number of the quantum wells and improving the overlap between the electric and optical fields are necessary for achieving larger contrast ratios and better sensitivity 8.2 Devices with a Gain Active Region 8.2.1 Theoretical Waveguide circuits inevitably introduce losses that limit the scale of the system or that require the use of optical amplification for compensation. In this section we consider the effects and implications of introducing gain as a mechanism to vary the transfer function of the resonant coupler. Since electrical excitation of the resonant 155 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. medium changes both the loss and the index of refraction, it is expected and found that the functionality of the device is altered over the electroabsorption case studied in the previous section, where the dominant effect is variation of the loss. T ransm ission a n d Q uality F actor a t R=5jd m 0.16 0.14 0 .1 2 * o > c cl = 3 O o 0.1 ( D 5 o Q _ 0.08 0.06 0.04 0.02 • 1 .4 ' 0.8- .•1'5 -1 0 10 20 25 30 -5 0 5 Modal Gain g [cnrT1] Figure 8.7 Contours of equal transmission T and Q for a 5|im microdisk device with gain active region (g=-a) (the changes in the refractive index are not included). To separately demonstrate the effects resulting from the introduction of gain into the resonator, we have plotted in Figure 8.7 contours of equal transmission and Q using eqn.(A4.9) for a gain active region (g = -a ), where we have not included the inevitable changes in the refractive index. The contours are plotted assuming a rather high value of residual loss in the resonator of a=15cm I. We have chosen a value that is at the limit of usable device parameters to show that even a lossy cavity can provide a useful performance with small driving powers when gain is used as the active region modulation mechanism [4], 156 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In this case, the gain compensates for losses and the Q increases with gain, which is accompanied by a decrease of the transmission at resonance (in the presence of gain the definition of Q is questionable and in this case the linewidth is a better quantity for consideration). The cavity views the coupling to the waveguides as additional loss, and thus, with an increase of K , for given gain, Q decreases. Above a certain gain value, the stimulated emission from the cavity adds to the mode energy, and thus the transmission begins to increase rapidly. The clear maximum in the Q (Q goes to infinity) and in the transmission coefficient along the line connecting the coordinates (g=15, K >=0) and (g=40, j c =0.08) expresses the threshold condition: gain=loss. At that point the device is a laser. The data on the right of this line is invalid, because that line marks the condition for which the gain is pinned to its threshold value. Of course, in practice, a pure gain medium is impossible as there is inevitably a concomitant change in the index of refraction. Furthermore, the devices we model are high Q resonant cavities and any, even a small, change in n leads to detuning of the resonant frequency and decrease of the transmission at a given wavelength. Thus, the inclusion of An by a more realistic active region in the calculations is an important issue. The gain spectrum and the total recombination rate for an InxGaj. x A syP j./ InxGai-x AsyPi.y active region is calculated, and the results are shown in Figure 8 . 8 for the case of: Ao=1.55jum', gain peak positioned at l= 1.6jjm ; and quantum well compressive strain s=0.8% [3]. The index change is calculated using Kramers-Kronig (KK) relationships. 157 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. r * 0.01 , Nqw= 1 , Rload = 50n , R1 = 4.95fi m , R2= 16.95g m , R 3- 26.95g i 0.02 0.00 3000- -0.02 2000- 1000- -0.040 -a o 6 - 1000- -0.08 -2000- _ 10 -10 0.2 0.4 0.6 0.8 t 1.4 1.6 1.8 0 1.2 Figure I 2 4 6 8 10 Gamier density x101 8 [am2 ] Applied V oltage [V] (a) (b) i Material gain and index change as a function of the carrier density (a), and the calculated modal gain as a function of the applied voltage. Transmission and Quality Factor at R =4.95 p m 0.16 0.14 0 .1 2 C D o' O ' 0.08 O 0.06 0.04 0.02 0 . 8- .1.2- 30 35 40 -10 -5 0 5 10 15 20 25 4 Modal Gain a Icm 1 (c) Figure 8.9 Contours of equal transmission T and Q for a 5pm microdisk device with gain active region (g=-a). (the changes in the refractive index are included). The calculation of T and Q for a fixed input wavelength are shown in Figure 8.9. The maximum in the transmission around the coordinate (g=32, k=0.05) is due to the inclusion of the refractive index change, An, which shifts the resonance, and thus 158 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. affects T. In this case the coupling coefficient has to be properly tuned in order to achieve the desired transmission change. For the application of this device as a modulator or switch, AT has to be at least 70% to achieve a CR>10dB. Applying a coupling of k=8% and having an initial modal gain of g i-1 5 c m l and final modal gain of g2=27cmI we can meet this target. If we assume only one well, confinement factor of r-1 % , and load resistance of Ri=50£2, calculated drive currents result in voltage changes of less then 0.2V to achieve the required Ag, as shown in Figure 8 .8 (b). These considerations are ‘area’ based considerations, and thus by varying the disk and electrode sizes different voltage-to-gain characteristics can be achieved. The bandwidth of the device will be limited by the minority carrier lifetime. A microdisk device with a gain active region could be useful as a laser, ON/OFF switch, modulator and add/drop filter. The gain compensates for the loss and a high extinction ratios and drop efficiency could be obtained with low power consumption at the expense of a speed limited by the carrier lifetime. The device can be operated in two regimes. The first is in the limit of a small coupling coefficient and small index change, where increasing the gain compensates for the loss, increases the loaded Q and decreases the transmission. The second regime is when the device is used with a bias point set to achieve zero transmission and increase of the gain leads to increase of T. Either mode is potentially useful and requires very low switching power. Each also permits the bias tuning of the operating point necessary to compensate for fabrication or temperature variations. 159 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 8.2.2 Experimental In this section we experimentally demonstrate microdisk switches with gain active region [6 , 8 ]. To our knowledge, these are the first active microdisk resonators utilizing a gain active medium as a way of trimming the frequency response of the cavity. These devices might be useful as active switching/routing elements in a photonic circuit or as filters with tunable bandwidth and gain. The epitaxial structure is the same as one described in the previous section, but now we apply a forward bias to the same type of devices as in the previous section. Figure 8.10 Gain trimming of the transmission response of a microdisk/waveguide coupler with radius By applying a dc current of 5 mA, the modal loss in the cavity decreases by A a = \lc m l, which leads to increase of the quality factor by zl <2=5300 and decrease of the transmission by AT=0.5. Figure 8.10 shows the measured transmission properties of a microdisk with a radius of f?=10pm with TE input polarization. The device has a free spectral range FSR of lOnm, quality factor Q of 5700, fmess F of 40, and transmission at resonance R=10 pm; TE; Gain Modulation; 4 QWs; 1550 1560 1570 1580 1590 Wavelength [nm] 160 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. T of 0.2 around /l=l.594pm without bias. The coupling coefficient is estimated to be k=5.6% and the internal loss at this wavelength, which is below the bandgap, is i Oo=5cm . This number includes the material loss and the loss due to the scattering from sidewall roughness. Note that T increases and Q decreases at shorter wavelengths, where the active region is absorbing without bias and the material loss is larger. Applying a dc current to the active region alters the transmission characteristics of the resonator. First, far from the bandgap wavelength (/lv v =1.55pm), due to the change of the refractive index caused by the injected free carriers, the resonance is simply shifted without significant decrease of Q. For 7= 1mA drive current, the resonant wavelength is blue shifted by A Jh= Inm, and the estimated modal refractive index change is An=2xl0'3. At larger currents, the cavity heats up, and the increased temperature red-shifts the resonant wavelength - see for example the graphs with 1=5 and 10mA. There is a balance between both processes, which restricts the maximum obtainable wavelength shift. Second, the behavior is completely different for wavelengths inside the bandgap of the active region. As shown in the insert to the same graph, by increasing the drive current, a decrease in the loss inside the cavity is observed, which leads to higher Q and lower transmission. More interesting is the case when 7=10mA. There the transmission not only increases, but also is even larger than unity. To investigate this behavior and to understand the influence of the gain medium on the resonator performance, the power transmission coefficient T and the dropped 161 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. coefficient D (derived with the coupling of modes in time theory approach) are plotted at resonance for R=10iim disk in Figure 8.11, as a function of the modal gain in the cavity, g. For this calculation we assumed that O o =5cm"1 and k=3.6%, which were shown above to be the actual values for our resonator. The modal gain is varied from -15cm'1 (20cm' 1 total cavity loss coefficient) to -t-lOcm'1 (5cm 1 total cavity gain) and ^O cm ' 1 means that the active material is transparent and only the residual absoiption is present, ob= 5cm'1 . R=10pm; Gain Trimming; k=3.6% ; a0 =5cm'1 j___ i ___ i ___ i ___ i ___ i ___ i ___ i ___ i — Transmission Coefficient Dropped Coefficient ▼ Measured Values u-u-1 — i ----- 1 ------- 1 ------ ' ------ 1 ------ 1 ------ 1 ------ *t----- 1 ------r -15 -10 -5 0 5 10 Modal Gain [cm'1 ] Figure 8.11 A calculation of the transmission and dropped coefficients as a function of the cavity modal gain/loss. The microdisk have radius R-lOgm, coupling coefficient k= 3.6% and internal loss coefficient ot=5cm "1 around A=1552nm. The measured values of the transmission are also plotted with triangles. At zero bias, the measured quantities are Qi— 2500 and 77=0.6 at l=1552.97nm (see the inset to Figure 8.10). From Figure 8.11 it could be inferred that the total loss in the cavity is «/=20cm'1 (15cm 1 material and 5cm' 1 residual absorption). By increasing the drive current to / 2=lm A and I3=5mA, the total loss in the cavity is 162 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. decreased to £ 6 = 12cm' 1 and arj=3cm' 1 respectively, and the resultant quality factor and transmission of the device are Q2=3200, 7 2 =0 . 4 4 and <2j=7800, T?=0.15 respectively. These conditions are also indicated in Figure 8.11 with triangles. Further increase in the gain will lead to decrease of the transmitted power and increase of the dropped power. When the modal gain is equal to the internal losses (^ Scm '1 ), the cavity is ‘ideal’ and the transmission 7=0, i.e. there is a complete power transfer to the dropped port, D= 1. If the gain is increased even further, but at the same time its value is not enough to overcome the loss associated with the power coupled to the bus waveguides (otherwise the microdisk will be a laser), then the transmission starts to increase again. The light coupled into the cavity will be amplified, the spontaneous emission will add to the mode output, and thus the total transmission and dropped coefficients will increase and become even larger than unity. This is the case measured with pump current 7= 10mA, shown in Figure 8.10 to lead to large T. The values of the quality factor and the transmission cannot be determined exactly from the graph, because of the relatively large wavelength step of our tunable laser. Note that Figure 8.11 does not include the change in the refractive index, An linked to Ag by the Kramers-Kronig relationships. This An blue-shifts the resonance as seen by the experimental graph, and if the switched wavelength is fixed, then it will experience a different transmission dependence from the one plotted. Switching can easily be achieved by using both ports and could be utilized with or without an initial bias. 163 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R=10|im; TE; Gain/Temperature Tuning; 4 QWs m n o m 03 2 0.7 £ 0.6 ® 0.4 - I= 0 m A; T=23 C l= 1 mA; T=33 C l= 5 m A; T=39 C 1551 1552 1 5 5 3 15 5 4 Wavelength [nm] Figure 8.12 Microdisk filter with tunable bandwidth. The refractive index change due to the temperature tuning has the opposite sign and the same magnitude compared to the refractive index change due to the injected carriers, and thus the resonant wavelength stays constant. Temperature tuning could also be incorporated into this device, since increasing the temperature will red-shift the resonant wavelength. Combined with current tuning, a filter with adjustable bandwidth could be demonstrated (Figure 8.12), where by simultaneously varying the temperature and the drive current the quality factor is changed, while the resonant wavelength is kept constant. Figure 8.13 shows the measured dropped coefficient of the same microdisk device, the one with transmission properties plotted in Figure 8.10. First far from the bandgap wavelength the resonant peak is simply shifted, while maintaining relatively constant intensity. At energies above the bandgap, the intensity of the dropped power is substantially increased - we observe an amplification factor of five with respect to the peaks at longer wavelengths. This fact is in agreement with the calculated behavior of the dropped coefficient, Figure 8.11. Note that when applying gain to the 164 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. cavity, the transmission coefficient initially decreases, and then increases after the point where the gain compensates for the cavity loss. In contrast, the dropped coefficient increases monotonically with the gain. R=10gm; TE; Gain Modulation; 4 QWs _J ■ I I ■ I i I i L. 1=0 mA 1= 1 mA l= 5 mA 1=10 mA 1550 1560 1570 1580 15 9 0 16 0 0 1610 Wavelength [nm] Figure 8.13 Gain trimming of the dropped response of a microdisk/waveguide coupler with radius i?=10p,m. The dropped power is amplified above the bandgap energies and the dropped coefficient is larger that unity. 8.2.3 Summary InP vertically coupled microdisk resonator/waveguide switches with gain active regions were demonstrated. These devices are small, versatile and can be used as building elements for dense photonic integrated circuits. Active switches, routers, tunable filters and filters with tunable bandwidth could be built as part of a WDM system. The devices have high Q (5700) of operation, which shows that the free carrier absorption associated with the doping introduced into the cavity was minimized with the doping profile used. Gain provides a way to trim the resonant 165 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. characteristics. Switching could be easily achieved on both ports and an initial bias could be applied for better contrast ratios. The resonant nature of these devices guarantees their excellent sensitivity. Drive current change as low as zU=200pA corresponding to voltage change riV=0.1V was measured to be enough to completely shift the resonant wavelength, i.e. to turn a microdisk switch ON or OFF. With larger currents, in the order of a few miliamperes, the quality factor of the cavity could be changed by an order of magnitude. Much smaller drive currents and voltages could be expected with further optimization of the cavity, decreasing the cavity dimensions, improving the current uniformity and ensuring better overlap between the optical field and the injected carriers. 8.3 Devices with Free Carrier Injection Active Region 8.3.1 Theoretical In cases where gain is not desired or required, processes that merely tune the resonant frequency offer an interesting counterpart to the loss modulation mechanism considered previously. In this section we investigate the behavior of a region into which free carriers (FC) are injected [4]. To maximize the effects of the free carrier density on the effective index of the mode in the resonator we consider a thick, highly confining waveguide with a bandgap energy higher than the operating photon energy. This region is an intrinsic, high refractive index layer, sandwiched between n- and p-doped low-index layers. Free carriers are injected into the /-region by applying forward voltage to the p-i-n junction. We restrict the injection level to 166 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. levels low enough to avoid population inversion and gain in the region. The first task is to find the carrier density in the steady state condition from the rate equation without stimulated recombination: = — - - BN2 — CN3 = 0. Here we assume dt qdt rji=0.6, B=10'10[cm3/s] is the spontaneous emission rate coefficient, C=3.10~ 2P [cm6/s] is the Auger recombination rate coefficient, J [A/cm2 ] is the current density and di-0.45fim is the intrinsic layer thickness. Solving for N [cm'3] we obtain carrier densities up to 4 x l0 17[cm'3] for current densities up to 200A/cm2. The next task is to obtain the refractive index change due to the free carriers injected into the intrinsic region. For this hypothetical device we choose to estimate the free carrier effects using an empirical relationship that captures the magnitude of the effect in the absence of a specific design. The empirical formula used is: An = d% N AN = - r y O - 20 A A . Here is the confinement factor of the mode, which is approximately 50% and N is the carrier density in [cm'3]. For the current density change mentioned above, index changes are on the order of An=0.002. The injected carriers absorb light and the absorption coefficient, in this case, is also found by using an empirical relation: a = r^(3.10~1 8 n + 7.1(T18p) = r t ),l(r 17AAf. Here a is in [cm'1 ] and N is in [cm'3]. This relationship was empirically determined for GaAs at energies below the band gap. The magnitude is not expected to be strongly dependent on the materials choice for energies below the bandgap. The 167 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. change in the absorption found for the carrier densities of interest is Aoc^lcm1, which does not strongly affect the Q. T ran sm issio n a n d Q uality F acto r a t R = 5 n m - 1000- -1000- 0.16 0.14 0.12 -1500-j O ) -1500- Q. (5 0.08 — 2000- -2000- -2000- 0 .9 * ^ 0.06 ^3000- ~30£ -3000- 0.04 -4000- -4000- -4000- 0.02 -6000- -6000- -6000- 160 180 100 120 140 40 60 80 0 20 C u rren t d e n sity J[A /cm 2] Figure 8.14 Contours of equal power transmission T=\t\2 and quality factor Q derived with the CMT theory are plotted at resonance for R~5fim as a function of the power coupling coefficient and the injected current density. The performance of an active microdisk with this active region is calculated by assuming a background absorption of Ob=5cm~1, which is a typical value measured in semiconductor laser diodes and includes the material absorption and the absorption from the p- and n- doped regions. Figure 8.14 shows a calculation of the transmission change and the total quality factor of the cavity. The change of the refractive index affects the cavity performance more than the FCA, and switching is achieved, for example, with AJ=80A/cm2 at k=4%. At these conditions Qtot remains almost constant Q-3500 and AT-0.8, which gives CR=20dB. This result shows the attractive features of a FC injection region as an 'active' material. Also another 168 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. attractive feature is that the device could be fabricated in a variety of direct bandgap materials without the need to match the energy gap to the input wavelengths. FC injection could be a useful approach for building active semiconductor microdisk devices. The index change has much larger influence then the absorption change and the device could be tuned without substantially degrading the quality factor and the filter bandwidth. This may find application in photonic circuits, where an active tuning of the resonant frequency is required in order to compensate for fabrication imperfections, aging processes and change in the temperature. Also this shift of the resonant frequency could be used in modulator, ON/OFF switch and add/drop filter applications, where large contrast ratios could be achieved with small power consumption. 8.3.2 Experimental In this section, we present tunable vertically coupled microdisk devices using a p-i-n active region, where the resonant wavelengths are tuned by the free carrier plasma effect. Free carriers injected into the intrinsic region change the effective refractive index of the whispering gallery modes, and thus blue-shift the resonant wavelengths. To our knowledge, these are the first tunable semiconductor microresonators utilizing a FCI active region to attain successful tunable characteristics [9]. 169 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The tunable microdisk filters demonstrated in this paper basically have a p-i-n diode structure incorporated into the disk cavity. The epitaxial structure is shown in Table 8.3. Table 8.3 Epi-structure of a microdisk resonator with FCI active region. .Laye r Description Material d[A] Doping 14 n+ - cladding InP 3,000 n+, l e i 8 13 n - cladding InP 7,000 n, 5el7 1 2 Bus core Q (1.1pm) 4,000 n, 3el7 1 1 Coupling Layer InP 8 , 0 0 0 n, 3el7 1 0 Disk core Q (1.25pm) 4,000 i 9 Zn - setback InP 1 , 0 0 0 i 8 p' - cladding InP 1 , 0 0 0 p, 3el7 7 Low p - cladding InP 1 , 0 0 0 p, 5el7 6 High p - cladding InP 6 , 0 0 0 p, l e l 8 5 p+ contact InGaAs 2 , 0 0 0 p+, le l9 4 Etch stop 2 InP 1 , 0 0 0 i 3 Etch stop 1 InGaAs 2 , 0 0 0 i 2 Buffer InP 1 0 0 0 n, 3el8 1 Substrate InP - n-i- By applying forward bias, carriers are injected into the /-region, and due to the free carriers plasma effect, a large absorption change, A a, appears in the far infrared spectral region. The Kramers-Kronig relationships link this change in the absorption to change in the refractive index, An, which shifts the resonant wavelength towards shorter wavelengths. The maximum value of this blue shift is limited by the cavity 170 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. heating caused by the injected current (the resonant wavelength is anticipated to increase with increase in the temperature). Therefore, for large tuning ranges, proper heat sinking is necessary. R=10pm; TE; FCI tuning; 1.1 m 0.9 r o.8- £ 0.7- S . 0.6 - ■O 0.5- < D tS 0.4- £ 0 .3 - c 0.2 .2 0.1-1 • l=0mA; 1 = 1 mA; l=5mA iill t * M h p M il l| I I * | Vi P ' ! i| 1 s ! i ] ■ 1570 1610 1580 1590 1600 Wavelength [nm] Figure 8.15 Microdisk tunable filter. The free carriers injected into the microdisk cavity change the modal refractive index, which blue-shifts the resonant frequency. Figure 8.15 shows the measured transmission characteristic of a microdisk tunable filter and switch with radius J?=1Q pm with TE input polarization. The device has a free spectral range FSR= 10 nm, quality factor <2=5500, finess F=40, and transmission at resonance T=0.1 around /l=1.587 pm. Applying a dc current to the active region alters the transmission characteristics of the resonator. Owing to the change of the refractive index caused by the injected free carriers, the resonance is simply shifted without significant decrease in Q. For 7=lmA drive current, the resonant wavelength is blue shifted by zU =l.lnm , and the estimated refractive index change is An= -2.1X10’3. The linewidth of this device, 54=0.28nm (34GHz), is 171 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. narrow enough to extract a specific WDM channel. By shifting the resonance with less than lnm (124GHz), the filter could be tuned to an adjacent channel (spaced at 100GHz), and thus individual channel selection is possible. For /= 5mA, one would expect five times larger index change due to the linear relationship between I and n, i.e. An= -10x10' . From the same figure, the resonant wavelength shifts only by A4=2nm, and the estimated effective refractive index changes by An= -4x1 O '3. The discrepancy between these numbers suggests that the increased temperature accounts for An= +6x10' change in the modal index. R=10jim; TE; FCI tunning; O 9.0 E 2.0 / 1.0 0.8 0.6 > 0) O ) 0.4 (0 o > 0.2 0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Current [mA] Figure 8.16 Switching behavior of a microdisk tunable filter at /t=1598nm. Change in the drive current by zt/=200pA is enough to toggle the switch from OFF to ON state. The voltage change required for this transition is riV=0.1V. This device may be used as an ON/OFF switch or active router. For example, if the input wavelength is tuned to the resonance of the switch, >4=1598 nm, then the transmission is T0 =0.12 without bias. By applying enough current to shift the 172 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. resonant wavelength by an amount larger than the full width half maximum (FWHM) of the peak, the transmission will increase to a value Tj=0.9. The switching behavior is shown in Figure 8.16. It can be seen that by increasing of the drive current from I0 =200 pA to Ij=400 pA the switch toggles from OFF to ON state. The voltage change required for this transition is zSV=0.1V. The measured contrast between OFF and ON states is less than lOdB, but this can be improved by designing optimized microdisk cavities with smaller transmission at resonance, or by the use of multiple disks. The sensitivity of the current dependence and possibly the range of tuning can be enhanced by minimizing the total current. This can be done by limiting the current flow to the edges of the disk where the index change is desired and by passivating the edges to reduce surface recombination there. R=10fim; TE; FCI tuning; 1545 1550 1555 15 6 0 15 6 5 1 5 7 0 1575 Wavelength [nm] Figure 8.17 Dropped port tuning of microdisk with radius i?=10pm. The device has a high quality factor Q of 7000. 173 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The dropped characteristics of a different device with the same radius are shown in Figure 8.17. The device has a larger quality factor Q~7,000 and is more sensitive to the loss inside the cavity. The figure shows a slight dependence of Q on I, and the quality factor drops to 5000 when the drive current is 7=5mA. This decrease in Q suggests an additional absorption coefficient of aj =3.5 cm' 1 due to the free carrier absorption process. R=10 (im; QW device R=10 ixm; p-i-n device „ 2.2 E 2.0 f . 1 .8 - i r H 1.2 jj 1-01 f> 0 .8 | 0.6 0.4 i - A — - 1=1 mA - l=5mA FCI + Band Filling FCI only — A — =1 mA E 2.0 f . 1.8 =5mA W 1.4 FCI only 1545 1560 1575 1590 1605 1620 W avelength [ nm ] 1545 1560 1575 1590 1605 1620 W avelength [ nm 1 (a) (b) Figure 8.18 Typical tuning characteristics of microdisk devices with gain (a) and FCI (b) active regions. The graphs show the achievable wavelength shift as a function of the wavelength (measured at the resonance), and the drive current is used as parameter. It is interesting to compare the amount of wavelength shift for given bias as a function of the resonant wavelength for two different devices utilizing gain and FCI active regions respectively. Both devices are driven by current and thus there is a change in the modal index and a shift in the resonant wavelength due to the injected carriers. However, the bandgap wavelength of the gain region is l=1.55|im . At these wavelengths, band filling and bandgap shrinkage effects are present, which contribute an additional index change [10], The measured tuning characteristics of 174 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. microdisk devices with gain (a) and FCI (b) active regions are shown in Figure 8.18, where the drive current is used as parameter. The wavelength shift of the FCI devices is independent of wavelength, which is to be expected from the Drude model, predicting a constant refractive index change. The devices with gain active regions show somewhat different behavior. The wavelength shift at shorter wavelengths (when approaching the bandgap) is larger, due to the contribution from the band filling effect. Band filling is expected to lead to negative index changes [10], and thus this contribution has the same sign as the FCI effect. An, due to band filling, gradually decreases at energies below the bandgap. A similar trend is observed in Figure 8.18(a), where the wavelength shift is large at shorter wavelengths and gradually decreases and levels out at longer wavelengths, were the dominant contribution is only from FC injection. 8.3.3 Summary Tunable microdisk resonant filters/switches with a free carrier injection active region were demonstrated for the first time. The devices exhibit single mode operation, large free spectral range (FSR= lOnm) and a high Q of 7000. The cavity has the design of a simple p-i-n junction. Free carriers injected into the intrinsic region change the effective index of the whispering gallery mode, and therefore blue- shift the resonant wavelength at a rate of 1 nm/mA. These devices may be useful elements in DWDM systems, where the active tuning is a way to compensate for the fabrication imperfections and to adjust the 175 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. resonant wavelength to the desired value. In this work we also demonstrated an active switch or router, which can toggle from the OFF to ON state by a change in the drive current of A/=200pA. The voltage change required for this transition is AV=OAV. 8.4 Devices with Electrooptic Active Region 8.4.1 Theoretical InP has a Zink-Blende-Structure and thus is 43m crystal. It is optically isotropic without applied field and the index ellipsoid in the presence of an electric field is 2 2 2 x y z given by: — + — + — + 2 rn Exyz + 2 rA 1 E xz + 2 rn Ezxy = 1. no no no A good reference for the EO effect is [11]. We will look for two possible configurations, which could be applicable for building u-disk modulators. The first example is sketched in the left column of Table 8.4. The electric field is applied in the z direction and the top surface is the (001) plane. This is the usual substrate orientation for growing high quality material. Unfortunately, as we can see, the refractive index change in the x'direction is positive under an applied field when the refractive index change in the y ' direction is negative. Thus the accumulated An per roundtrip for a microdisk (when the light propagates in a circle) is zero and this orientation is not applicable for building active devices. However, this is not the case for the example shown in the second column in Table 8.4. Here the index change in the x'and y'directions is the same so the total 176 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. accumulated An is not zero. Unfortunately the top surface is the (111) plane, which is difficult for both growing and processing. There are few examples in the literature of successful growth of high quality materials on this surface. The index change is Table 8.4 Electrooptic effect and its application to microdisk modulators. Electric Field E l (001) Ex =Ey=0; EZ =E; E 1 ( 1 1 1 ) Ex=Ey=EZ =E/V 3; nx' n / nz' n0 + y 2 fi3 0r4}E n0 - Y 2 nlr4 X E «o n< ‘ + / 2 S nlr" E n^ } i S nl^ E no ~ ~ Y Y n°r41^ Confi guration i 1 z (0 0 1 ) A Z ' (111) IM I ' i 2 .---- ... * C x' X If we assume a coupling of ic=2% (Qc=20,000), symmetric coupling between the disk and both bus waveguides, loss a=2.5ctri1, this gives us Qtot=6000 and T=0.1 at the resonance wavelength of Ao=1.55/jm for the case of a 5|im disk. The refractive index change, which is needed to shift the resonant wavelength (without changing the loss) and produce AT=0.7 (CR=10dB) at the same wavelength, is An=3.9xl0~4. For InGaAsP r,«=2xlO'10cm/V, and using the formula above we obtain E=230kV/cm. If we assume the thickness of the intrinsic layer is di=0.4/Jm, then the applied 177 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. voltage, assuming uniform distribution, is V=E*di=9.2V. The very weak electrooptic effect in InP results in the required voltage being high and therefore does not meet our objectives for low-voltage modulators. To enhance the EO effect a series of cascade resonators could be used for real devices. 8.5 References 1 M.K.Chin, W.S.C.Chang., “Theoretical Design Optimization of Multiple-Quantum-Well Electroabsorption Waveguide Modulators”, IEEE J. Quantum Electronics, vol.29, no.9, pp.2476-2488, 1993 2 M.K.Chin, “Comparative analysis of the performance limits of Franz-Keldysh effect and quantum-confined Stark effect electroabsorption waveguide modulators”, IEE Proc. Optoelectron., vol. 142, no.2, pp. 109-114, 1995 3 S.L.Chuang, “Physics o f optoelectronic devices ”, J. Wiley & Sons, Inc., 1995 4 Kostadin Djordjev; Seung June Choi, Sang Jun Choi, and P. Daniel Dapkus, “Active Semiconductor Microdisk Devices”, IEEE Journal o f Lightwave Technology, vol.20, no.l, January 2002, pp.105-113. 5 B.E. Little, H.A. Haus, J.S. Foresi, L.C. Kimerling, E.P.Ippen and D.J. Ripin, “Wavelength Switching and Routing Using Absorption and Resonance”, IEEE Phot. Technology Lett, vol. 10, no. 6 , pp. 816-818, 1998 6 Kostadin Djordjev, Seung-June Choi, Sang-Jun Choi, P.D.Dapkus, “Active Semiconductor Microdisk Switching Devices Utilizing Gain and Electroabsorption Effects”, Optical Fiber Communication Conference and Exhibit (OFC), March 2002, Postdeadline session pp. FA2-1/FA2-3 7 Kostadin Djordjev, Seung-June Choi, Sang June Choi, P.D Dapkus, “Vertically Coupled InP Microdisk Switching Devices with Electroabsorptive Active Regions”, IEEE Photonics Technology Letters, vol. 14, no.8 , August 2002, pp. 1115-1117 8 Kostadin Djordjev, Seung-June Choi, Sang June Choi, P.D. Dapkus, “Gain Trimming of the Resonant Characteristics in Vertically Coupled InP Microdisk Switches”, Applied Physics Letters, vol.80, no.19, 13 May 2002, pp.3467-3469 178 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 9 Kostadin Djordjev, Seung-June Choi, Sang June Choi, P.D Dapkus, “Microdisk Tunable Resonant Filters and Switches”, IEEE Photonics Technology Letters, vol. 14, no. 6 , June 2002, pp.828-830 10 B.R. Bennet, R.A. Soref, J.A.D. Alamo, “Carrier-Induced Change in Refractive Index of InP, GaAs, and InGaAsP”, IEEE J. Quantum Electronics, vol.26, no.l, January 1990 11 A.Yariv, P.Yeh, “Optical Waves in Crystals ” , J. Wiley & sons, Inc., 1984 179 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 9 Summary 9.1 Semiconductor optical equalizer Two-segment semiconductor optical amplifiers/equalizers (SOA-EQ) with current adjustable gain spectra were demonstrated for the first time. They were built as a component of a particular system, an optical analog-to-digital converter, but have much broader applications. The equalizers have spatially and spectrally inhomogeneous gain medium. It is spatially inhomogeneous, because both sections are separated in space and an independent control of their biasing currents is possible. It is spectrally inhomogeneous, because both sections have different spectral properties, and thus by varying the relative strength of the bias currents, the overall device spectrum (which is a convolution of the spectra of both sections) could be equalized. Selective area growth techniques are used as a main enabler for the monolithic integration. First, the devices were theoretically analyzed to find the optimal regime of operation. We solved for the coupled set of multimode equations for the carrier density and the photon flux in the separate modes, in both pulsed and CW regimes. It was shown that self-phase modulation (SPM) is a dominant source of spectral broadening. SPM appears as a result of gain saturation that is responsible for time- dependent variations in the carrier density, and hence the refractive index. The temporal and spectral changes occurring during amplification depend on the relative values of the pulse width and the carrier lifetime. When the pulse width is much 180 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. shorter than the carrier lifetime, both the shape and the spectrum of the amplified pulses are asymmetric. For input pulses longer than the carrier lifetime the spectrum is broadened on both the red and blue sides. SPM induces frequency chirp, i.e. the local frequency of the pulse changes with time. This self-induced chirp has a large impact on the output spectrum, when the input pulse is unchirped. With an increasing of the chirp of the input pulse the effect of this nonlinearity weakens and in the limit of a large input chirp the effect is insignificant. After studying and understanding the general properties of a TWSOA we underlined the following design ‘portrait’ of an SOA-EQ: A multi-section device with the longest-wavelength section having the largest and broadest gain than all other sections. The spectrum of each consecutive shorter-wavelength section has to overlap and be included into the spectrum of each preceding, longer-wavelength section. Independent biasing of all sections is needed. CW equalization could be achieved at all input levels and without using gain saturation, by adjusting the relative strengths of the biasing of the separate sections. To achieve equalization in the pulse regime, the pulse has to have large linear chirp (i.e. the frequency linearly increases with time) and duration much larger than the carrier relaxation time (quasi-CW regime). In this case the equalizer has to have large total gain to assure gain saturation at low input levels. If saturation is always achieved the device is insensitive to variation in the input signal. The wafers were grown by low-pressure MOCVD on n+ [001] InP substrates. The selective area growth (SAG) techniques were utilized to simultaneously define regions with different spectral properties with one growth run. This technique requires patterning of the epitaxial wafers with a dielectric mask prior to the growth process, and then performing the growth with the mask on the wafer surface. SiNx was used for this purpose and the SAG regions were defined within a 15/im opening area between dielectric stripes with width w, positioned at every 250/im. The 181 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. calibration runs showed that varying the width w from 0 to 50/im changed the bandgap wavelength by more than 160nm, compared to the wavelength in the area without a dielectric mask. Broad area lasers were fabricated to confirm this wavelength shift and the devices showed uniform threshold current distribution (150A/cm2 /well) up to 30/mi SAG stripe width, where the wavelength was red shifted by 90nm with respect to the wavelength in the plain area. Lasers processed between wider stripes showed somewhat higher threshold currents, which is explained by increased defect density in the QW structure due to the higher compressive strain. Experiments to find the best wafer cleaning procedures were also performed. It was found that the CF4 etching of the dielectric mask has a detrimental effect on the device performance, and therefore BOE was implemented instead, followed by a long O2 plasma ashing to remove photoresist residues. To fabricate the semiconductor optical equalizers we employed broad area, ridge waveguide and buried-heterostructure structures. The fabrication process for all of the approaches were developed and applied to real devices. For defining BA and BH mesas we used a Cl2 - based ECR process to improve etching depth uniformity and reproducibility. RWG mesas were etched with HBr - based solution to define a reverse mesa profile for improved thermal and electrical characteristics. Lasers processed with different approaches showed reasonable performance, with the BH lasers somewhat dependent on the used etching chemistry and achieved uniformity. A polyimide process was developed and used for planarization and defining the bond-pads. 182 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. TWSOAs were fabricated by applying a two-layer AR coating on both facets of the lasers to reduce the feedback. CW gain equalization was obtained over a lOOnm range by varying the relative magnitudes of the drive currents in both sections. The devices show 12dB of gain and lOdBm saturation power. These equalizers may play an important role in optical analog-to-digital converting systems. Another potential application for this spectrally inhomogeneous active region is as a gain media in wide-range tunable DFB laser applications. 9.2 Microdisk resonant devices High-Q, InP vertically-coupled microdisk couplers were fabricated using a thermal wafer-to-wafer bonding technique to pattern both sides of an epitaxial structure containing two stacked waveguides. The devices exhibit single mode operation and large free spectral range. The vertical geometry has the advantage of precise control of the coupling coefficient by the epitaxial growth. Another advantage is that waveguides and resonator can be grown with different material compositions, facilitating the design of active microdisk devices - ON/OFF switches, modulators and microdisk lasers. An innovative feature in our design is the circular post defined below the resonant cavity. Basically this is a second disk etched between the I/O waveguides. Its purpose is threefold: (i) it improves the mechanical stability of the whole 3D structure; (ii) it improves current/field uniformity in active devices; (iii) it acts as a limiter for the higher order modes improving the transmission characteristics. 183 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Extensive numerical simulations were undertaken to design and optimize the coupler geometry. It was found that low-effective-index, submicron, single-mode rib waveguides are needed for effective phase matching to the lower-effective-index whispering gallery modes (WGMs) of the microdisk cavity. Phase synchronism is critical for achieving high-extinction-ratio resonant microresonators. The coupling coefficient was calculated using the coupling of modes theory for curved transmission lines, where the propagation constant of the W G M s,/^ j, is a function of the interaction coordinate z, depending on the device geometry. The performance of the coupled disk-waveguides system was studied with the coupling of modes in time theory, and the effect of incorporation of EA, gain and FCI active regions on the transmission of the coupler were investigated. It was found that EA can change the transmission by introducing additional loss into the cavity; gain can compensate for the residual loss, forming an ideal, lossless cavity; the FCI effect simply shifts the resonant wavelength without significant effect on the quality factor. The resonant couplers were fabricated using a wafer-to-wafer bonding technique to pattern both sides of an epitaxial structure. The samples were first grown by low- pressure MOCVD on n+ [001] InP substrate. QWs were incorporated into the resonant cavity to utilize EA or gain active regions, and a bulk intrinsic material was used as an active region to utilize the FCI effect. The doping profile was optimized for low optical loss and low electrical resistance. In order to achieve smooth sidewalls during the etching, the photolithography and the dielectric masking were optimized separately, by design of experiment studies. A new two-step REE, CH4- 184 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. based dry etch process was developed and optimized, which resulted in vertical and smooth sidewalls by carefully controlling the C H 4 / H 2 ratio and the chamber pressure. In order to design high-Q, high-contrast-ratio resonators, different phenomena influencing their performance have to be understood. Therefore, passive microdisk devices were initially fabricated with the aim to optimize the coupler geometry and find the optimal values of the coupling coefficient, assuring critically coupled conditions. In this study we investigated how different structural dimensions affect the performance of the microresonators. We varied the coupling separation dc, the disk radius R, the etch depth of the bus waveguides dwc, and the thickness of the thin InP membrane between the waveguides and resonators that remains after fabrication, t. The experimental results were compared with numerical simulations obtained from a full-vectorial mode solver. The quality factor of the resonators was shown to be coupling limited in the space of parameter values investigated, which was confirmed by the linear dependence of Q on R at constant dc, and by the exponential dependence of Q on the separation distance dc at constant R. The phase matching of the disk and waveguide modes was found to affect the measured power extraction efficiency and by varying the waveguide etch depth, dwc, a better phase synchronism and better contrast ratios were achieved. High-(2 cavities are very sensitive to any external influences and even a small perturbation caused by the waveguide etch depth spoils the Q. Increasing the thickness of the InP membrane remaining after the resonator fabrication also decreased the quality factor by coupling energy from the disk cavity to the membrane slab waveguide. 185 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Electroabsorptive, InP, vertically-coupled microdisk switches with active region utilizing the QCSE were demonstrated for the first time. The devices exhibited single mode operation, large free spectral range (F5, i?=10.5nm) and a high Q of 5700. The QCSE provided a way of loss-trimming of the resonant characteristics: by increasing the reverse bias, an increase in the transmission and decrease in the dropped power was observed at the resonant wavelengths. These devices are viewed as building blocks for future photonic integrated circuits. Miniature EA active switches, routers and fast modulators amenable to large-scale integration are envisioned as part of a WDM system. Their small size, the use of polyimide to reduce the parasitic capacitance, and the QCSE itself, assure a fast frequency response. The resonant nature of these devices guarantees their excellent sensitivity. Further improvements in the design, such as, an increase in the number of the quantum wells and improving the overlap between the electric and optical fields are necessary for achieving larger contrast ratios and better sensitivity Gain-trimmed, InP, vertically-coupled microdisk switches were also demonstrated for the first time. ‘Ideal’ active switches, routers, tunable filters and filters with tunable bandwidth could be built as part of a WDM system. The devices had high-Q (5700) operation, which implies that the free carrier absorption associated with the doping introduced into the cavity was minimized with the doping profile used. Gain provided a way to trim the resonant characteristics. Switching could be easily achieved on both ports and an initial bias could be applied for better contrast ratios. The resonant nature of these devices guarantees their excellent 186 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. sensitivity. Drive current change as low as zt/=200pA corresponding to voltage change zlV=0.1V was measured to be enough to completely shift the resonant wavelength, i.e. to turn a microdisk switch ON or OFF. With larger currents, in the order of a few miliamperes, the quality factor of the cavity could be changed by an order of magnitude. Much smaller drive currents and voltages could be expected with further optimization of the cavity, decreasing the cavity dimensions, improving the current uniformity and ensuring better overlap between the optical field and the injected carriers. Tunable microdisk resonant filters/switches with a free carrier injection active region were also demonstrated for the first time. The devices exhibited single mode operation, large free spectral range (FSi?= lOnm) and a high Q of 7000. The cavity had the design of a simple p-i-n junction. Free carriers injected into the intrinsic region changed the effective index of the whispering gallery modes, and therefore blue-shifted the resonant wavelength at a rate of 1 nm/mA. These devices might be useful elements in DWDM systems, where the active tuning is a way to compensate for the fabrication imperfections and to adjust the resonant wavelength to the desired value. In this work we also demonstrated an active switch or router, which can toggle from the OFF to ON state by a change in the drive current of A/=200pA. The voltage change required for this transition is zlV=0.1V. The amount of wavelength shift, for given bias, was found to be independent on the wavelength, while operating at energy below the bandgap of the core material. The wavelength shift at shorter wavelengths (when approaching the bandgap) is larger, due to the contribution from 187 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the band-filling effect. Band filling is expected to lead to negative index changes, and thus this contribution has the same sign as the FCI effect. 9.3 Future Research Directions Microresonators are very attractive components for photonic circuitry due to their compactness and versatility. In this dissertation we have already demonstrated their functionality; different active regions were utilized to trim and adjust the resonant properties of the coupled microdisk/waveguides system. In order to integrate the individual components into a functional and commercially viable photonic chip, the following challenges have to be solved. First, the device structure has to be finely tuned, so that the individual resonators show better extinction ratio. This could be accomplished by carefully adjusting the coupling efficiency between the resonator and the waveguides and by optimizing the etching process of the bus, so the phase-matching conditions are maintained at all times. A large contrast ratio is necessary for improving the crosstalk between the separate communication channels. For larger tuning range it is imperative to have a microdisk cavity with good heat sinking. The current device structure with air-guided bus waveguides suffer from the poor thermal conductance of the air and the non-uniform current/field distribution close to the edge of the cavity. Thus a new design has to be developed. For example, a device with buried bus waveguides and a small air-guided microdisk is expected to have excellent thermal properties and uniform potential/carrier distribution. 188 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Moreover, a BH waveguide will have excellent coupling properties to a single-mode fiber, thus reducing the overall insertion loss. This structure has to be studied and an approach to phase-match the low-effective-index microdisk whispering gallery mode to high-effective-index buried waveguides has to be developed. Third, there are several novel devices that could be demonstrated using the same platform, which was developed with this dissertation. Some examples include: microdisk lasers, tunable lasers with microdisks used as feedback elements, resonant detectors, tunable filters, fast modulators, multiplexers, second- and higher-order filters with improved square-like spectral response, etc. The latter could be tunable filters, where one can tune the shape of the spectral response and at the same time tune the filter central frequency. All of these devices impose different challenges to the designer. For example the design of a microdisk modulator will require an optimization of the active region to improve the modulation speed. If an electroabsorption region is the preferred choice, then a maximization of Aa/Oo is necessary. If a FCI region is employed, then a method to reduce the carrier lifetime (which is the main restriction on device speed) has to be developed; a proton- implanted active region may offer a significant increase in device speed. And finally, the integration of hundreds of devices into a single photonic chip has to be demonstrated. The accomplishment of this milestone requires an unprecedented control on the uniformity and reproducibility. The resonant frequency of the microdisk is very sensitive to the variation in the cavity dimension, and thus lithography with high-fidelity image transfer is a necessity. 189 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Bibliography P.PAbsil, J.V.Hryniewicz, B.E.Little, F.G.Johnson, KJ.Ritter, and P.T.Ho, “Vertically coupled microring resonators using polymer wafer bonding,” IEEE Phot. Techn. Letters, vol. 13, no. 1, pp. 49-51, January 2001. P.P.Absil, J.V.Hryniewicz, B.EJLittle, R.A.Wilson, L.G.Joneckis, and P.T.Ho, “Compact Microring Notch Filters”, IEEE Phot. Technol. Lett., vol. 12, no.4,pp.398-400, 2000 G. P. Agrawal and N. K. Dutta, Long-Wavelength Semiconductor Lasers. New York: Van Nostrand Reinhold. 1986, ch. 2 G.P.Agrawal, N.A.Olsson, “Self-Phase Modulation and Spectral Broadening of Optical Pulses in Semiconductor Laser Amplifiers”, IEEE JQE, .vol.25, pp.2297-2306, 1989 Masahiro Aoki, Masaaki Komori, Tomonobu Tsuchiya, Hiroshi Sato, Kouji Nakahara, and Kazuhisa Uomi, “InP-Based Reversed-Mesa Ridge-Waveguide Stmctures for High- Performance Long-Wavelength Laser Diodes”, J. Selected Topics in Quant. Electronics, vol. 3, no.2, April 1997, pp.672-683 M.Aoki, H.Sano, M.Suzuki, M.Takahashi, K.Uomi, and A.Takai, Electron. Lett, vol.27, 1991, pp.2138-2139 R. Bellman. G. Bimbaum. and W. G. Wagner, ‘Transmission of monochromatic radiation in a two-level system,” J. Appl. Phys. vol.34, pp. 780-782, 1963 B.R. Bennet, R.A. Soref, J.A.D. Alamo, “Carrier-Induced Change in Refractive Index of InP, GaAs, and InGaAsP”, IEEE J. Quantum Electronics, vol.26, no.l, January 1990 Chin and S. T. Ho, “Design and modeling of waveguide-coupled single-mode microring resonators,” Journal of Lightwave Technology, vol. 16, pp. 1433-1446, August 1998. Seung June Choi, Kostadin Djordjev, Sang Jun Choi, and P. Daniel Dapkus, “CH4 -Based Dry Etching of High Q InP Microdisks’, J. Vac. Sci. Technol. B, vol.20, no.l, pp.301-305, Jan/Feb 2002. Seung June Choi, Kostadin Djordjev, Sang Jun Choi, and P. Daniel Dapkus, 'Smooth, CH4 - Based Dry Etching of InP Microdisks', SCCVAS Leading Edge Student Symposium, Anaheim (CA), October, 2001 Seung June Choi, Kostadin Djordjev, Sang Jun Choi and P. Daniel Dapkus, ‘Fabrication of Vertically Coupled InP Microdisk Resonators by using Smooth, CH4-based Reactive Ion Etching Methods’, 14th Annual LEOS’ 2001 Meeting Conference Proceedings, November, 2001., pp.628-629 190 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. M.K.Chin, W.S.C.Chang., “Theoretical Design Optimization of Multiple-Quantum-Well Electroabsorption Waveguide Modulators”, IEEE J. Quantum Electronics, vol.29, no.9, pp.2476-2488,1993 M.K.Chin, S.T.Ho, “Design and Modeling of Waveguide-Coupled Single-Mode Microring Resonators”, J. Lightwave Technology , vol. 16, no.8 , pp. 1433-1446, 1998 M.K.Chin, “Comparative analysis of the performance limits of Franz-Keldysh effect and quantum-confined Stark effect electroabsorption waveguide modulators”, IEE Proc. Optoelectron., vol. 142, no.2, pp. 109-114, 1995 S.T.Chu, B.E.Little, W.Pan, T.Kaneko, S.Sato and Y.Kokubum, “An eight-channel add-drop filter using vertically coupled microring resonators over a cross grid”, IEEE Photon. Technol. Lett, vol. 11, pp.691-693, 1999 S.L.Chuang, “Physics of optoelectronic devices”, J .Wiley & sons, Inc., 1995 L. Coldren and S. Comizine, “ Diode lasers and Photonic Integrated circuits ” , Wiley Series in Microwave and Optical Engineering, 1995 F.Coppinger, A.S Bhushan, and B.Jalali, “Time Magnification of Electrical Signals Using Chirped Optical Pulses”, Electronics Letters, vol. 34, no. 4, pp. 399-400, February 1998 Djordjev, K.D.; In Kim; Dapkus, P.D., “Chirped Pulse Propagation in Saturated Traveling Wave Semiconductor Amplifiers, ” LEOS '99. IEEE Lasers and Electro-Optics Society 1999 12th, Annual Meeting, Volume: 2, 1999, pp: 786 -787 Kostadin Djordjev, Sang Jun Choi, Won-Jin Choi, Seung June Choi and P. Daniel Dapkus, “Spectral Equalizers Based on Saturable Semiconductor Optical Amplifiers,” 14th Annual LEO S’ 2001 Meeting Conference Proceedings, November, 2001, pp.103-104 Kostadin Djordjev, Sang-Jun Choi, Won-Jin Choi, Seung-June Choi, In Kim, and P.D.Dapkus, “Two-Segment Spectrally Inhomogeneous Traveling Wave Semiconductor Optical Amplifiers Applied to Spectral Equalization”, IEEE Phot. Technology Lett., vol. 14, no.5, May 2002, pp. 603-605. Kostadin Djordjev, Seung-June Choi, Sang-Jun Choi and P.D. Dapkus, “High-Q Vertically- Coupled InP Microdisk Resonators”, IEEE Phot. Technology Lett., vol. 14, no.3, pp.331-333, March 2002. Kostadin Djordjev, Seung June Choi, Sang Jun Choi, and P. Daniel Dapkus, ‘High-Q, Vertically-Coupled Microresonators Built by Wafer-Bonding Technique’, 14th Annual LEOS’ 2001 Meeting Conference Proceedings, November, 2001., pp.509-510 Djordjev, Kostadin; Choi, Seung-June; Choi, Sang June.; Dapkus, P.D; ‘Study of the Effects of the Geometry on the Performance of Vertically-Coupled InP Microdisk Resonators’, IEEE Journal of Lightwave Technology, vol. 14, no. 8 , August 2002 to be published 191 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Kostadin Djordjev; Seung June Choi, Sang Jun Choi, and P. Daniel Dapkus, “Active Semiconductor Microdisk Devices”, IEEE Journal of Lightwave Technology, vol.20, no.l, January 2002, pp. 105-113. Kostadin Djordjev, Seung-June Choi, Sang-Jun Choi, P.D.Dapkus, “Active Semiconductor Microdisk Switching Devices Utilizing Gain and Electroabsorption Effects”, Optical Fiber Communication Conference and Exhibit (OFC), March 2002, Postdeadline session pp. FA2- 1/FA2-3 Kostadin Djordjev, Seung-June Choi, Sang June Choi, P.D Dapkus, “Vertically Coupled InP Microdisk Switching Devices with Electroabsorptive Active Regions”, IEEE Photonics Technology Letters, vol. 14, no.8 , August 2002, pp. 1115-1117 Kostadin Djordjev, Seung-June Choi, Sang June Choi, P.D. Dapkus, “Gain Trimming of the Resonant Characteristics in Vertically Coupled InP Microdisk Switches”, Applied Physics Letters, vol.80, no.19, 13 May 2002, pp.3467-3469 Kostadin Djordjev, Seung-June Choi, Sang June Choi, P.D Dapkus, “Microdisk Tunable Resonant Filters and Switches”, IEEE Photonics Technology Letters, vol. 14, no. 6 , June 2002, pp.828-830 S.Donati, G.Guliani, “Noise in an Optical Amplifier: Formulation of a New Semiclassical Model”, IEEE J.Quantum Electronics, vol.33, No.9, pp.1481-1488, 1997 Fan, Riantm Verdiell and Dagenias: “Real-time insuty monitoring of Ar coating for SOA by ellipsomtry”, IEEE Phot. Technology Lett., vol.4, no9, September 1992 E. Fill and W.Schmid. “Amplification of short pulses in C02 laser amplifiers, ” Phys. Lett., vol.45A, pp. 145-146, 1973. L. M. Frantz and J. S. Nodvik, “Theory of pulse propagation in a laser amplifier,” JAppl. Phys., vol. 34. pp. 2346-2349, 1963. N. J. Frigo, “Ultrashort pulse propagation in saturable media: A simple physical model,” IEEE J. Quantum Electron., vol. QE-19, pp. 511-519, 1983. M. L. Gorodetsky, A. D. Pryamikov and V. S. Ilchenko, “Rayleigh Scattering in high-Q microspheres,” J. Opt. Soc. Am. B, vol. 17, No.6 , pp.1051-1057, June 2000 H.A. Haus, “Waves and Fields in Optoelectronics”, Prentice Hall, Englewood Cliffs, NJ, 1984. H.A.Haus, W.P.Huang, S.Kawakami, and N.A.Whitaker, “Coupled-Mode Theory of Optical Waveguides”, J. Lightwave Technology, vol. LT-5, no.l, pp. 16-23, 1987 C. H. Henry, “Theory of the linewidth of semiconductor lasers.” IEEE J. Quantum Electron., vol. QE-18. pp. 259-264. 1982 192 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. J.V. Hryniewicz, P.P. Absil, B.E. Little, R.A. Wilson, and P.T. Ho, “Higher Order Filter Response in Coupled Microring Resonators”, IEEE Phot. Technol. Lett., vol. 12, no.3, March 2000, pp.320-322. A. Icsevgi and W. E. Lamb, Jr., “Propagation of light pulses in a laser amplifier,” Phys. Rev., vol.185, pp. 517-545, 1969. T.Krauss and P.Layboum, “Monolithic integration of a semiconductor ring laser and a monitoring photodetector”, SPIE, vol.1583, pp.150-152, 1991. P.G.Kryukov .md V.S.Letokhov, “Propagation of a light pulse in a resonantly amplifying (absorbing) medium,” Sov. Phys. Usp., vol. 12, pp. 641-672. 1970. Lee, D.L., “ Electromagnetic Principles of Integrated Opics”, John Wiley ans Sons, New York, 1996 Chao-Kun Lin, “Wafer-Bonded Bottom-Emitting 850nm VCSELS for Short Distance Free- Space Optical Interconnects”, PhD. Dissertation, USC, December 1999 B.E.Little, S.T.Chu, H.A.Haus, J.Foresi, and J.P.Lain, “Microring resonator channel dropping filters,”, J.Lightwave Technology, vol. 15,pp.998-1005, 1997. B. E. Little and S.T.Chu, “Estimating surface-roughness loss and output coupling in microdisk resonators,” Optics Letters, vol. 21, No.17, pp. 1390-1392, September 1996 B. Little, Sai T. Chu, H. Haus, “Second-order filtering and sensing with partially coupled traveling waves in a single resonator”, Optics Letters, vol.23, pp.1570-1572, October 1998 B.Eiittle, H.A.Haus, J.SJForesi, L.C.Kimerling, E.P.Ippen, and D.J.Ripin, “Wavelength switching and routing using absorption and resonance”, IEEE Phot. Technol. Lett., vol. 10, pp.994-996, 1998 E.A.J. Marcatili, “Bends in optical dielectric guides,” Bell Syst. Tech. J., vol.48, pp.2103- 2132, 1969 Marcatili, E.AJ, and Hardy, A., “The azimuthal effective-index method”, IEEE J. Quantum Electron. QE-24, 1988, pp.766-774 M.Matsuhara, Aa.Watanabe, “Coupling of curved transmission lines, and application to optical directional couplers”, Journal of the Opt. Soc. Of America, vol.65, no.2., pp. 163- 168, 1975 M. Osinski and J. Buus, “ Linewidth broadening factor in semiconductor lasers-An Overview,” IEEE J. Quantum Electron., vol.QE-23, pp. 9-29, 1987 O. Painter, R.K.Lee, A.Yariv, A.Scherer, J.D.O’Brien, P.D.Dapkus, I.Kim, “Two Dimentional Photonic Crystal Defect Laser”, Science, 284, pp.1819-1821 (1999) 193 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. W.H.Press, B.P.Flannery, S.A.Teukolsky, W.T.Vetterling, “Numerical Recipes in C”, Cambridge University Press, 1988 D.R.Rowland, and J.D.Love, “Evanescent wave coupling of whispering gallery modes of a dielectric cylinder”, IEE Proc.-J, vol. 140, no3, pp.177-188, 1993 E. O. Schulz-Dubois, “Pulse sharpening and gain saturation in traveling-wave masers,” Bell Syst.Tech. I., vol.43., pp. 625-658, 1964 K.Shimoda, H.Takahashi, and C.H. Townes, “Fluctuation in amplification of quanta with application to maser amplifiers,” J. Phys.Soc.Jap, vol. 12, pp.686-700, 1957 L. Silvestre, A. Ougazzaden, D. Delpart, A. Ramadane, C Daguest, and G. Patriarche, “Study of the Growth Rate and Composition Variations in Metalorganic Vapor Phase Selective Area Epitaxy at Atmospheric Pressure and Application to the Growth of Strained Layer DBR lasers”, J. Crystal Growth, vol. 170, 1997, pp.639-644. T. Tanbun, Y.K. Chen, J.A. Grenko, E.K. Byrne, J.E. Johnson, R.A. Logan, A.Tate, A.M. Sergent, K.W. Wecht, P.P. Sciortine Jr., S.N.G. Chu, “Integrated DFB-DBR Laser Modulator Grown by Selective Area Metalorganic Vapor Phase Epitaxy Growth Technique”, J. Crystal Growth, vol. 145, 1994, pp. 902-906. E.J. Thrush, M.A. Gibbon, J.P. Stagg, C.G. Cureton, C.J. Jones, R.E. Mallard, A.G. Norman, and G.R. Booker, J. Crystal Growth, vol. 124, 1992, pp.249 D.V. Tishinin, P.D. Dapkus, A.E. Bond, I. Kim, C.K. Lin, and J.O'Brien, “Vertical resonant couplers with precise coupling efficiency control fabricated by wafer bonding,” IEEE Phot. Technology Lett., vol. 11, pp. 1003 -1005, August 1999. Y.Yamamoto, “Noise and Error Rate Performance of Semiconductor Laser Amplifiers in PCM-IM Optical Transmission Systems”, IEEE J.Quantum Electronics, vol. QE-16, No.10, pp. 1073-1081, 1980. I. Ury, PhD Thesis, CalTech, 1980 J. P. Wittke and P. J. W arter, “Pulse propagation in a laser amplifier,” J. Appl. Phys., vol.35, pp.460-461, 1964. A.Yariv, P.Yeh, “Optical Waves in Crystals ” , J.Wiley & sons, Inc., 1984 194 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Appendices Appendix 1 SOA/EQ device process flow-chart Table Al.l STEP 1: SAG stripe definition Process Description Process Place Parameters Process Purpose SiNx deposition PCVD chamber 275°C/10min/30W Deposit dielectric mask TCE+ACE +ME clean Squeeze bottle + beaker 30s + 2min + 2min Remove particles and organics Dry blow N2 gun Removes moisture Pre-Bake + cool down Hot plate 120C / 2 min + lmin Removes moisture AZ5214 PR apply Spinner 4000rpm / 30s Photolithography PR bake Hot plate 120C / lmin Solvent removal ‘SAG’ mask exposure in Oil direction Aligner 70 mJ/cm2 Pattern definition Develop - AZ 400K, 1:4 Beaker 40s/ gentle agitation Develops exposed areas DI water rinse + dry blow (N2 gun) Beaker + running H2 0 5min DI rinse Removes the PR residues 0 2 ashing RIE 50W /200mtorr/ 25s Removes the PR residues BOE (1:10) etch of the SiNx Beaker 3 min Defines the SAG pattern DI water rinse + dry blow (N2 gun) Beaker + running H2 0 lmin Cleans from BOE Flood exposure of the PR Aligner 300mJ/cm2 Removes the PR Develop AZ400K Beaker lmin Removes the PR DI water rinse + dry blow (N2 gun) Beaker + running H2 0 2min Removes the PR residues Acetone + ME clean Squeeze bottle + beaker 2min + 2min Remove PR residue DI water rinse + dry blow (N2 gun) Beaker + running H2 0 5min Cleaning the surface 0 2 plasma cleaning RIE 100W/200mtorr/5min Cleans the surface and forms thin oxide r z n . ; r~ ~ : i n r u i i 195 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table A 1.2 STEP 2: SAG pre-growth treatment + growth Process Description Process Place Parameters Process Purpose Ultrasonic clean in DI bath Ultrasonic bath lmin Cleans the particles Sample cleaning - H2 SO4:H2O2:H20 Beaker (1:1:3); 65C / 20s Removes the oxide and cleans the surface DI water rinse DI bath- 18MOh 2 hours Remove the S 0 4 +ions from the surface Spin Dry Spinner lmin Removes the moisture uniformly MOCVD growth 1st step MOCVD reactor Growth of the active region BOE etch of the SiNx SAG stripes Beaker 10 min Etches the SiNx SAG pattern DI water rinse DI bath- 18MOh 2 hours Cleaning the surface Spin Dry spinner lmin Removes the moisture from the surface MOCVD growth 2n d step MOCVD reactor Growth of the top cladding layers SAG - SOA resion plain region SAG - SOA region 196 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table A1.3 STEP 3: Isolation trench formation. Process Description Process Place Parameters Process Purpose TCE + ACE + ME clean Squeeze bottle + beaker 30s + 2min +2min Remove particles and organics Pre-Bake + cool down Hot plate 120C / 2 min + lmin Removes the moisture AZ5214 PR apply Spinner 4000 rpm / 30s Trench formation PR bake Hot plate 120C / lmin Solvent removal ‘Trench’ exposure, the pattern is 10x30pm in size Aligner 70 mJ/cm2 Trench definition Develop AZ 400K 1:4 Beaker 20-40s Develop the -exposed pattern DI water rinse + dry blow (N2 gun) Beaker + running H2 0 2min Cleans the developer 0 2 ashing RIE 50W /200mtorr/ 20s Removes the PR residues Wet etch in H2 S04/H2 0 2 /H2 0 (1/1/3) Beaker RT/lOs; observe undercut with microscope Complete removal of the top InGaAs contact layer DI water rinse + dry blow (N2 gun) Beaker + running H2 0 lmin Cleaning from the acid Wet etch in HC1/H2 P 0 4 (1/6) Beaker + stirrer plate RT/stirring 10-40s Defining the trench depth ~ 0.4pm deep DI water rinse + dry blow (N2 gun) Beaker + running H20 lmin Cleaning from the acid TCE + ACE + ME clean Squeeze bottle + beaker 30s + 2min +2min Removes the PR 0 2 ashing RIE 50W /200mtorr/ lmin Removes the PR residues 197 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table A1.4 STEP 4: Mesa stripe formation Process Description Process Place Parameters Process Purpose NH4OH/H2 0 - (1/6) Beaker 30s Removes the surface oxide DI water rinse + dry blow Beaker + H2 0 2mm Cleans the surface SiNx deposition PCVD 275C /2500A /30W Dielectric mask TCE + ACE + ME Squeeze bottle + beaker 30s + 2min + 2min Remove particles and organics Dry blow N2 gun Pre-Bake + cool down Hot plate 120C / 2 min + lmin Removes the moisture HMDS apply Spinner 4000rpm/30s Improves PR adhesion AZ5214 photoresist (PR) apply Spinner 4000rpm / 30s Photolithography PR bake Hot plate 120C / lmin Solvent removal ‘Mesa’ mask align/exposure, 10pm wide Aligner 70 mJ/cm2 Pattern definition Develop - AZ 400K 1:4 Beaker 20-40s/ gentle agitation Develops exposed areas DI water rinse + dry blow (N2 gun) Beaker + running H2 0 5min Removes the PR residues 0 2 ashing RIE 50W/200mtorr/25s Removes the PR residues PR hard bake Hot plate 120C/ lmin Hardens the PR and smoothes the edges CF4 plasma etch RIE 120mW/1 OOmtorr/10 O s SiN* stripe formation ACE+ME Squeeze bottle + beaker 2min + 2min Remove PR 0 2 plasma cleaning RIE 100W /200mtorr/ 2min Cleans the surface Mesa etch ECR, 3pm deep ECR chamber Prp =50W BCl3 /A r, 10/3, 150C, 2mtorr, Pu w=1000W, Dry etch of the mesa 0 2 plasma cleaning RIE 100W /200mtorr/ 2min Cleans the surface BOE (1:10) etch of the SiNj Beaker 5 min Cleans the mask DI water rinse + dry blow (N2 gun) Beaker + running H2 0 5min Cleans the surface SiNx deposition PCVD 275C /1000A /30W Dielectric mask for the planarization P SiN s z . substrate 198 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table A1.5 STEP 5: Polyimide planarization Process Description Process Place Parameters Process Purpose TCE + ACE + ME clean Squeeze bottle + beaker 30s + 2min +2min Remove particles and organics Dry blow N2 gun Removes the moisture DI water rinse + dry blow (N2 gun) Beaker + running H2 0 2min Cleans the surface Sample pre-bake + cool-down Hot plate 120C / lOmin + 4min Moisture removal Polyimide PI-2737 spin Spinner 150rpm for 10s; 1800rpm for 60s Polyimide application Polyimide soft-cure + cool down Oven 60C/2min + 90C / 3 min + 2min cooling down Remove solvents ‘Polyimide’ mask exposure, 7pm wide Aligner 150 mJ/cm2 Alignment of the opening (negative image) Polyimide develop - DE-9040 Beaker RT; 7-10s Developing of the opening Polyimide rinse-RI 9180 + dry blow Beaker + N2 gun 1 min Removes the PY residue Polyimide cure Oven with N2 atmosphere. from 130C @ 5C/min to 350C, 30min cure @ 350C, cool down Hard cure of polyimide Observe the opening with microscope. Microscope Observe if the mesa top is clean 0 2 plasma etch-back of the PY (repeat this step until top is clean) RIE 0 2 /200mtorr/ 100W/30s Removing the PY residue on top of the device mesa substrate 199 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table A1.6 STEP 6: P-contact definition Process Description Process Place Parameters Process Purpose TCE + Acetone clean Squeeze bottle + beaker 30s + 2min Remove particles and PR Methanol clean Beaker 2min + agitation Remove acetone residue Dry blow N2 gun Removes moisture Pre-Bake + cool down Hot plate 120C12 min + lmin Removes moisture AZ 5214 PR apply - three layers Spinner 3 x 30Q0rpm / 30s Photolithography PR pre-bake Hot plate 120C /30s Solvent removal ‘P-contact’ mask exposure Aligner 70 mJ/cm"1 Pattern definition PR post-bake Hot plate 120C/ lmin Polymer cross-link Flood exposure Aligner 250 mJ/cm^ Negative image Develop - AZ 400K 1:4 Beaker 10 -30s/ gentle agitation Develops un-exposed areas DI water rinse + dry blow (N2 gun) Beaker + running H2 0 2min Removes the PR residues 0 2 ashing RIE SOW /200mtorr/ 30s Removes the PR residues CF4 plasma etching RIE 100 W /1 OOmtorr/ lmin Cleans the SiNx over the device mesa Metal Evaporation Metal Evaporator Ti/Pt/Au - 300A/500A/2000 A p-metal formation Metal lift-off - Acetone Beaker Until lifts-off Lift-offs the unnecessary metal Methanol clean Beaker 2min + agitation Remove acetone residue tpad substrate 200 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table A1.7 STEP 7: Substrate lapping and N-metals definition Process Description Process Place Parameters Process Purpose Substrate mounting on metal holder Lapping station Use mounting wax Mounting for lapping Clean all of the remaining wax around the sample Lapping station Acetone/TCE using a swab Cleans the wax for lapping Lapping using 5pm Alx Ov powder + DI Lapping station Up to 4 mils - 100pm Thins the substrate DI rinse the sample on the holder Beaker + DI rinsing 5 min Removes particles Substrate unmount Hot plate Use TCE + ACE TCE + Acetone clean In a beaker 5min + 2min Remove the wax Methanol clean In a beaker 2min + agitation Remove acetone residue Dry blow N2 gun Very gently / sample is fragile Removes the moisture Mount on microscope glass slide Bench Use vacuum tape on the edges Mount for n-metal evaporation n-metal evaporation Metal evaporator AuGe/Ni/Au - 1000/500/2000A n-metal evaporation Metal annealing RTA From 20C ramp with 5C/s to 430C, stay 30s; cool with 5C/s to 150C Metal annealing and forming an ohmic contact Cleaving into bars Cleaving station Scribe the edge Cleaves into bars for measurement AR coating, SiOx Dielectric evaporator n=1.8, d=A/4 8 Two Cleavingjplanes .u, ‘p’ -contact 1 | Au, I— ~ I; ‘n’-contact 201 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Appendix 2 Wafer bonding Table A 2.1 Summary of the wafer-bonding process. • Preparing the samples: o Cleave the samples and the transfer substrate to lcm 2 in size, o O2 plasma cleaning (2min, 150mtorr, 100W) of both samples, o Etching of the remaining polymers by a dip in BOE (1:10, 5min). • Cleaning the surface: o Rinse in DI water (5 min). o Clean with TCE (5min, beaker, agitating), o Clean with acetone (5min, beaker, agitating), o Clean with methanol (5min, beaker, agitating), o Rinse in DI water (5 min). o Etch with NH4OH/H2O (1:6, 5min, beaker, agitating), o Final rinse in DI water for lOmin (leave samples in the water). • Bringing the samples into contact: o Bring the samples into contact under DI water, o Gently dry blow with N2 leaving some water on the interface. • Wafer Bonding: o Load both samples (brought close together still with water on the interface) into a graphite fixture, o Load the fixture into the bonding furnace, o Evacuate the chamber by mechanical pump, o Heat for 5min at 100°C. o Fill the chamber with H2 to atmospheric pressure for 15min. o Ramp for 5min to 520°C. o Bond for 30min at 520°C. o Slowly cool down for 15min to room temperature. Unload. 202 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Appendix 3 Properties of a Microdisk Resonant Cavity A3.1 Modal fields distribution Microdisks and microrings support traveling wave modes, which propagate by total internal reflection from the semiconductor/air interface, see Figure A3.1. In resonance, the overlapping waves in the resonator must be in phase after each round-trip in the cavity, i.e. the round-trip phase difference must be equal to an integer multiple of 2n. The resonant condition is k0ne ff 2nR — 2 mri, where ne jf is the mode effective index, R is the disk radius and m is an integer number. These modes are called whispering gallery modes (WGM) and they are spatially distributed close to the outer edge of the disk. Figure A3.1 WGM in a micronng cavity. n0 ne ft => a) b ) Figure A3.2 Conformal transformation method for calculating the modal field distribution The technique used for describing modal distribution is the conformal transformation method [1] combined with the effective index method [2]. The rather involved problem of a three-dimensional dielectric disk (Figure A3.2(a)), is first solved in the vertical direction by converting the disk into a slab waveguide with infinite in-plane dimensions (Figure A3.2(b)). The modal distribution in a slab waveguide is known [2 ] and we find its effective index n^jf in the vertical direction. 203 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The problem, which remains to be solved in the radial direction, is that of a dielectric cylinder with index ne jf and infinite vertical dimensions. This is still rather complicated and includes finding the solutions of the Helmholz wave equation in cylindrical coordinates (A3.1) in both dielectric regions, and then matching the boundary conditions. d2Hz 1 dm 1 d2Hz 12 tt A -^ ~ T + — T - + — - T - r +k‘ffHz:=0 (A3.1) dp2 p dp p l d f l n A better approach is to apply the so-called conformal transformation, which involves change of the variables (/?, (p ) with a new set of variables (u,v): u = R In \R j and v = Rep (A3.2) The solution is assumed in the form, Hz(u,v) = F(u)&xp(jkvv). The resultant equation has the simpler form (A3.3) of an equation for a straight waveguide, with the refractive index a function of the new variable u (A3.4): d 2F r du - + ® n(u)2 F = 0 (A3.3) n{uf = (M<0) exp(2a/R ) - n 2 (A3.4) I «o 1“ > 0 ) The advantage of this approach is that Eqn.(A3.3) is the well known wave equation and there are many numerical techniques for solving it, even with arbitrary refractive index profile (A3.4) as in our case. We used the Finite Difference numerical scheme [2 ], which is capable of finding the bound states in a waveguide with arbitrary refractive index distribution. The Dirichlet boundary conditions (E=0 204 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. at the boundaries) are used. After completing the calculations for the separate fields in both vertical and radial directions, the total field distribution could be found as their product and the accuracy is restricted by the effective index method itself. 235 X C D ■U 3.3 c a ,325 o a2 C O £:ai5 C D c c ai _ j_______ ■ — i i i i i ■ 0 Q5 1 1.5 2 25 3 a5 4 4.5 5 . 3 x 1 0 1 2 4 6 0 3 5 x 10 — a : C O 0.2 .■=0.15 0.1 |O05 0.5 4.5 25 25 0.2 C D -0.2 •a o ^ -0 .4 Distance [irj xio* a) solution in the vertical direction Distance [m ] xio6 b) conformal transformation (radial direction ) Figure A3.3 Example graphs from the solution of a microdisk problem. Figure A3.3 shows an example calculation for the field distribution in a microdisk resonant cavity with radius i?=5pm. Figure A3.3(a) is the solution in the vertical direction. The disk waveguide consists of a 0.45jim InGaAsP core with refractive index n=3.34 and InP cladding with n=3A7. The solution found has a field distribution as plotted on the bottom left insert with an effective index of nen=3.25165 shown as a bound state with a dashed line in the refractive index profile plot on the top left graph. A more interesting plot is Figure A3.3(b) showing the solutions in the radial direction. The top right graph is the refractive index profile of the disk cavity after the conformal transformation is made. On the horizontal axis, we plot the distance in 205 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the radial direction, i.e. d=0 is the center of the disk and J=5|im is its outmost edge. Notice the exponential increase of the refractive index (see. eqn.(A3.4)), which confines the light close to the disk edge. The first two modes are plotted in the bottom right graph. Another characteristic property of a microdisk cavity can be also seen on the same graph. The conformally-transformed refractive index still exponentially increases beyond the disk edge (<i>5|im). Therefore, there is a turning point (outside this graph) where the refractive index becomes larger than the modes effective index, shown with the dashed line. After that point, the propagation constant fi is again real, and the field exhibits oscillatory pattern characteristic of a traveling wave - i.e. the field propagates by ‘tunneling’ through the triangular barrier formed from the semiconductor/air interface and the energy flows out of the cavity. Strictly speaking, the microdisk modes are not bound states (they are quasi-bound) because of this energy loss. Therefore, the use of finite difference techniques, which is good only for finding the bound states with Dirichlet boundary conditions (E=0 at the boundaries), is questionable. Fortunately, the energy loss associated with the bending of the cavity in smooth and high index contrast rings is negligible. Thus, the error in the estimated ‘bound’ state is very small, if the calculation window is truncated before the turning point, as shown in Figure A3.3. From the above graphs it can be inferred that the field is concentrated very close to the disk/air interface. Furthermore, Reyleigh scattering theory proves that the light scattered from dielectric imperfections is proportional to the integral of the product of the electric field and any refractive index perturbation. Thus, the performance of 206 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the microresonators is greatly affected by scattering from sidewall roughness. Such scattering leads to two deleterious effects. First, the energy is lost due to the coupling into the radiation modes, which leads to low Q microcavities. Second, the surface roughness may also cause coupling between the forward and backward propagating waves, if the correlation length of the surface perturbation phase matches them. This will split the degeneracy existing in the resonant frequencies of both modes by 8(0, proportional to the strength of the coupling, and the measured transfer characteristics will exhibit double peaks. This will effectively lower the Q value, but will depend on the details of the surface roughness and will not be considered hereafter. A3.2 Calculation of the power coupling coefficient After having the field solutions, it will be our task to find the power coupling coefficients between the bus waveguides and the disk. This is of particular importance since the resonator performance is strongly dependent on this coefficient. Here the problem is more involved, because the coupling occurs between straight and curved waveguides. Furthermore, the well-known coupling of modes in space theory [3] accounts only for the phase mismatch due to the difference in propagation constants. We have to account for the phase mismatch caused by the bending of the wavefronts in a curved waveguide. The approach we undertook was to use the coupling of modes theory (CMS) for curved transmission lines [4], simplified by Love [5] for dielectric waveguides. 207 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Now we will discuss the coupling between curved waveguides. The difficulty lies in the fact that the phase fronts of the fields are no-longer planar and perpendicular to the propagation direction. The geometry is shown in Figure A3.4. Figure A3.4 Coupling between curved waveguides x and y are one -dimensional curvilinear coordinates, the amplitude of the fields are aj and az and the propagation constants are J3i(x) and /^(y) respectively. The increment of the field dai(x) along x results from contribution due to internal mechanisms and contribution from line2 , due to the mutual coupling. It is assumed that the main coupling occurs between points x and y such that 0 = 0 i= 0 2 , where 0j and 0 2 are the angles between the vector xy and the tangents to the lines. This is a first order approximation. From energy considerations and introducing the coupling coefficients c, we can write: da, (x) = jfi, (x)a, {x)dx - jc2 {y)a2 (;y)dy ( A 3 5) da2 (y) = jp 2 (y)a2 (y)dy - jc* (x)a, {x)dx If we assume that there is no power dissipation in the coupling region, and the power is conserved along each infinitesimal section (dx and dy), then it could be shown that c,dx = c * 2d y . The coupling coefficient is proportional to the scalar-vector product of the fields in the lines. Therefore, it could be assumed that the coupling coefficients between nonparallel lines can be obtained by multiplying the coupling coefficient of two parallel lines, C o, by cos(t c -20)\ 208 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C j = c0 cos(tt - 20). c2 = c0 cos(/r - 2©). ' d y ' 1'2 dx dx \ dy j V " / V/2 dx (A3.6) From the above equations one can obtain the following coupled mode equations (CMEs) for curved lines: dax dx da = jfii® 1 + jc ta 2 2 _ (A3.7) — j c 2ai These equations could be further simplified [5] if we note that the waveguide divergence angle ^=^r-2 0 have to be very small < p«l, otherwise negligible coupling will take place. Hence the approximation cos(ft-20)= 1 can be made in the equations for the coupling coefficients. The second simplification is to ‘project’ the CMEs onto a straight line with coordinate z, which represents an effective propagation direction. Rewriting (A3.7) with respect to z: dax dz da2 dz -J + / ci A dz ) dy^ dz az + j dx' dz ) dy_ 1 dz (A3.8 ) where a, fd and c’s are functions of z (aj=aj(x(z))). We note that c2dyI dz = c2 (dy/ dx)(dx/ dz) = cx dx/ d z, and thus the coupling coefficients are identical. Finally, the simplified CMEs are: 209 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. JPldf a2 jCa\ (A3.9) where - / V = A “ . (A3.10) and C~c0, the coupling coefficient for uniform, parallel waveguides with the same separation. The equations (A3.9) appear as the standard CMEs for ‘parallel’ waveguides whose effective propagation coefficients are determined from the geometry of the coupler by eqn.(A3.10), whilst C has the same functional dependence on separation as for uniform couplers with straight waveguides. Now we will find the total power-coupling coefficient by solving numerically (A3.9). The additional information we need is how to relate the propagation constants to the particular geometry we have, namely vertically coupled microresonators. The coupling is between linel (a straight waveguide) and line2 (curved microdisk with radius R, vertically coupled with minimal separation distance dc). Each waveguide is assumed to be uniform with propagation constants J3j and fh respectively. We choose the z-axis to coincide with the x-axis along the straight waveguide, so that x=z and dx/dz= 1, and thus fiie ff=(3i. The y-axis is curvilinear and is aligned along the periphery of the disk cavity. Using the criteria that the corresponding coupling points (having the largest contribution to the coupling coefficient) are defined as &=©i-& 2, it follows from simple geometric 2 1 0 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. X ( y | consideration that these points are related by: — = tan — . Hence the effective 2R K.2R J propagation constant /heff is found from eqn.(A3.10) as: (A3.11) where z « R . Note that at z=0 (the point where the bus waveguide is exactly below the disk cavity - the smallest separation) the effective disk propagation constant is equal to fh.- The coupling coefficient C is equal to the coupling coefficient calculated for straight waveguides, c0, positioned at distance I from each other. This distance, between the corresponding coupling points as defined above, is a function of z and could be written: The coupling coefficient for parallel waveguides could be found [2] as an overlap integral between the disk, and bus waveguide fields multiplied with the refractive index distribution of both waveguides, integrated over the cross sectional area: Thus, the final CMS equations (A3.9) for curved waveguides resemble the “usual” ones but with a disk propagation vector and coupling coefficient dependent C = c0 (/(z)) (A3.12) (A3.13) (A3.14) 211 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. on the position z. The approach to find the power-coupling coefficient is the following: 1. Find the 2D modal field distribution and the propagation constants fa and 0 2 of the bus waveguides and the microdisk, with the method described in the first section of this appendix. 2. Find the coupling coefficients between the disk and bus waveguide modes as an overlap integral, eqn.(A3.14), of the fields with the refractive index distribution, integrated over the cross sectional area. Find this coupling coefficient, c0(l), as a function of the separation distance by simply shifting spatially the fields and the index distributions and calculating the coupling with (A3.14). 3. Solve (A3.9) numerically, with j3uff=/3i and by using eqns. (A3.11) and (A3.13) with C = c0( l - l(z))- The CMS differential equations are solved using the fifth order Runge-Kutta method [6 ]. The overall power- coupling coefficient is calculated as the ratio of the power coupled into the ring (measured at a distance far from the coupling region) to the input power in the bus waveguide. The equations are integrated with unity excitation in the bus waveguide, i.e. with the following initial conditions: ai(0)=l anda2(0 )=0 . Figure A3.5 is an example of the calculations described in this section. The simulated structure is a disk with a radius i?=5|im; bus and disk waveguide core have a thickness of <f=0.45fxm; and a bus waveguide width of w=0.7|im. Figure A3.5(a) 212 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. plots the first two disk modes calculated as discussed above by using the conformal transformation method. 250 1 50 1 S O 100 1Q O n i l n H K 50 10O 1 SO 200 250 300 n = 2 . 9 6 8 4 50 10O 150 200 250 300 n « = 2 . 7 0 0 1 a) microdisk modal distribution in a disk with R=5|im. W aveguide thickness =4.Se-OQ7 ; width =7e-OD7 SO 100 150 200 250 300 n =3.1 565 S O T O O 1__ n =2.8069 b) modal distribution in the bus waveguide with d=0.45|im and w=0.7pm. r e d - o - ( d = 1 ); m an g s n t a - * - ( d = 2 ) ; g r e e n - x - ( d = 3 ) ; b lu e - v- (d = 4 ) 3 5 0 . 3 2 5 0 . 2 1 5 0 . 1 0 5 0 lm J Sep D Is ta n c x 1 0 i i T 3 0 . 7 0 . 6 0 .5 0 . 3 0 . 2 0 . 1 0 6 8 4 2 Se p . D is tan ce [ m] x 1 0 '7 c) power coupling coefficient between the top four fields, o-first disk mode, * - second disk mode. Figure A3.5 Calculation of the model field distribution and the power-coupling coefficient. 213 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The vertical line in the middle of the graph is the semiconductor - air interface and the two horizontal lines are the disk waveguide core layer (the four graphs show the cross-sectional view of the region where the bus and the disk waveguides are closest to each other in vertical direction.). Figure A3.5(b) shows the spatial distributions of the first two modes in the bus waveguide. Figure A3.5(c) plots the calculated coupling coefficient. This is the overall power-coupling coefficient as a function of the separation distance in the vertical direction between the bus and microdisk waveguides. Here the lines with style ‘-o-’ describe the coupling to the first disk mode (Figure A3.5(a) - left) and lines with style describe the coupling to the second one (Figure A3.5(a) - right). Also, the left inset of Figure A3.5(c) describes the interaction of the first bus-waveguide mode (Figure A3.5(b) - left) and the right graph is for the second bus-waveguide mode (Figure A3.5(b) - right). In other words, Figure A3.5(c) shows the 2x2 coupling interactions between the four fields shown in a) and b) graphs. Notice that the first bus mode couples only to the first disk mode, and changing the separation changes the coupling exponentially. This dependence is to be expected since the process is an evanescent coupling through the separation layer. To obtain a coupling coefficient of 2% we need a waveguide separation of about 0 .6 -0 .8 }im. Also, the second bus mode couples strongly to the second disk mode (Figure A3.5(c)-right , ‘-*-‘) because of the smaller mode mismatch: we obtain K~0.55 @ J=0 .2 pm compared to k~0.35 @ J=0.2(a.m for the coupling between the first modes. The bus waveguide modes are not orthogonal to the microdisk modes. For example, 214 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the second bus mode is coupled somewhat strongly to the first disk mode, thus leading to second order cross-coupling effects. In conclusion to this appendix we have developed modeling tools, which allows us to predict the coupling efficiency in a vertically coupled microdisk structure. Second order effects, such as the mutual coupling between the modes, are detected and have to be avoided by carefully designing the u-disk and waveguide sizes. 215 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Appendix 4 Coupling of modes in time formalism In this formalism the microresonator is a lumped device and only the total energy and power in the ring are considered [7]. The cavity supports a traveling wave of amplitude A(t), which is normalized such that |A (f|2 represents the total power flowing through any cross section of the ring waveguide at time t. The ring may also be viewed as a lumped oscillator of energy amplitude a(t) normalized so that |a ( t|2 represents the total energy stored in the ring. Stored energy and power flow in the ring are related through: \ a { t f = \ A { t f — (A4.1) a relation which could be used to translate between the ‘power picture’ and the ‘energy picture’. R is the ring radius and vg is the group velocity. The energy in the ring is supplied by the incident wave of amplitude S,. S; St The filtered output wave Sd travels to the Figure A4.1 CMT theory approach. detector and power not picked up by the ring travels as the transmitted wave St. (Figure A 4.1 ). Here [Sj2 is the power in the waveguides. The lumped oscillator has a resonant frequency of (D 0 and an amplitude decay time-constant of 1/t. This decay rate is related to: (i) the power leaving the disk 1/td 216 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. towards the detector, (ii) the power coupled to the transmitted wave l/tt and (iii) the power lost inside the disk due to scattering or absorption, 1/tf. 1/v= 1/T t+1/Td+1/% From energy flow considerations: where /I is the mutual coupling coefficient between the ring and the waveguide, S t = S ( - jf ia . Now the task is to relate the mutual coupling coefficient j L i to the amplitude decay rate 1/r and the power coupling coefficient k between the waveguides. The relation between fi and the external decay rates i/Tf and 1/fa is determined by power conservation. One considers the case in which the ring is excited with an ! ! 2 initial energy |a0| , there is no detector waveguide l/td-0, and no incident signal Si=0. From (A4.2) the energy decay rate, neglecting the internal losses, is: If we consider the same case in the power picture, then the traveling wave of amplitude A(t) in the ring couples to the transmitted wave St : where iris the fraction of power coupled out of the ring over the interaction region, i.e. this is the power coupling coefficient. Now the power leaving the ring is the time (A4.2) (A4.3) (A4.4) 217 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. rate of change in energy, assuming Si=0, and using (A4.2) and St = S( - jjLia (@ Si=0) we have: = K ' f — = |s ,f < A4-5> From the above equations the sought-after relationship between j n , K , and 1 /% could be written: f j 2 = Kvg = 2/rt (A4.6) This equation serves to convert between coupling coefficients with different normalizations (power or energy). The power appearing in the detector waveguide could be written by power conservation: M 2 = M 2 -I s . f = — M* < A4-7> 7d After solving for the amplitude transmission coefficient, by considering a steady state incident signal St with time dependency -exp(jax), we found from St - St - j/Lia and (A4.2): j{o)-a)0)+ ( y + y - y t _ _ L — ___________________ V / d / 1 ' ( AA S ') The quality factor of a cavity is given by the time averaged stored energy per optical cycle, divided by the power coupled or scattered out of the resonator. We make use of the fact that 1 / r = o 0/2 g , where Q is the quality factor associated with the loss rate r. The Q associated with internal loss and scattering is defined as 218 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Ql 1 =aXQ j m eff (here a is the absorption coefficient [cm-1 ], and ne g is the mode effective index). The Q associated with the coupling loss is defined as index, ic t j is the power coupling coefficient between the transmission waveguide ‘t \ detector waveguide ‘d’ and the ring). We can rewrite (A4.8) as: where Aco is the detuning from the resonance and An accounts for the change of Eqn.(A4.9) is a very important and useful tool for understanding the resonant and assume resonant conditions, Ad)=0. We see from (A4.9) that t=0 (no transmission) when Q + Q^x = Q~l , i.e. even in the presence of loss all of the input power might be extracted, although this requires that the input and detector waveguides are no longer symmetrically coupled to the ring & Q~l . Also, if we use this device in the so called overcoupled regime Q^1 « Q~x with only a single bus waveguide, then at resonance, from (A4.9), we see that the transmission is unity, t = - 1. This means that the amplitude of the signal is not changed, but the ring (where R is the ring radius, neff is the mode effective t = (A4.9) resonance caused by the change of the refractive index (terms of the An2 order are neglected). behavior of a ring or disk. Let’s first set aside the change of the refractive index An 219 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. introduces a phase change of n into the bus waveguide. Thus we may use this configuration to build a Mach Zender interferometer type of modulator. We now have three mechanisms for affecting the transmittance of the resonator. These mechanisms are: (i) change in the absorption, i.e. g / 1 by the EA effect ( Qd ,Qt are fixed by the coupling coefficient, i.e. by the device geometry); (ii) introduction of gain into the medium (g= -a); (iii) change in the refractive index An by the EO effect or PCI, which shifts the resonance. Disk with R=10|im; k =4% ' ■ ' ______, ______I ______.______L. W 1.0 - - g =a=0 cm \ . a=10 cm - a=30cm - a.=50 cm ■•--g=oc=0 cm' ■ ■ - gt =2.5 cm'1 - - g =3 cm'1 — - g=3.4 cm'1 Normalized Detuning ( ( o - a > ^ /( o 0 xIO"4 Figure A4.2 The effect of changing the cavity loss on the transmission and quality factor of a resonant coupler. A graph showing the effect of the first two mechanisms on the performance of a microdisk resonator (change in the quality factor of the cavity) is Figure A4.2 . The simulation (eqn.A4.9) is done for a i?=10|im disk, using a realistic value for the power coupling coefficient of k=A%. Without internal loss (a=0cml) the microdisk 2 2 0 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. can completely extract the power from the input waveguide, 7= 0. Introducing gain into the cavity will compensate for the loss and eventually, when the gain is larger than the loss, will amplify the propagation mode. The quality factor of the cavity will increase and the transmission will initially decrease until the gain equals the internal loss (T=0 at this point). Further increase of the gain will lead to even larger then unity transmission. When an electroabsorption region is used, the increase of the modal loss will lead to constant decrease of the quality factor and increase of the transmission at resonance. Disk with R=10|im; k =4%, ct=4cm'1 c 1 -0 - o ’ 5 5 m o.8- E m C 0.6- C B — An=0 — - An=0.0006 -• - An=0.0012 — An=0.0019 — An=0.0025 j _ N 0.4- 1 0 2 " O z 0.0 -10.00 -5.00 0.00 5.00 10.00 x10 Normalized Detuning ( g > - c o 0 ) / ( d 0 Figure A4.3 The effect of changing the refractive index of the cavity on the transmission and quality factor of a resonant coupler. The third mechanism for affecting the cavity response is the change of the resonant frequency by affecting the modal refractive index. The simulated disk has a radius of i?=10pm, coupling coefficient of k=A% and internal loss of cN cm '1 . The change in the resonant frequency, shown in Figure A4.3 , can be used for 221 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. modulation of the input laser mode, by positioning the mode at resonance and varying the modal cavity index. The transmission at the particular wavelength will increase from the value at resonance, to unity far from it. 2 2 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Appendix 5 Quantum Confined Stark Effect A5.1 Finding the excitonic binding energy A S . 1.1 T w o -p a r tic le s S c h r d d in g e r e q u a tio n In this appendix we consider the excitonic absorption in a quantum well structure in the presence of a uniform applied electric field, and will follow closely [8 ]. The effective mass equation for the electron and hole in a quantum well could be written: 2 ' A 4m?s\re - r hu where He,h are the single particle Hamiltonians: H , = - f r V 2 .+V,( O and = ^ - V j - V „ ( r J (A5.2) 2 me 2mh The electron potential Ve (re) and hole potential Vh (r,,) may also include the effect of the electric field |c|F.re or - \e\F.rh. The electric field is assumed to be applied along the z axis, which is the growth direction of the quantum well: 223 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A S . 1 .2 S o lu tio n o f th e e le c tr o n - h o le effective m a s s e q u a tio n w ith e x c i to n i c e ffe c ts Using the transformations for the difference coordinate and the center-of-mass . _ m o -m * hph coordinate systems for the x and y components: p = pe - p h, R t = —— ----- S -JL, M where M =m*e +m*h, m* = , and p = xx + yy , we obtain: me +mh h2 y2 _ h2 d2 n 2 a 2 2mr p 2me d2 H E - E J Z e 2ml d2Zh +ve{ze)+vh{zhy $(re,rh)= 0 (A5.3) The dependence on Rt leads only to a kinetic energy term, and the solution can be written: t K j .R t ^ ( re ’ rh ) = - J j - F { p , Z e,Zh ) , whereF ( p ,ze,zh) is the exciton envelope function, which satisfies: (A5.4) 2 mr f h2K 2 N 4 ^ J r e - r J E - E - v ' 2 M , F(p,ze,z j = 0 (A5.5) The center of mass of the electron-hole pair is thus moving freely with a kinetic energy h2K 2/2M . Note that H e = - ^ ~ + Ve{ze) and H h = - ^ ^ T ~Vh{zh) 2m„ dz: 2mh dzh are the single particle Hamiltonians, which satisfy the one-dimensional Schrodinger 224 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. equation for a particle in a box model: H(ze)fn{ze) = Eenf n(ze) and H(zh)gm(zh) = Ehm gm( z j . In this equation f n(ze) and gm(z h) are the free-electron and free-hole wave functions in the absence of any mutual interactions. Now the task is to find the exciton envelope function. With the Coulomb interaction term the solution F(p,ze,zh) is more complicated. However, using the completeness properties of the solutions of the single particle equations, we can express the envelope function as a linear combination of them: ^(p> = (p )fn )sm ) (A 5 -6) n m With this expression substituted in (A5.5) we obtain: / 2 2 " N * - V ? + £ „ . - £ w - — r ------------ r - ( £ - £ , ) / . ■ f e , ) s , - ( z „ k v ( p ) = 0 ( A 5 . 7 ) v 2m, ' “ “ 4 ffi,|r ,-r J ' ‘ j Multiplying the above equation by f*{ze) and g*n{zh) and integrating over and Z h, one obtains: v 2 m r where: Vp - Vnm (p) (p) = Eex(j)nm (p), (A5.8) Vn m (p) = X \dzef* (te )fn k )/* * £ « (zh )g (Zh )------ f------------------ w 4*e,\p2 + \ z , - z > \ T « l dze\fn(zef \dzh\gm{zh] (A5.9) e2 4m:s \p 2+\ze- z h\2Y2 and Eex= E - ( E g + E en- E hm ) (A5.10) 225 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In the last equation we ignore the coupling between different subbands and only ri = n and m - m are taken into account. The solution to eqn.(A5.8) will provide us with the exact exciton bound energy, Eex under different bias conditions. For the solution of this problem we will need the single particle bound energies, Een and Ehm and the single particle wavefunctions f n (ze) and gm (zh) for the particular quantum well under bias conditions. They can be found with the transmission matrix approach [9]. A 5 . 1.3 V a r ia tio n a l m e t h o d fo r th e e x c ito n ic p r o b le m The most common approach to solve the excitonic problem is the variational approach. Noting that the 1s t state solution of ^(p) behaves like the solution of a hydrogen atom: < * (P (A 5 .ll) \ 7 t A then one can write: (A 5 J2 ) The wavefunction satisfies the normalization condition: { < / > \ < f > ) = 1 • We find the variational relationship of the exciton energy with respect to the variational variable X: ■ l r A 2m A 40S,. k, 2 1 ( 2 Ip + \ Z e - Z h \ iy2 (A5.13) 226 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. This is a variational equation of the excitonic binding energy. With known single particle functions f n(ze) and gm{zh), the excitonic energy can be found by minimizing the above equation with respect to A The value of A at which the energy minimum is the radius of the exciton, Re x = A . A5.2 Absorption coefficient The general formula for the absorption coefficient in SI units is: a(»a>)= S(e, - E , - » « ) [ / ( £ , ) - / ( £ , ) ] , (A5.14) 2 7te where, C0 = nrc£0m la When the solution to (A5.14) is obtained it will contain a set of binding energies and continuum of states. The quantum number for the exciton state is denoted as x. The absorption coefficient is obtained as: a{hco) = C0^ Y j ^ E C (P = 0)«'P J n S{EX -hco) (A5.14) where <pZ is the exciton wavefunction (A5.11), In m = jf*(z)g*m(z)dz is the overlap between the conduction band and the valence band single particle wavefunctions and Ex = EZ + Ee x =Eg + Ee n - E h m + Ee x is the exciton binding energy. |e.pC T [ is the matrix element and is found to be: 2 2 7 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. TE polarization: — (l + cos2 6n m ), heavy - hole exciton P -P „ f =MlM(E,) = M U 4 (A5.15) I ^ — (5 - 3 cos 2 6nm ), light - hole exciton TM polarization: f 0 , heavy - hole exciton ! 1 b e.v\ = M 2 bM{Et) = M 2 b \ 1/ \ , , , • (A.5.16) 1 (1 + 3 cos 6nm ), light - hole exciton where M l = ^ -E „ is the bulk matrix element, which is tabulated for different h 6 p materials and cos2 6n m « (Ee n + \Ehm |)/(Em + \Ehm \ + Et) depends on the transverse energy detuning Et from the minimum of the conduction band. If we assume that there is no mixing between the different subbands and consider pairs of transmission n=Cl and m=HHl, and thus drop the double summation over nm and treat each pair independently, then: = C „|-£|V a c ( p = 0)e.p„ /„ f S(EX -h a,). (A5.17) * X A 5 .2 .1 D is c r e te - s ta te c o n tr ib u tio n s : The first state contribution is: a{hm) = CQ H - ^ l| c . p j nm \2 — — ^ ----- 7 (A5.18) Vyrc A1 / {Ex-hco) + y where the delta function is replaced by a Lorentzian function with finite linewidth 2 y. 228 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A 5 .2 .2 C o n tin u u m - s ta te c o n tr ib u tio n s : In this case the sum over the states x becomes the sum over a continuum distribution of states kx and ky, <a 5 - 1 9 ) x kx ky Then the continuum state contribution to the absorption coefficient is: = C« I ' - f "Ue,M{E, )\r ( o f r - ^ L - ^ -r (A5.20) A 5 .2 .3 T o ta l a b s o r p tio n s p e c tr u m : The complete absorption spectrum can be written as the sum of the bound-state and continuum-state contributions: 2 kl cc[h(o) = Ct3 —M l\l L lin t ..+ )|f» '(o f 7 r — . * ( E Z - E . - n c ) K r 4 J „ M (0) states 1 2 + f + |2 2 + y (A5.21) A 7 C E 1 7 1 where a 0 = V > = ArfTT— ^ and = E * + E - = + + ’ mre 2 ft (4;k J A5.3 Solving the QCSE problem. In this section we will apply the above formalism to solving for the QCSE in Ino.47Gao.53 As/InP quantum well. The thickness of the QW is 85A and has emission energy of 0.8eV (1.55pm). The effective electron, heavy-hole and light-hole masses 229 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. are 0.043mo, 0.292mo and 0.041mo respectively. The bandgap energy is jB^=0.75eV. Conduction band discontinuity is obtained to be AEc=0.237eV and the valence band discontinuity is zlEv=0.356eV. First the quantum well problem is solved with the transfer matrix approach, which is capable of finding the single particle bound energies Ee n and E h m and the corresponding electron and hole wavefunctions f n(ze) and gm(zh), in a well with applied external bias. The values of the energies are shown in Figure A5.1 (c) for field ranging from 0 to lOOkV/cm, and the wavefunctions are plotted in Figure A 5.1 (b) for the boundary values of the fields. wavefunctions is calculated and the values for the c-hh and c-lh transitions are plotted in Figure A5.1 (d). The exciton binding energy is found from minimizing eqn.(A5.13) with respect to the variational parameter A To evaluate the integrals, the single-particle wavefunctions, calculated in the previous step are needed. The minimum value of the expression is the exciton binding energy Eex and the value of X at which this happens is the exciton radius. Both parameters are plotted in Figure A5.1 (e) and (f) respectively. The overlap integral between the conduction and valence 230 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1S000 14000 E = 0 kV/cm *12000 E_ = 100 k V /c i 10000 3000 ■S 6000 I " 4000 | 2000 0.78 0.80 0.82 0.84 0.86 1 .0 08 06 S T a > 0 4 ro 0.2- Jj O tO - -02 -04 E^IOOMSfem 06 04 02 00 -02 ■04 Energy [eV] (a) absorption spectra with and without applied field 0 40 80 120 160 0 40 80 120 160 Distance [A ] Distance [A| (b) QW under the conditions shown in (a) — E h 9 £ 0.90 § ■ °-88 t : 0.8E I 20 40 60 80 Electric Field [kV/cm] 100 2 0 4 0 60 8 0 100 Electric Field [kV/cm] (c) the energy of the QW states under different field (d) the overlap integral as a function of the field - E x h 0 20 40 60 80 Electric Field [kV/cm] < 200 A' ■ ■ A 20 40 60 80 100 Electric Field [kV/cm] (e) exciton binding energies as function of E (f) exciton radius as a function of E Figure A5.1 Properties of the QCSE The total absorption coefficient is calculated from eqn.(A5.21). The broadening factor y is assumed to be 3meV. To evaluate the integral, one needs the values of the 231 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. overlap integral, In m , the values of the exciton and single-particle bound energies Eex, Ee n , E fim ., and the exciton wavefunction at zero radius. All of these parameters were evaluated during the previous steps. The remaining constants are defined above. The absorption spectrum with the first two transitions (e-hh, e-lh) is plotted in Figure A5.1 (a). Figure A5.1 (a) shows the calculated absorption at 0 and lOOkV/cm. If we have a mode below the bandgap (for example at 0.79eV) and by changing the applied field we vary the absorption at that energy, modulation of the light intensity could be achieved. Figure A5.1 (b) shows the QW states and electron, heavy hole and light hole wave functions under these conditions (Ino.53Gao.47As/InP QW). The basic properties of the QCSE can be understood from these graphs. By increasing the electric field: (1) the energy of the QW states decreases Figure A5.1 (c) - this is so- called Stark Shift; (2) the wave functions are shifted to the opposite ends of the QW (Figure A5.1 (b)) and thus the overlap integral between CB-HH and CB-LH wave functions decreases (Figure A5.1 (d)), causing decrease of the oscillator strength of the absorption peak; (3) the exciton binding energies also decrease (Figure A5.1 (e)), but this change is small compared to (1); (4) the exciton radius increases (Figure A5.1 (f)). All of these four effects lead to the overall behavior of the QCSE shown on Figure A5.1 (a). 232 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. References 1 M.K.Chin, S.T.Ho, “Design and Modeling of Waveguide-Coupled Single-Mode Microiing Resonators”, J. Lightwave Technology , vol. 16, no.8 , pp. 1433-1446, 1998 2 L.A.Coldren, S.W.Corzine, ‘‘ Diode Lasers and Photonic Integrated Circuits ”, John Wiley & Sons, Inc. 1995 3 H.A.Haus, W.P.Huang, S.Kawakami, and N.A.Whitaker, “Coupled-Mode Theory of Optical Waveguides”, J. Lightwave Technology, vol. LT-5, no.l, pp. 16-23, 1987 4 M.Matsuhara, Aa.Watanabe, “Coupling of curved transmission lines, and application to optical directional couplers”, Journal of the Opt. Soc. Of America, vol.65, no.2., pp. 163- 168,1975 5 D.R.Rowland, and J.D.Love, “Evanescent wave coupling of whispering gallery modes of a dielectric cylinder”, IEEProc.-J, vol. 140, no3, pp.177-188, 1993 6 W.H.Press, B.P.Flannery, S.A.Teukolsky, W.T.Vetterling, “Numerical Recipes in C”, Cambridge University Press, 1988 7 B.E.Little, S.T.Chu, HA.Haus, J.Foresi, and J.P.Lain, “Microring resonator channel dropping filters,”, J.Lightwave Technology, vol.l5,pp.998-1005, 1997. 8 S.L.Chuang, “Physics of optoelectronic devices ”, J. Wiley & sons, Inc., 1995 9 L. Coldren and S. Comizine, “Diode lasers and Photonic Integrated circuits ”, Wiley Series in Microwave and Optical Engineering, 1995 233 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Dispersive properties in photonic crystals applications to beam-steering and wavelength demultiplexing

PDF

III-V compound semiconductor nanostructures by selective area growth using block copolymer lithography

PDF

High performance components of free -space optical and fiber -optic communications systems

PDF

Dispersive and nonlinear effects in high-speed reconfigurable WDM optical fiber communication systems

PDF

All-optical devices based on carrier nonlinearities for optical filtering and spectral equalization

PDF

Full vectorial finite element analysis of photonic crystal devices: Application to low -loss modulator

PDF

Coupled multiple micro-resonators design and active semiconductor micro-resonator fabrication

PDF

Experimental demonstration of techniques to improve system performance in fiber -optic communication systems using subcarrier -multiplexed and digital baseband signals

PDF

III nitride selective area growth and application

PDF

Electro-optic and thermo -optic polymer micro-ring resonators and their applications

PDF

Finite-difference time-domain modeling of photonic crystal microcavity lasers

PDF

Optical modulators for integrated photonic systems

PDF

Design, fabrication, and integration of a 3-D hybrid electronic/photonic smart camera

PDF

Experimental demonstrations of all -optical networking functions for WDM optical networks

PDF

High -speed CMOS continuous -time switched -current sigma -delta modulators

PDF

Gallium arsenide based semiconductor laser design and growth by metal-organic chemical vapor deposition

PDF

Photonic microcavities and nanocavities embedded with InAs quantum dots

PDF

Admission control and burst scheduling in multiservice cellular CDMA systems

PDF

An evaluation of ultra -wideband propagation channels

PDF

Experimental demonstration of techniques to improve system performance in non-static microwave frequency analog and digital signal transmission over fiber -optic communication systems

Asset Metadata

Creator Djordjev, Kostadin Dimitrov (author)

Core Title Active microdisk resonant devices and semiconductor optical equalizers as building blocks for future photonic circuitry

School Graduate School

Degree Doctor of Philosophy

Degree Program Electrical Engineering

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag engineering, electronics and electrical,OAI-PMH Harvest

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Dapkus, Daniel P. (committee chair), [illegible], Edward (committee member), O'Brien, John (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-247492

Unique identifier UC11339323

Identifier 3093754.pdf (filename),usctheses-c16-247492 (legacy record id)

Legacy Identifier 3093754.pdf

Dmrecord 247492

Document Type Dissertation

Rights Djordjev, Kostadin Dimitrov

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA