UNIVERSITY OF CALIFORNIA. Santa Barbara. The Integration of Mach-Zehnder Modulators with Sampled Grating DBR. Lasers

Transcription

1 UNIVERSITY OF CALIFORNIA Santa Barbara The Integration of Mach-Zehnder Modulators with Sampled Grating DBR Lasers A Dissertation submitted in partial satisfaction of the requirements for the degree Doctor of Philosophy in Materials by Jonathon Scott Barton Committee in charge: Professor Larry A. Coldren, Chair Professor Daniel J. Blumenthal Professor Nadir Dagli Professor Steven DenBaars Professor Evelyn L. Hu September 2004

2 UMI Number: UMI Microform Copyright 2004 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 PO Box 1346 Ann Arbor, MI

3 The dissertation of Jonathon Scott Barton is approved. Evelyn L. Hu Daniel J. Blumenthal Steven DenBaars Nadir Dagli Larry A. Coldren, Committee Chair September 2004

4 The Integration of Mach-Zehnder Modulators with Sampled Grating DBR lasers Copyright 2004 By Jonathon Scott Barton iii

5 TABLE OF CONTENTS Abstract... vii Vita... ix Symbols and acronyms... xv Acknowledgements... xvi INTRODUCTION Direct Modulation External Modulation Electro-absorption Modulators Mach-Zehnder Modulators Mach-Zehnder Bias Approaches Traveling Wave Devices...20 References...21 CHAPTER I: DEVICE INTEGRATION Widely-tunable laser design Semiconductor Optical Amplifier(SOA) Integration Dual SOAs Optical Feedback and Reflection Linewidth Measurements Relative Intensity Noise...47 References...50 CHAPTER II: MOCVD GROWTH & FABRICATION Semiconductor Epitaxial Structure Quantum Well Design Conducting Substrate Base structure Growth Characterization Semi-insulating substrate growth Regrowth Zn Doping of InP and InGaAsP Transmitter Fabrication...70 References...73 CHAPTER III: LUMPED MODULATOR DESIGNS Device Efficiency DC Extinction Curves Franz-Keldysh Absorption...81 iv

6 3.4 Electric Field Effects...84 Linear Electro-optic effect...84 Kerr Effect Carrier Based Effects...88 Plasma Effect...88 Bandfilling Effect...91 Carrier induced bandgap shrinkage Temperature induced bandgap shrinkage Accumulation of Effects High Speed Design Junction Capacitance Minimization Parasitic Capacitance Minimization Fringing Capacitance Multimode Interference (MMI) Design Phase Shifter st Generation Designs nd Generation Designs References CHAPTER IV: SERIES PUSH-PULL DESIGNS Lumped Series push-pull bandwidth Dual RF series push-pull devices Traveling wave modulators Traveling wave matching Transmission line model Traveling wave bandwidth RF loss CPS T-Electrode devices Characteristic Impedance Comparison Measured Bandwidth References CHAPTER V: DEVICE COMPARISONS DC Modulation Efficiency RF Extinction of devices Bandwidth Comparison Chirp Measurements Chirp Measurement Techniques Linearization of Modulators References v

7 CHAPTER VI: CONCLUSIONS AND FUTURE WORK Wavelength Converters References Appendix A: Relevant Material Constants Appendix B: RF Spectrum Analyzer Deimbedding Appendix C: Process vi

8 ABSTRACT JONATHON S. BARTON THE INTEGRATION OF MACH-ZEHNDER MODULATORS WITH SAMPLED GRATING DBR LASERS Some of the latest results of InP based widely-tunable optical transmitters will be presented. Widely-tunable transmitters are seen as a crucial component in Dense Wavelength Division Multiplexing (DWDM) communication systems. Integration of different optical components reduces the costs, insertion losses, and footprint of the device. This work outlines the material design and fabrication aspects to produce high bandwidth, low drive voltage modulation without degradation of the laser and modulator device characteristics with optical/electrical crosstalk. Photonic integrated circuits are particularly susceptible to optical reflections - which can cause excessive chirp, gain ripple, lasing of the Semiconductor Optical Amplifier, higher noise figure, and inter-modulation distortion. Careful design is required to minimize refractive index discontinuities in active/passive interfaces, Multi-mode interference devices, and through angling and flaring at the output waveguide. With the use of a Mach-Zehnder modulator, the chirp parameter can be tailored to maximize the transmission distance through fiber - particularly important at high data rates (10Gbit/s). Using traveling wave electrodes in a series push-pull electrode structure, we are able to demonstrate some of the highest speed widely tunable lasers to date with 40GHz bandwidth. DC extinction Vpi as low vii

9 as 0.6V is demonstrated with high saturation power dual SOA structures. This is all achieved using low k dielectrics (BCB) and highly efficient PN junctions in which compromises must be made to insure high performance in each integrated device region. viii

10 VITA JONATHON S. BARTON 1975 Born in Sacramento, California 1993 Bachelor degree in Electrical Engineering and Material Science - University of California, Davis 2003 Intel Fellow 2004 PH.D in Electronic Materials - University of California Santa Barbara LIST OF PUBLICATIONS [1] Mason B, Barton J, Fish GA, Coldren LA, Denbaars SP. Design of sampled grating DBR lasers with integrated semiconductor optical amplifiers. IEEE Photonics Technology Letters, vol.12, no.7, July 2000, pp [2] Blumenthal DJ, Olsson B-E, Rossi G, Dimmick TE, Rau L, Masanovic M, Lavrova O, Doshi R, Jerphagnon O, Bowers JE, Kaman V, Coldren LA, Barton J. All-optical label swapping networks and technologies. Journal of Lightwave Technology, vol.18, no.12, Dec. 2000, pp [3] Hanxing Shi, Cohen D, Barton J, Majewski M, Coldren LA, Larson MC, Fish GA. Relative intensity noise measurements of a widely tunable sampledgrating DBR laser. IEEE Photonics Technology Letters, vol.14, no.6, June 2002, pp [4] Majewski ML, Barton J.S., Coldren LA, Akulova Y, Larson MC. Direct intensity modulation in sampled-grating DBR lasers. IEEE Photonics Technology Letters, vol.14, no.6, June 2002, pp [5] Shi HX, Cohen DA, Barton J, Majewski M, Coldren LA, Larson MC, Fish GA. Dynamic range of widely tunable sampled grating DBR lasers. Electronics Letters, vol.38, no.4, 14 Feb. 2002, pp ix

11 [6] Skogen EJ, Barton JS, DenBaars SP, Coldren LA. Tunable sampledgrating DBR lasers using quantum-well intermixing. IEEE Photonics Technology Letters, vol.14, no.9, Sept. 2002, pp [7] Skogen EJ, Barton JS, Denbaars SP, Coldren LA. A quantum-wellintermixing process for wavelength-agile photonic integrated circuits. IEEE Journal of Selected Topics in Quantum Electronics, vol.8, no.4, July-Aug. 2002, pp [8] Raring J.W., E. J. Skogen, L. A. Johansson, M. N. Sysak, J. S. Barton, M. L. Mašanović, L. A. Coldren Demonstration of Widely-Tunable Single-Chip 10 Gb/s Laser-Modulators Using Multiple-Bandgap InGaAsP Quantum-Well Intermixing, Photonics Technology Letts July [9] Barton J.S., Skogen E.J., Mašanović M.L., DenBaars S.P., and Coldren L.A., Widely-tunable high-speed transmitters using integrated SGDBRs and Mach-Zehnder modulators. IEEE Journal of selected topics in quantum electronics, Vol. 9, NO. 5, pp September /October [10] Coldren LA, Fish GA, Akulova Y, Barton JS, Johansson L, Coldren CW. Tunable semiconductor lasers: a tutorial. Journal of Lightwave Technology, vol.22, no.1, Jan. 2004, pp [11] Masanovic M.L., V. Lal, J. A. Summers, J. S. Barton, E. J. Skogen, L. A. Coldren, and D. J. Blumenthal, "Design and Performance of a Monolithically- Integrated Widely-Tunable All-Optical Wavelength Converter with Independent Phase Control," accepted for publication in IEEE Photonics Technology Letters, [12] Masanovic M.L., V. Lal, J. S. Barton, E. J. Skogen, J. A. Summers, L. Rau, L. A. Coldren, and D. J. Blumenthal, "Widely-Tunable Monolithically-Integrated All-Optical Wavelength Converters in InP," to be published in IEEE Journal of Lightwave Technology, [13] Hutchinson J.M., J. F. Zheng, J. S. Barton, M. L. Masanovic, M. N. Sysak, J. A. Henness, L. A. Johansson, D. J. Blumenthal, L. A. Coldren, H. V. Demir, V. A. Sabnis, O. Fidaner, J. S. Harris, and D. A. B. Miller, " Indium Phosphide based Wavelength Conversion for High Speed Optical Networks," Intel Technology Journal, [14] Barton JS, Masanovic ML, Sysak MN, Hutchinson JM, Skogen EJ, Blumenthal DJ, Coldren LA. 2.5-Gb/s error-free wavelength conversion using a monolithically integrated widely tunable SGDBR-SOA-MZ transmitter and x

12 integrated photodetector. IEEE Photonics Technology Letters, vol.16, no.6, June 2004, pp [15] Masanovic M.L., V. Lal, J. S. Barton, E. J. Skogen, L. A. Coldren, and D. J. Blumenthal, "Monolithically integrated Mach-Zehnder interferometer wavelength converter and widely tunable laser in InP," IEEE Photonics Technology Letters, vol. 15, pp , [16] Masanovic M.L., E. J. Skogen, J. S. Barton, J. M. Sullivan, D. J. Blumenthal, and L. A. Coldren, "Multimode interference-based two-stage 1 * 2 light splitter for compact photonic integrated circuits," IEEE Photonics Technology Letters, vol. 15, pp , [17] Skogen EJ, Raring JW, Barton JS, DenBaars SP, Coldren LA. Postgrowth control of the quantum-well band edge for the monolithic integration of widely tunable lasers and electroabsorption modulators. IEEE Journal of Selected Topics in Quantum Electronics, vol.9, no.5, Sept.-Oct. 2003, pp [18] Sysak, M.N., J. S. Barton, L. A. Johansson, J. W. Raring, E. J. Skogen, M. L. Mašanović, D. Blumenthal, and L. A. Coldren, Single Chip Wavelength Conversion using a Photocurrent Driven (PD) EA Modulator integrated with a Widely Tunable Sampled Grating DBR (SGDBR) Laser. Submitted to Photonics Tech. Letts. CONFERENCE TALKS [19] Mason B, Fish GA, Barton J, Coldren LA, DenBaars SP. Characteristics of sampled grating DBR lasers with integrated semiconductor optical amplifiers. Optical Fiber Communication Conference. Technical Digest Postconference Edition. Trends in Optics and Photonics Vol.37 (IEEE Cat. No. 00CH37079). Opt. Soc. America. Part vol.1, 2000, pp vol.1. [20] Majewski ML, Barton J, Coldren LA, Akulova Y, Larson MC. Widely tunable directly modulated sampled-grating DBR lasers. Optical Fiber Communications Conference. (OFC). Postconference Technical Digest (IEEE Cat. No.02CH37339). Opt Soc. America. Part vol.1, 2002, pp vol.1. [21] Barton J, Coldren L, Fish GA. Tunable lasers using sampled grating DBRs Digest of LEOS Summer Topical Meetings: Advanced Semiconductor Lasers and Applications/Ultraviolet and Blue Lasers and Their Applications/Ultralong Haul DWDM Transmission and Networking/WDM Components (IEEE Cat. No.01TH8572). IEEE. 2001, pp.2 pp. Invited xi

13 [22] Skogen EJ, Barton J, DenBaars SP, Coldren LA. Tunable buried ridge stripe sampled grating distributed Bragg reflector lasers utilizing quantum well intermixing. LEOS th Annual Meeting of the IEEE Lasers and Electro- Optics Society (Cat. No.01CH37242). IEEE. Part vol.1, 2001, pp vol.1. [23] Barton JS., Skogen, E.J., Masanovic M., S. Denbaars, L. A. Coldren, Integration of a Mach-Zehnder Modulator with Sampled Grating Distributed Bragg Reflector Laser, Proc. Integrated Photonics Research Conference, paper no. 1FC3-1, Vancouver, Canada, July [24] Skogen E.J., Barton J.S., DenBaars S.P., Coldren, L.A., On Tuning Efficiency of Sampled Grating DBR Lasers using Quantum Well Intermixing, Proc.Integrated Photonics Research Conference, paper no. IFC2, Vancouver, Canada July [25] Mašanović M., Skogen E.J., Barton J.S., Sullivan J., Blumenthal D.J., Coldren L.A., Cascaded Multimode Interference-Based 1x2 Light Splitter for Photonic Integrated Circuits Integrated Photonics Research Conference, Vancouver, Canada, July [26] Majewski ML, Barton J, Coldren LA, Akulova Y, Fish G. Wavelength monitoring in widely tunable sampled-grating DBR lasers integrated with semiconductor optical amplifiers. Technical Digest. Summaries of papers presented at the Conference on Lasers and Electro-Optics. Conference Edition (IEEE Cat. No.02CH37337). Opt. Soc. America. Part vol.1, 2002, pp vol.1. [27] Barton JS, Skogen EJ, Masanovic ML, DenBaars SP, Coldren LA. Tailorable chirp using integrated Mach-Zehnder modulators with tunable sampled grating distributed Bragg reflector lasers IEEE 18th International Semiconductor Laser Conference. Conference Digest (Cat. No.02CH37390). paper no. TuB3, Garmisch, Germany (Sept. 29- Oct. 3) IEEE. 2002, pp [28] Skogen EJ, Barton JS, Masanovic ML, Getty JT, DenBaars SP, Coldren LA. Use of post-growth control of the quantum-well band edge for optimized widely-tunable laser-x devices IEEE 18th International Semiconductor Laser Conference. Conference Digest (Cat. No.02CH37390). IEEE. 2002, pp [29] Barton JS, Skogen EJ, Masanovic ML, Raring J, Sysak MN, Johansson L, DenBaars SP, Coldren LA. Photonic integrated circuits based on sampledgrating distributed-bragg-reflector lasers. SPIE-Int. Soc. Opt. Eng. xii

14 Proceedings of Spie - the International Society for Optical Engineering, vol.4998, 2003, pp Invited [30] Mašanović M., Skogen EJ, Barton JS, Lal V, Blumenthal DJ, Coldren LA. Demonstration of monolithically-integrated InP widely-tunable laser and SOA- MZI wavelength converter International Conference Indium Phosphide and Related Materials. Conference Proceedings (Cat. No.03CH37413). IEEE. 2003, pp Santa Barbara, California (May 12-16, 2003) [31] Skogen E.J., J.S. Barton, J.W. Raring, L.A. Coldren, S.P. DenBaars, "High Contrast InP/InGaAsP Grating MOCVD Regrowth Using TBA and TBP", Conference Proceedings from ICMOVPE conference, Hawaii [32] Skogen EJ, Barton JS, DenBaars SP, Coldren LA. Wavelength agile photonic integrated circuits using a novel quantum well intermixing process. Optical Fiber Communications Conference. (OFC). Postconference Technical Digest. Postdeadline Papers (IEEE Cat. No.02CH37339). Opt Soc. America. Part vol.2, 2002, pp.fb8-1-3 vol.2. [33] Johansson LA, Barton JS, Coldren L. High-performance EAM-integrated SGDBR laser for WDM microwave photonic applications International Topical Meeting on Microwave Photonics. Technical Digest (IEEE Cat. No.02EX638). IEICE. 2002, pp Tokyo, Japan. [34] Barton J. S., Milan L. Mašanović, Matthew N. Sysak, Erik J. Skogen, John Hutchinson, Daniel J. Blumenthal, Larry A. Coldren, A Novel Monolithically- Integrated Widely-Tunable Wavelength Converter Based on a SGDBR-SOA- MZ Transmitter and Integrated Photo-Detector, Proc. Photonics in Switching 2003, paper no. PS.Mo.A9, pp , Versailles, France (September 2003) [35] Mašanović M.L., Roopesh R. Doshi, Vikrant Lal, Jonathon. S. Barton, Larry A. Coldren, Daniel J. Blumenthal, First Demonstration of both Analog and Digital Wavelength Conversion using a Monolithically-Integrated InP Widely Tunable All-Optical Wavelength Converter (TAO-WC), Proc. Photonics in Switching 2003, paper no. PS.Mo.A10, pp , Versailles, France (September 2003) [36] Mašanović M. L., Vikrant Lal, Jonathon S. Barton, Larry A. Coldren and Daniel J. Blumenthal, Wavelength Conversion Over a 50nm Input and 21nm Output Wavelength Range Using a Monolithically Integrated Tunable All- Optical MMI-MZI (TAOMI) Wavelength Converter, Proc. ECOC-IOOC 2003, paper no. Th1.6.5, pp , Rimini, Italy (September 21-25, 2003) xiii

15 [37] Johansson L.A., J.S. Barton, M.L. Masanovic, J.M. Hutchinson, J.A. Henness, Y.A. Akulova, G.A. Fish and L.A. Coldren, Integrated Optical Components for WDM Optical/Wireless Applications, Proc. Microwave Photonics, pp , Budapest, Hungary (September 2003) [38] Johansson L.A., J.S. Barton and L.A. Coldren, G.A. Fish, Generation of High-Speed Optical Frequency Modulation Using a Phase Modulator OFC [39] Hutchinson J.M., Jonathon S. Barton, Milan L. Mašanović, Matthew N. Sysak, Jeffrey A. Henness, Leif A. Johansson, Larry A. Coldren, Monolithically integrated InP-based tunable wavelength conversion, Presented at Photonics West, San Jose, [40] Mašanović M.L., Vikrant Lal, Leif A. Johansson, Jonathon S. Barton, Larry A. Coldren, Daniel J. Blumenthal, Characterization of the Chirp Properties of a Monolithically-Integrated Widely-Tunable All-Optical Wavelength Converter (TAO-WC) Proc. LEOS 2003, paper no. TuCC3, pp , Tucson, Arizona (October 2003) [41] Hutchinson J.M., Jeffery A. Henness, Leif A. Johansson, Jonathon S. Barton, Milan L. Mašanović, Larry A. Coldren, 2.5 Gb/sec Wavelength Conversion Using Monolithically-Integrated Photodetector and Directly Modulated Widely-Tunable SGDBR Laser, Proc. LEOS 2003, paper no. WU4, pp , Tucson, Arizona (October 2003) [42] Raring J.W., E. J. Skogen, L. A. Johansson, M. N. Sysak, J. S. Barton, M. L. Masanovic, and L. A. Coldren, "Quantum Well Intermixing for Monolithic Integration: A Demonstration of Novel Widely-Tunable 10Gb/s Transmitters and Wavelength Converters," presented at Integrated Photonics Research Conference, San Francisco, California, USA, [43] Masanovic M.L., V. Lal, J. A. Summers, J. S. Barton, E. J. Skogen, L. A. Coldren, and D. J. Blumenthal, "10 Gbps and 2.5 Gbps error-free operation of a monolithically integrated widely-tunable all-optical wavelength converter with independent phase control and output 35nm tuning range," presented at Optical Fiber Communications Conference, OFC, Los Angeles, California, USA, [44] Choquette KD, Barton JS, Geib KM, Allerman AA, Hindi JJ. Short wavelength bottom-emitting VCSELs. SPIE-Int. Soc. Opt. Eng. Proceedings of Spie - the International Society for Optical Engineering, vol.3627, 1999, pp xiv

16 SYMBOLS AND ACRONYMS AFM BCB BER CPW CPS DBR EAM EO FESEM FKE FWHM LEO MOCVD MQW MZ MZM PIC RIE SOA SEM SIMS SGDBR SMSR TW QCSE QWI WDM WG Atomic Force Microscope Benzocyclobutene Bit Error Rate Coplanar Waveguide. Coplanar Stripline Distributed Bragg Reflector Electroabsorption Modulator Electo-optic Field Emission Scanning Electron Microscope Franz-Keldysh effect Full-width Half Maximum Linear Electrooptic effect Metal-organic chemical vapor deposition Multiple Quantum Well Mach-Zehnder Mach-Zehnder Modulator Photonic integrated circuit Reactive Ion Etch Semiconductor Optical Amplifier Scanning Electron Microscope Secondary Ion Mass Spectroscopy Sampled Grating Distributed Bragg Reflector (laser) Side Mode Suppression Ratio Traveling wave Quantum Confined start effect Quantum Well Intermixing Wavelength Division Multiplexing Waveguide SYMBOLS A. amplitude factor, damping factor, r. Angular relaxation resonance frequency. go differential gain i internal quantum efficiency v cavity volume xv

17 ACKNOWLEDGEMENTS I have been lucky to have the opportunity to work with a great set of people in a first-rate facility that truly allows the full process of events to take place from MOCVD growth, to processing, to complicated RF testing of the devices. This work would not be possible without the help of my committee members which span a wide knowledge base giving me the opportunity to grow in the MOCVD lab under Steve DenBaars, processing expertise from Evelyn Hu, RF experience from Nadir Dagli, Opto-electronic systems work from Dan Blumenthal, and optoelectronic design expertise from Larry Coldren. Everyone in the Coldren group has made an impact in one way or another. Special thanks to Erik Skogen for showing me how to grow in MOCVD and process InP based materials, Beck Mason and Greg Fish for helping me in my early career in SGDBR design and testing, Dan Lofgreen for showing me the intricacies of MATLAB. Milan Mašanović and Leif Johansson for helping me out with RF testing. The folks at Agility Communications who helped with some of the regrowths and AR coating runs. Also, thanks to Intel for providing me with the 2003 Intel fellowship. Last but not least, special thanks to Jennifer Hale, who has endeared the long hours and stress. xvi

18 INTRODUCTION Photonic integrated circuits[11] are being pursued to keep up with the everincreasing desire for high optical bandwidth with a small footprint, high power, cost effective packaging and high degree of functionality. With the recent downturn in the telecom market, optical components that will provide value to optical networks at low cost are particularly desired. The cost primarily comes from packaging and the yield improvement of chips in tunable laser fabrication. Because of this, tunable lasers are seen as able to provide value by the integration of lasers with optical amplifiers and modulators on one chip reducing fiber alignments from as high as 5 in the discrete case to 1 in the integrated case. Each fiber alignment adds 3-5dB of insertion loss which demonstrates the obvious desire to integrate. Also integrated devices enable new small form factor 10 Gbit/s transponders in which packaging of conventional bulky discrete parts simply is not possible due to space limitations. 1

19 Mach-Zehnder SOA SGDBR Fig. 1. Integrated SGDBR-SOA-MZ This dissertation explores the integration of interferometric Mach-Zehnder (MZ) modulators, semiconductor optical amplifiers (SOA) and sampled grating (SGDBR) tunable lasers as illustrated in fig. 1. This introduction will attempt to outline the choices of opto-electronic building blocks available to achieve highperformance optical transmission. As will be outlined in the following chapters, these devices are capable of achieving 10 Gbit/s operation with tailorable chirp. Monolithic integration of these devices presents a number of challenges. Optimization of the laser and modulator structures needs to take into account often competing design specifications. Additionally, careful design is necessary to minimize optical reflections[1,227,224] and retain single mode operation. Electrical crosstalk must be reduced to control unwanted chirp and improve the tuning mechanism. Also, thermal crosstalk plays a role in the 2

20 integrated device performance[25]. An in-depth look at these integration challenges is presented in Chapter 1. Desire for increased bandwidth has prompted the change of SONNET standards as shown in table 1. Suppliers are gearing up to provide data rates greater than 40 Gb/s to meet perceived bandwidth constraints as fiber is being implemented in not only long-haul applications, but increasingly, metro networks as well. Table 1. Sonnet Data Rates OC-12 OC-48 OC-192 OC-768 Rate 622 Mb/s 2.5 Gb/s 10 Gb/s 40 Gb/s Cost has also driven the push for devices that enable CWDM and DWDM systems as new fiber tends to be expensive to deploy however replacement of transmitters and receivers considerably less expensive. System designers generally would like single devices that cover the C- Band ( ) and/or L-Band ( ) and S-Band( ). Another major concern for communication applications particularly with higher bit rates (10 and 40 Gbit/s) is the control of a parameter called chirp. Chirp () is defined as the ratio between the change in the real part to the imaginary part of the refractive index. 3

21 n real [1] n imag This parameter strongly affects the maximum transmission distance possible before signal regeneration. Figure 2a&b shows the influence of the chirp parameter [] on the transmission distance for standard non dispersion shifted fiber. Fig 2.A &2B Dispersion Penalty vs transmission distance for OC-192 signal 9.95 Gbit/s for various alpha parameters for standard non-dispersion shifted Corning SMF-28 fiber. Maximum transmission distance for 10 and 40 Gbit/s signal The dispersion penalty is given by[14] : c log [(1 8B L) (8B ) ] [2] 10 L Where is chirp parameter, is group velocity dispersion parameter, B is the bit rate, L is the fiber length. standard Corning SMF-28 fiber =16.45 ps/nm/km at 1555nm This chirp is often split into three major components[521] including: 1.Transient chirp due to the sudden changes of the current in the device 2. Adiabatic chirp due to the induced change in refractive index 4

22 3. Thermal chirp due to the large resistance change and resulting heating/cooling of the active region. The power penalty is also influenced by degradation of the extinction ratio as shown in fig. 3, the signal to noise ratio, and jitter [24]. Fig. 3. Power penalty due to extinction ratio degradation The total transmission distance is often described at the point at which the total power penalty reaches 2dB. With proper chirp management, for a 2dB power penalty at 10Gbit/s one can transmit over 125km in between repeaters. At 40Gbit/s, this distance is dramatically lower just above 7km. This means that chirp management is crucial at high bit rates and for 40 Gbit/s this is more important than high output power as the signal needs to be regenerated frequently without dispersion compensation. 5

23 First we should explore the modulation possibilities of tunable lasers. In principle tunable lasers may be either directly modulated 1 or externally modulated 2 with either a separate modulator or as an integrated device. Next we will compare the performance of different external modulator designs/materials which typically use LiNbO 3, AlGaAs/GaAs, Polymers, or InP in interferometric Mach-Zehnder modulators or in Electro-absorption based modulators InP devices. Chapter 1 explores the integration platform and the influence of design and materials on laser, Semiconductor Optical Amplifier (SOA) and modulator properties and an analysis of the reflections in the device. Others [306,322] have attempted integration of MZ modulators with lasers, however without very careful design the performance can suffer from optical reflections. In fact in some cases these problems have lead people to believe that integration yields diminishing returns and lead to exploration of the copackaging of discrete components[26]. In chapter 2 we will examine the material properties with respect to the growth structure and doping of the device. 1 As discussed in Section 2 2 As discussed in Section 3 6

24 0.1 DIRECT MODULATION The direct modulation of widely-tunable lasers is desirable[15] due to its simplicity and reduced optical absorption path. Fairly large optical bandwidths have been demonstrated for the direct modulation of DFB lasers, narrowly tunable DBR lasers 3, and SGDBR lasers 4. A demonstration of the small-signal modulation response S 21 for an SGDBR is shown in fig. 4. As the bias to the laser is increased the response exhibits less dampening. 6 Intensity Modulation Magnitude (db) mA 60mA 70mA 80mA 100mA 120mA 150mA Frequency (GHz) P[RF]= -20dBm Fig. 4 Directly Modulated SGDBR laser provided by Agility Comm. Small-signal intensity modulation responses of the SGDBR laser for different gain section currents. λ = nm. A maximum in the modulation bandwidth occurs for short cavity lengths taking into account the carrier-density dependent differential gain and the 3 31GHz [27] GHz [28-30] 7

25 photon lifetime. However, in order to form a SGDBR the gain section length generally needs to be fairly long 5 to achieve enough gain for operation. At these lengths one can expect a maximum bandwidth on the order of GHz ignoring parasitics[312]. A higher internal loss leads to a reduced photon lifetime and improved modulation bandwidth. High-speed direct modulation of SGDBR lasers requires high differential gain, low nonlinear gain saturation, high optical confinement, and short and narrow cavities. The small signal frequency response is given by[312]: A H w 1 1 ( ) [3] 2 1 j s 1 j RC r 2 j where A is an amplitude factor, is the damping factor, r is the angular relaxation resonance frequency. The first two terms in (1) correspond to the low-frequency limitations imposed on the modulation response. The first term depends on the carrier transport time ( s ) over the separate carrier confinement region, the second term is determined by the time constant RC of the RC parasitic elements of the chip. These two parameters are often related through the so called K-factor. The K-factor calculated for standard SGDBR lasers is close to 0.6-ns, therefore the maximum achievable 3-dB modulation bandwidth will be 14.8-GHz. The K- Factor is related to the maximum achievable bandwidth by: 5 >450m 8

26 f max 2 2 K [4] The K factor is given by K 4 2 ( p ) vg go [5] where g o is the differential gain, v g is group velocity, is dielectric constant. Published K-factors are usually in the range 0.13 to 2.4ns. The maximum modulation bandwidth is limited by RC parasitics, device heating, and maximum power handling capabilities [312]. Unfortunately, modulating the gain section will modulate the phase and front mirror sections due to current leakage. So without adequate isolation between the sections, this can cause the wavelength to change broadening the linewidth of the laser known as chirp. When adjacent electrodes are not biased the impedance is high leading to little modulation [314] however the isolation resistance decreases with frequency. Fortunately, since the SGDBR consists of multiple sections, modulating the gain section alone will only affect part of the phase within the cavity[330]. Chirp is fairly large and positive for direct modulation of widely tunable lasers as typical linewidth enhancement factors range from 2-9[28] depending on the tuning range of the laser and the placement of the grating with respect to gain spectra[314]. This means the maximum transmission distance before a repeater is necessary is fairly short 9

27 on the order of 10 s of km as shown in fig 1. For small-signal modulation the chirp during direct modulation can be written as: v f m P 2P f 1 f g m 2 [6] where f g is the characteristic frequency of the chirp and is the chirp parameter, also termed the linewidth enhancement factor. A number of approaches have been employed to minimize the chirp of direct modulation. Chirp has been shown to be improved with tensile-strained MQW material giving a smaller linewidth enhancement factor (). Also, prechirping has been employed which involves the simultaneous modulation of the laser and an external modulator at the same time[532]. 0.2 EXTERNAL MODULATION External modulators refer to modulators that operate external to the cavity of the laser. The most common materials for use in external modulators are LiNbO 3, electro-optic polymers and III-V compound semiconductors. LiNbO 3 has been the material of choice due to its high linear electro-optic coefficient and low optical loss(>5db) as shown in table 2. 10

28 Table 2 Chemical Formula LiNbO 3 Crystal Structure Hexagonal Space Group: R3c Point Group: 3m Lattice Parameters, Е a = c = Density, g/cm Thermal Expansion Coefficients, a a = C -1 a c = Transparency Range, mm Propagation loss 0.2dB/cm [4] Index of refraction (n o ) 2.15 (typically) [4] Electro-Optical Coefficients, pmv -1 r 33 = 30.8; r 31 = 8.6; r 22 = 3.4;r 51 = 28 Where the index shift is given by: 3 o 2 r33ez s33e n n [7] 2 r 33 is the linear electro-optic coefficient and s 33 is the quadratic coefficient which is negligible with LiNbO 3. Also, with traveling wave electrodes, and Ti-diffused ridge waveguide optical structures, LiNbO 3 modulators have been demonstrated with bandwidths greater than 100GHz[28] or with drive voltages <1V [21]. Some of the best modulator results to date are shown in table 3. 11

29 Table 3. Comparison of LiNbO 3 and InP Modulators Material Ref V 3dB Frequency Extinction Figure of pi Comments System Response Ratio (db) Merit (GHz) (GHz/V) LiNbO 3 [32] Noguchi Highest speed to date [21] Sugiyama M et. al. OFC Lowest Drive voltage LiNbO 3 suitable for SiGe driver Mitomi, [23] Dolfi D. W <3dB insertion loss z-cut <4dB insertion loss x-cut InP This Work 0.6 * First 40 Gbit/s Integrated widely-tunable laser Takeuchi [31] 40Gbit/s 40Gbit/s EAM-DFB [20] Tsuzuki K mm long MZ not integrated Rolland Note: DC V pi does not reflect RF performance as will be discussed in chapter 5 InP based modulators can be fabricated significantly smaller than conventional LiNbO 3 modulators and with much lower driving voltage requirements (1-3V rather than 6-7V typically). These InP based devices have high efficiencies enabling much shorter devices due to a number of linear and higher order effects [2]. The bias point and modulation amplitude of LiNbO 3 modulators has been shown to be highly dependent on stress, temperature, humidity, and DC voltage[13]. Also, as active devices with quantum wells cannot be made with LiNbO 3, InP is the material of choice for integrated photonic circuits at 1.55µm. 12

30 A more desirable and less costly approach with respect to packaging would be to integrate the modulator with the laser chip. Common modulator designs employ either the Franz-Keldysh effect (FKE)[319] or the Multi-Quantum Well (MQW) Quantum Confined Stark effect (QCSE)[339] with either electroabsorption (EA) modulators or Mach-Zehnder phase modulators. Additionally, electro-absorption modulators can be either designed as lumped circuit components or in more sophisticated traveling wave designs to reach bandwidths exceeding 50GHz[16,17]. Of course, each design has tradeoffs. Key parameters to consider are drive voltage, bandwidth, optical powerhandling capability, bias stability, wavelength sensitivity and insertion loss. Due to the decoupling of the laser functions from the modulation, external modulation leads to a simpler tuning mechanism of the SGDBR, with a simpler layout of the control circuits, and the prevention of chirping and mode-hopping. Additionally, the extinction ratio can be far more desirable for external modulation[214]. External modulators, however introduce loss to the transmitted power due to their coupling and insertion losses. To compensate for these losses optical amplifiers are typically added to the transmitter or the laser must be very high power (>10dBm). This approach leads to increased costs and complexity of the transmitter. Another inconvenience associated with the use of semiconductor modulators is the wavelength sensitivity of their extinction ratio which is present in both Franz-Keldysh(FKE) and Quantum 13

31 Confined Stark-effect(QCSE) based EAMs and Mach Zehnders. The different types of modulators will be outlined in the following sections. 0.3 ELECTRO-ABSORPTION MODULATORS EAMs are attractive due to their short length (75-400µm), ability to integrate with lasers[31], and low drive voltages. Most discrete devices are based on the QCSE that offers superior attenuation to FKE devices. However, QCSE devices although well suited to single wavelength lasers such as DFBs, can be highly wavelength dependent without careful design and not as desirable for widely-tunable laser integration. FKE devices exhibit a lowering of the chirp parameter with higher reverse biases - however achieving negative chirp is difficult as demonstrated in fig. 5 without very high biases and large insertion losses. 5 Chirp factor, alpha nm 1545 nm 1560 nm Reverse bias (V) Fig 5. Alpha parameter vs wavelength for a FKE EA-Modulator. Printed with permission from L. A. Johansson 14

32 EAMs, particularly with high power integrated lasers, will dissipate large amounts of power due to the photocurrent generated in the EAM. The thermal management of integrated devices require careful material design, heatsinking, and thick metal electrodes to dissipate heat[25]. Having said this, quantum well intermixed shallow well EAMs have shown exceptional promise [228] in providing wideband operation with high bandwidth, low drive voltage and negative chirp[34]. EAMs based on QCSE have shown that the chirp parameter may be set negative/positive by adjusting the bias voltage of course increases in the bias results in more optical loss[34]. 0.4 MACH-ZEHNDER MODULATORS Mach-Zehnder modulators are a class of interferometric based modulators that rely on the relative phase shift of one branch with respect to the other to achieve partial to full canceling of the signal at the output as shown in fig

33 1 0.8 Intensity Voltage (V) Fig. 6 Mach-Zehnder configuration and idealized light intensity output at the output. In practice, particularly with the use of InGaAsP waveguides with compositions corresponding to the emission wavelengths close to the operating wavelength, this index change is accompanied by an absorption change as well due to the Kramers-Kronig relations. The argument for using Mach-Zehnder modulators consists of a few reasons. First of all, MZMs offer better power handling than EAMs since less photocurrent is generated in most designs and the optical power is split into two. High optical bandwidths (>10 GHz) are achievable, with fairly low drive voltages V pi <3V. Also, one of the more compelling advantages of Mach- Zehnder modulators over EAMs is the ability to produce negative chirp in a more tailorable fashion. In the EAM case, one can also obtain negative chirp, however, only at higher biases usually resulting in higher optical loss. Pishifted MZ modulators can easily achieve negative chirp with high extinction as 16

34 the device is off without bias. The other advantage of these devices is their higher wavelength independence which is beneficial for WDM systems. Despite the above advantages, Mach Zehnder devices must be fairly long in comparison with EAMs. The length of the modulator is an optimization between the insertion loss and efficiency of the modulator. Most Mach- Zehnder devices provide phase-modulation due to the linear electro-optic effect. The devices shown in this dissertation make use of carrier based effects and electric field based effects that are fairly efficient by designing the waveguide bandgap close to operating wavelength and doping the waveguide. Doing this makes the device more efficient and compact at the expense of bias dependent absorption loss and wavelength dependence as the device is a cross between an EAM and typical MZ modulator. These losses can be mitigated by integration of a laser. Unlike LiNbO 3 modulators, high performance InP based Mach-Zehnder modulators induce considerable loss with reverse bias due to the Franz-Keldysh effect. 0.5 MACH-ZEHDNER BIASING APPROACHES There are a number of different approaches to biasing MZ modulators as shown in table 4. The simplest approach uses a single RF input on one branch. The drive voltage requirements may be reduced by using a push-pull bias scheme. By using both the normal data output and inverting output from a 17

35 modulator driver, the drive voltage requirement for each is cut in half with a parallel push-pull bias scheme. In Chapter 3, lumped modulators are presented using the above modulation approaches and the performance explored with respect to efficiency and speed and the relations to material properties of the passivation dielectric and semiconductor materials. Alternatively, by using a series push-pull configuration on the two modulator sections, the bandwidth can be roughly doubled with the same drive voltage as the single-sided case (see Chap. 4). The ultimate in figure-of-merit is achieved with the dual RF push-pull modulation approach. This uses both RF inputs with 4 electrodes to produce two sets of series push-pull electrodes. This not only doubles the bandwidth with respect to the single sided case but additionally requires half the drive voltage. For an integrated modulator with a high power tunable laser the insertion loss is not as much of a problem as the drive voltage and bandwidth. These devices are also shown in chapter 4. 18

36 Table 4 MZ Bias Techniques Single- Sided Modulation Doublesided Parallel push-pull Doublesided Series Push-pull Dual RF Series Push-pull Bandwidth Voltage Device Length Advantages B V pi L Simple Configuration Easy to get negative chirp harder for 0 chirp B V pi /2 L Utilize both inverting and noninverting outputs from driver 2B V pi L Higher speed, better wavelength sensitivity. Easy to get 0 chirp more difficult to get negative chirp 2B V pi /2 2L Highest figure of merit for integrated devices. Disadvantages Not highest figure of merit. 2 RF inputs more complexity Requires decoupling and bias circuitry. Semiinsulating substrate is necessary. 4 electrodes. Highest complexity good isolation is necessary Chapter 4 examines the higher speed series push-pull devices. Although lumped electrode devices operate at sufficient bandwidths to enable 10 Gbit/s operation, even higher speeds are possible by taking advantage of traveling wave effects[5,6,16,19,23]. These devices exhibit higher optical bandwidth, and with improved transmission line characteristics, lower return losses (S 11 ). Finally Chapter 5 gives some comparison of performance with respect to the chirp and bandwidth of different designs. Additionally, a conclusion and future work session explores the gains that could be made to the device with improved doping, bias schemes, etc. as well as more complex PICs that could be made such as photocurrent-driven wavelength converters. 19

37 0.6 TRAVELING WAVE DEVICES Traveling wave modulators have electrodes that are transmission lines to distribute the capacitance over the whole device length. For lumped electrode devices, low drive voltages require long devices, however large bandwidths require short devices. Fig. 6 Scanning Electron Microscope image of traveling wave electrode device Traveling wave devices such as shown in Fig. 6 are not limited by the RC time constant so they may be made longer to achieve superior extinction and lower drive voltages. The maximum length of the electrodes is limited primarily by the optical and electrical propagation losses. Optical losses can be high if doped waveguides are used 6, and microwave loss can be high if highly doped layers are underneath the contacts. TW structures benefit greatly by decreasing the capacitance per unit length and lowering the microwave losses. 6 where 10cm -1 is typical for a SGDBR structure with 3m wide ridge 20

38 This increases the modulator impedance and the microwave phase velocity. A more extensive look at these designs is presented in Chap. 4. REFERENCES [1] Brosson P, Delansay P. Modeling of the static and dynamic responses of an integrated laser Mach-Zehnder modulator and comparison with an integrated laser EA modulator. Journal of Lightwave Technology, vol.16, no.12, Dec. 1998, pp [2] Mendoza-Alvarez JG, Coldren LA, Alping A, Yan RH, Hausken T, Lee K, Pedrotti K. Analysis of depletion edge translation lightwave modulators. Journal of Lightwave Technology, vol.6, no.6, June 1988, pp [3] Koren U, Koch TL, Presting H, Miller BI. InGaAs/InP multiple quantum well waveguide phase modulator. Applied Physics Letters, vol.50, no.7, 16 Feb. 1987, pp [4] Dagli N. Wide-bandwidth lasers and modulators for RF photonics. IEEE Transactions on Microwave Theory & Techniques, vol.47, no.7, pt.2, July 1999, pp [5] Li GL, Sun CK, Pappert SA, Chen WX, Yu PKL. Ultrahigh-speed travelingwave electroabsorption modulator-design and analysis. IEEE Transactions on Microwave Theory & Techniques, vol.47, no.7, pt.2, July 1999, pp [6] Zhang CZ, Yi-Jen Chiu, Abraham P, Bowers JE. 25 GHz polarizationinsensitive electroabsorption modulators with traveling-wave electrodes. IEEE Photonics Technology Letters, vol.11, no.2, Feb. 1999, pp [7] Walker RG. High-speed III-V semiconductor intensity modulators. IEEE Journal of Quantum Electronics, vol.27, no.3, March 1991, pp [8] Chin MK, Yu PKL, Chang WSC. Optimization of multiple quantum well structures for waveguide electroabsorption modulators. IEEE Journal of Quantum Electronics, vol.27, no.3, March 1991, pp [9] Chin M.K., Comparitive analysis of the performance limits of Franz- Keldysh effect and quantum-confined Stark effect electroabsorption 21

39 waveguide modulators, IEE Proc. Optoelectron. Vol 142. No. 2., April [10] Tipping A.K., G. Parry, P. Claxton, Comparison of the limits in performance of multiple quantum well and Franz-Keldysh InGaAs/InP electroabsorption modulators, IEE. Proc. Vol.136, Pt J, No. 4, August [11] Koch TL, Koren U. Semiconductor photonic integrated circuits. IEEE Journal of Quantum Electronics, vol.27, no.3, March 1991, pp [12] Chin M.K., An Analysis of the Performance of Franz-Keldysh Electroabsorption Waveguide Modulators, IEEE. Photonics Tech. Lett. Vol. 7, No. 3, March [13] Rolland C, Tarof LE, Somani A. Multigigabit networks: the challenge. IEEE Lts, vol.3, no.2, May 1992, pp [14] Agrawal, Govind P., Fiber-Optic Communication Systems Second Ed. Wiley Series in Microwave and Optical Engineering [15] Morthier G, Sarlet G, Baets R, O'Dowd R, Ishii H, Yoshikuni Y. The direct modulation bandwidth of widely tunable DBR laser diodes. [Conference Paper] Conference Digest IEEE 17th International Semiconductor Laser Conference. (Cat. No.00CH37092). IEEE. 2000, pp [16] Akage Y, Kawano K, Oku S, Iga R, Okamoto H, Miyamoto Y, Takeuchi H. Wide bandwidth of over 50 GHz travelling-wave electrode electroabsorption modulator integrated DFB lasers. Electronics Letters, vol.37, no.5, 1 March 2001, pp [17] Kawanishi H, Yamauchi Y, Mineo N, Shibuya Y, Mural H, Yamada K, Wada H. EAM-integrated DFB laser modules with more than 40-GHz bandwidth. IEEE Photonics Technology Letters, vol.13, no.9, Sept. 2001, pp [18] Chin MK. On the figures of merit for electroabsorption waveguide modulators. IEEE Photonics Technology Letters, vol.4, no.7, July 1992, pp

40 [19] Pascher W, Den Besten JH, Caprioli D, Leljtens X, Smit M, Van Dijk R. Modelling and design of a travelling-wave electro-optic modulator on InP. Kluwer Academic Publishers. Optical & Quantum Electronics, vol.35, no.4-5, March-April 2003, pp [20] Tsuzuki K, Ishibashi T, Ito T, Oku S, Shibata Y, Iga R, Kondo Y, Tohmori Y. 40 Gbit/s n-i-n InP Mach-Zehnder modulator with a pi voltage of 2.2 V. Electronics Letters, vol.39, no.20, 2 Oct. 2003, pp [21] Sugiyama M, Doi M, Taniguchi S, Nakazawa T, Onaka H. Driver-less 40 Gb/s LiNbO/sub 3/ modulator with sub-1 V drive voltage. Optical Fiber Communications Conference. (OFC). Postconference Technical Digest. Postdeadline Papers (IEEE Cat. No.02CH37339). Opt Soc. America. Part vol.2, 2002, pp.fb6-1 vol.2. [22] Noguchi K, Miyazawa H, Mitomi O. 75 GHz broadband Ti:LiNbO/sub 3/ optical modulator with ridge structure. Electronics Letters, vol.30, no.12, 9 June 1994, pp [23] Dolfi DW, Ranganath TR. 50 GHz velocity-matched broad wavelength LiNbO/sub 3/ modulator with multimode active section. Electronics Letters, vol.28, no.13, 18 June 1992, pp [24] Dorgeuille F, Devaux F. On the transmission performances and the chirp parameter of a multiple-quantum-well electroabsorption modulator. IEEE Journal of Quantum Electronics, vol.30, no.11, Nov. 1994, pp [25] Kozodoy P, Strand T, Akulova Y, Fish G, Schow C, Ping Koh, Zhixi Bian, Christofferson J, Shakouri A. Thermal effects in monolithically integrated tunable laser transmitters. Twentieth Annual IEEE Semiconductor Thermal Measurement and Management Symposium (IEEE Cat. No.04CH37545). IEEE. 2004, pp [26] Anderson K. Design and manufacturability issues of a co-packaged DFB/MZ module Proceedings. 49th Electronic Components and Technology Conference (Cat. No.99CH36299). IEEE. 1999, pp [27] Kjebon O, Schatz R, Lourdudoss S, Nilsson S, Stalnacke B, Backbom L. 30 GHz direct modulation bandwidth in detuned loaded InGaAsP DBR lasers at 1.55 um wavelength. Electronics Letters, vol.33, no.6, 13 March 1997, pp [28] Mason B., S. L. Lee, M. E. Heimbuch, and L. A. Coldren, Directly modulated sampled grating DBR lasers for long-haul WDM 23

41 communication systems, IEEE Photon. Technol. Lett., vol. 9, pp , March 1997 [29] San-Liang Lee, Mark, E. Heimbuch, Dan A. Tauber, Larry A Coldren, Direct Modulation of widely tunable sampled grating DBR lasers. SPIE Vol 2690/223. pg [30] San-Liang Lee, D. A. Tauber, V. Jayaraman, M.E. Heimbuch, L.A. Coldren, Dynamic Responses of Widely Tunable Sampled Grating DBR Lasers, IEEE. Photonics. Tech. Lett., Vol. 8., No.12, Dec [31] Takeuchi J., K. Tsuzuki, K. Sato, M. Yamamoto, Y. Itaya, A. Sano, M. Yoneyama, T. Otsuji, NRZ Operation at 40 Gb/s of a compact module containing an MQW electroabsorption modulator integrated with a DFB laser IEEE. Phot. Tech. Letts. Vol. 9, No. 5, May [32] Noguchi K, Mitomi O, Miyazawa H. Millimeter-wave Ti:LiNbO 3 optical modulators. Journal of Lightwave Technology, vol.16, no.4, April 1998, pp [33] Raring J. W., E. J. Skogen, L. A. Johansson, M. N. Sysak, J. S. Barton, M. L. Mašanović, L. A. Coldren Demonstration of Widely-Tunable Single- Chip 10 Gb/s Laser-Modulators Using Multiple-Bandgap InGaAsP Quantum-Well Intermixing Photonics Technology Letts. July

42 Chapter 1 Device Integration This dissertation demonstrates an integration platform to enable widely-tunable laser functionality, semiconductor optical amplifier (SOA) technology, as well as efficient and high-speed modulation all on one chip. In this chapter we explore the laser, SOA, and modulator integration considerations. First we look at the SGDBR laser design and its suitability for integration. Next the SOA design is examined and finally the susceptibility of reflections is explored. Fig 1-1 Integration Platforms for 1.55m based photonic integrated circuits [128] Printed with permission from Dr. Erik Skogen 25

43 Modulator-Laser integration is usually performed using the following approaches: 1) Butt-coupling approach allows for independent optimization of device structures however there is a need for selective area growth. Also, more optical loss at the interface. Relatively complicated [125]. 2) QW intermixing low interfacial mode mismatch. Allows for multiple bandgaps due to differing amount of intermixing in the modulator and passive regions. This is a maturing technology for the fabrication of SGDBRs as developed by Skogen et al.[128] 3) Offset QW structure where QWs are removed in passive regions and Franz-Keldysh effect (Bulk effect) utilized in modulator sections. Devices in this dissertation use this approach. See Appendix D. 26

44 1.1 WIDELY-TUNABLE LASER DESIGN Over the last 20 years a few different implementations of monolithic-tunable lasers have been fabricated including Y cavity lasers, grating assisted Table 1.1 Tunable Laser Technologies Technology Tuning range (nm) Fiber Coupled power (dbm) Companie s SMSR RIN Tuning Speed Integrated Amplifiers or modulators DBR [208] 17 Agere Ns-ms Easy SGDBR /11 Agility/ 35-55dB -150dB/Hz Ns-ms Easy Marconi SSGDBR 100 NEL DS-DBR Bookham Easy - -OFC 2003 Y-Branch Syntune 40 Ns-ms Not possible with current implementation OFC GCSR 40 3 ADC/ Altitune Ns ms Not possible MEMS 20.3 Bandwidth Possible but very VCSEL 9 difficult electrically pumped MEMS VCSEL Optically pumped >50nm tuning >7mW achieva ble with 120mW pump Nortel, Cortek, Cielo, Picolight, 1-10ms Possible not easy Wavelength selectable DFB array with MMI NEC/ Fujitsu Temp Tune Possible DFB array with MEMs 20 Santur Tunable External Cavity Diode lasers (ECDL) Iolon, Intel Princeton Optronics, Blue sky Research dB/Hz Not possible 27

45 coupled cavity (GACC), grating coupled Sampled Reflector (GCSR) lasers, Superstructure Sampled Grating Distributed Bragg reflector (SSGDBR)[212,219], DS-DBR, and Sampled Grating Distributed Bragg Reflector (SGDBR) lasers[207,216,217,218], and narrowly tunable DBR lasers[208,209,211] as shown in table 1-1. For the work described here, a SGDBR laser was employed as it has many advantages over other tunable lasers. The simple holographic grating definition gives a wide tuning range without requiring facets for operation leading to highly integrated functionality. Also, there are no moving parts, wafer level testing is possible, and the technology has been demonstrated with high volume manufacturing as only one regrowth is required. Recently devices have been demonstrated commercially demonstrating high output power [129] with fast tuning(<10 ms electronics limited). Ultimately this leads to a high performance device that will help reduce costs. Recent improvements in the SGDBR design were geared toward improving the output power. This was achieved by decreasing the number of mirror periods on the front mirror yet retaining a high side mode suppression ratio (SMSR) (>35dB). Fig. 1. shows a SEM of an SGDBR device. While large tuning ranges have been demonstrated [as high as 72nm] by using a large mirror comb spacing difference this is a tradeoff with power output and flatness over the tuning range. 28

46 Detector Rear Mirror Phase Gain Front Mirror. Figure 1-2 Scanning Electron Micrograph of SGDBR Laser All tunable transmitters in this work use SGDBR devices with the same design parameters as shown in table 2-1. These parameters are fairly conservative as one may obtain higher power with less mirror periods on the front mirror. Table 2-1 SGDBR Device Parameters Front Mirror Burst width 4m (x5) Typical threshold current (ma) 27.5 ma Front Burst Period 68.5m Typical threshold Voltage 1.1V Rear Mirror Burst width 6m (x12) Max Power 20mW Rear Burst Period Gain Length Ridge Width 61.5m 550m 3m 29

47 The tunable laser structure includes two sampled grating mirror sections with parameters as shown in table 2-1. The basic operation is based on the fact that the SGDBR takes advantage of a tuning enhancement by using two mirrors with slightly different mirror periods which results in slightly differing reflection spectra from the Vernier effect [207,220]. The two reflectivity spectra are shown in Fig 1-3. Fig. 1-3 Reflectivity spectra for front and rear mirrors. The device relies on refractive index tuning with current injection into the front and rear mirror sections in which the wavelength can be tuned approximately 38nm with the design above as shown in Fig

48 Fig 1-4 Wavelength as a function of biases on each mirror section Typical IV and LI output plots are shown in Fig 1-5 for a untuned device. The laser threshold for devices ranges from approximately 25-35mA for different tuning biases. As current is injected into the mirrors this adds optical loss in the laser and increases the threshold current. 31

49 25 S020714B_DIE#8_Device# V [V] L [mw] 2.5 L [mw] CW V [V] I [ma] Fig Typical LIV plot for a ridge SGDBR device From Fig 1-6. one can see the supermode boundaries as the mirrors are tuned differentially. Fig 1-6a. Output Power (mw) vs tuning on front and rear mirrors Fig 1-6b corresponding Gain Voltage. Gain section = 100mA SOA = 120mA 32

50 There are optimum mirror alignments in the center of each hexagon region for a given phase section bias (in this case 0mA). The more drastic changes in power correspond to wrap-around points where the device tunes from one side of the tuning range to the other. These devices also demonstrate a low level of spurious reflections in the longitudinal cavity as the devices tune normally. Fig. 1-6b shows the corresponding gain section voltage as the device is tuned. This plot closely relates to the output power plot and can be used in electronic control circuitry to determine the best alignment of the mirrors for a given WDM channel wavelength. 1.2 SEMICONDUCTOR OPTICAL AMPLIFIER INTEGRATION As seen in the previous section, the SGDBR lasers can output as high as 20mW. However, due to the fairly thin waveguide (0.35m) and curved flared output the output mode does not couple well with lensed fiber. A modeconverter would be desirable to increase the output power without introducing more optical noise sources, however the integration adds complexity to the processing and was not pursued in this work. Coupling losses can be as high as 10dB for discrete SOA devices without mode converters[101]. In this device, with only one output, the typical coupling losses range from 4-5dB. 33

51 Additionally zero bias insertion losses in the modulator which range from 4-7 (see Fig. 1-7) as well as bias induced losses make SOA integration desirable. Fig 1-7 Unbiased insertion loss for a MZ Modulator for different wavelengths. Total length 1100m (10.5cm -1 at 1550) Semiconductor optical amplifiers were integrated[114] to improve the overall power output and to even out the wavelength dependent power as lower input optical power results in higher gain. The SOA was chosen to be before the modulator so that the extinction ratio would not be degraded. SOA integration provides for added functionality of the device. Not only higher output power is now possible but it can be used as a variable attenuator since it has high dynamic range. Also it can be readily applied for power leveling the channels across the wavelength range since it is an independent power control. 34

52 Additionally, SOAs have been used successfully as modulators, gates, frequency converters, or detectors[200]. SGDBRs require a sufficient AR coating [>10-3 ][100] for consistent tuning characteristics. Integration of an SOA and modulator requires an improvement in the reflectivity to to prevent device degradation as described in the following section. The most important measures of performance for linear applications are the maximum gain, saturation power and maximum noise figure[106]. In general, SOAs are not competitive with EDFAs because the noise figure is so high. Typical relative noise figures of EDFAs to SOAs at 1.55um (3.1 vs 5.2). [101] In the integrated device the SGDBR laser is a CW source in which the SOAs are biased to saturation so the noise characteristics do not suffer much. Devices 4-9 have 400µm and 500µm long SOAs before the MZ modulator. The gain from these devices is shown in Fig Fig 1-7 Gain at 1554um for a 400m and 500m long, 3m wide SOA at T = 16C 35

53 For high power from the SGDBR (>10dBm), this means the SOA only contributes approximately 6-7dB of gain at high SOA biases. This effectively cancels out the insertion losses of the modulator. 1.3 DUAL SOAS A number of different approaches have been explored to improve the saturation power and gain of SOA devices. Gradual tapering of the waveguide has been explored[103] as well as using multiple waveguides coupled using MMIs to improve the gain saturation[105]. Fig. 1-8 DUAL SOA Design Layout Saturation power is generally limited by the small cross sectional active area required for single-mode operation. By using N waveguides in parallel the saturation power is improved by 10log (N) db with respect to a single waveguide device without degrading the noise figure. The devices presented 36

54 in this dissertation have Dual SOAs ranging from 350m each to 575m each. The SOA region was tapered out to 3.5m wide ridges in the SOA region from 2.5m wide ridges in the modulator region. The gain of the Dual SOAs was measured with one SOA reverse biased then compared the gain with one biased to 150mA and the other SOA bias varied. Clearly thermal crosstalk affects the SOAs so that the gain is reduced 1-2dB due to heating from the adjacent SOA that is 16m away. Overall the saturation power is improved and the Dual SOA approach is beneficial. These heating effects are beneficial to the efficiency of the modulator as shown in more detail in chapter 5. 37

55 Fig 1-9a&b Output power for Dual SOA with 575µm electrodes 13.5um transparency current second SOA reverse biased 38

56 Fig 1-10a&b Output power and gain with SOA #2 biased to 150mA [575µm SOA] 39

57 1.4 OPTICAL FEEDBACK AND REFLECTION Highly integrated devices require the minimization of optical feedback in order to reduce unwanted chirp, lasing of the SOA, gain ripple, higher noise figure, and inter-modulation distortion. Reflections lower the gain bandwidth and output saturation power. The design needs to consider reflections due to the facets and active/passive interfaces[127], waveguide design, tapers, and multimode interference (MMI) devices within the Mach-Zehnder modulator. Reflections at the facets are minimized by flaring and angling the waveguide at the outputs and backside absorber facets. This reduces the requirements of the AR coating by an order of magnitude often providing continuous tuning even without an AR coating. In order to operate properly, SOA based devices require sufficiently low Anti reflection (AR) coating reflectivity - generally accepted as < A multi-layer AR coating is used to achieve greater than 10-4 reflectivity back into the optical cavity. Active/Passive interfaces are angled to minimize reflections due to the index discontinuity. Additionally, the MMI lengths are optimized for minimum reflections and are tapered so that reflections are not coupled back into the laser cavity mostly important in the off state. Due to the nature of the multimode interference splitter, changes in the widths, growth thicknesses etc. will influence the light propagation. Biasing of sections on top of the MMIs will allow for tuning the power splitting ratio[608] important in MZ design for the 40

58 chirp and extinction ratio tuning. Note however, that the imaging length is modified potentially leading to large reflections when not optimally biased. A safer approach is to adjust the different branch gain or loss in the two branches. Finally, the waveguide design is weakly-guided throughout the structure. Also, as demonstrated in [124] shallow ridge technology in the MZ lowers the reflections in the device considerably in comparison with previous deep ridge etched MZs[336, 337,344]. It has been shown that when parasitic reflections are generated in the MZ the extinction ratio is degraded when the chirp is best[306]. The optical feedback in the device can be explored using a few different approaches. Optical low coherence reflectometry (OLCR) is a good approach to measure the reflections at each interface in the structure[124]. Also, one can look at the tuning modemap as a function of biases on the rear detector and front detector biases. An example of this is shown for a device with two 575µm long SOAs in the branches of the Mach-Zehnder in Fig. 1-11a with poor AR coating. 41

59 Fig 1-11a Device #10 Comparison of Dual SOA with significant optical feedback with SOAs biased to 80mA. Fig 1-11b demonstrates reflection reduction by reverse biasing front and rear detectors -4V Reverse biasing the active rear absorber and the passive front detector 7 reduces the feedback from the facets thereby improving the tuning map profile as shown in fig 1-11b. Fig 1-12 Front detector reverse biased to absorb off-state light 7 As shown in Fig

60 Another indication of the optical feedback in the device is the degradation of the linewidth. 1.5 LINEWIDTH MEASUREMENTS Linewidth arises from coupling between variations of phase and intensity. The linewidth for a conventional Fabry-perot, DFB, or SGDBR is given by: where the threshold gain is given by: 2 2 vg hvgthnsp m (1 ) [1.1] 8P gth 1 i m [1.2] and the mirror loss: 1 1 m ln [1.3] 2L R ( ) R ( ) 1 2 P is optical power, is the linewidth enhancement factor, L is the cavity length, and R 1 and R 2 are the reflection coefficients for the mirrors. v g is the group velocity, n sp spontaneous factor, hv is the optical energy In general, linewidth decreases with increased laser power and increases with tuning of the mirror and phase sections due to misalignment of the mirrors[29]. 43

61 Accurate measurement of linewidth in multi-section devices requires careful control. Battery powered sources are preferable to source less noise and also biasing that shorts high frequency noise to GND are required particularly when biasing sections highly sensitive to noise such as the phase section. Also, the setup requires sufficient optical isolation(>50db). In order to explore the susceptibility of the device to optical reflections, a linewidth measurement setup that was used as shown in Fig SGDBR-MZ AMP LiNbO 3 2GHz Fig Linewidth measurement setup Delayed Self-Heterodyne Method The output light is inserted in a LiNbO 3 MZ modulator in order to frequency shift the linewidth away from 0Hz reducing the noise in the measurement. This shifted Full-Width Half Max (FWHM) linewidth is double the true linewidth of the laser assuming that the lineshape is near Lorentzian in shape due to the heterodyning of the two signals. Then the signal goes into another interferometer in which one arm has a large delay (1km) which must be longer than the reciprocal of the linewidth in order to have a small measurement error. This means a system with 3.5s delay will be able to make 44

62 linewidth measurements as low as 225kHz. With a 25s delay one can measure to 30kHz. The linewidth of a commercial SGDBR-SOA was compared with a device with a 400m SOA before the MZ (Device #8) and a device with Dual 350m long SOAs in Fig Each was compared at the same bias point of 150mA on the Gain section and SOAs. Fig 1-14 Linewidth measurement comparing Commercial SGDBR-SOA device to Device 8 (single 400µm SOA) and Device 1 (Dual SOA device with 350m long SOAs) As can be seen, there was very little difference between the linewidth of the three devices. This actually is a coincidence as the linewidth varied 45

63 approximately from 4-9 for different Gain and SOA biases. However, this demonstrates that the external cavity (MZ) does not adversely affect the laser s linewidth with optical feedback. Although this demonstrates the lack of optical reflections in the device, in order to look at the quality of signal under transmission, it is helpful to look at the linewidth as a function of frequency. The noise is not constant with frequency for the laser 8 and consists of non-white and white components. A relaxation oscillation induced noise resonance is found between 1-6GHz depending on the output power. Also, phase noise below 500MHz is fairly large due to carrier fluctuations in the tuning sections from noisy current sources. This low frequency non-white phase noise has been shown to not affect the transmission performance of SGDBR lasers after fiber[126]. As these devices are most likely to be used in a 10Gbit/s system where the phase noise to intensity noise conversion efficiency is greatest at 5GHz, the noise at higher frequencies is more important than at low frequencies[506]. The integration time affects the measurement and ideally the linewidth is measured as a function of frequency as shown in [126]. The instantaneous linewidth is much narrower than shown in this measurement as the line tends to drift over time. This is why the seemingly large linewidths shown here are not detrimental to practical applications as the white noise at higher frequencies are the most important to the transmission properties of the laser. Nonetheless, for a typical 8 See RIN section

64 SGDBR measured using the coherent optical frequency discriminator technique[131], linewidths are usually measured between 1-2 MHz[126]. 1.6 RELATIVE INTENSITY NOISE The relative intensity noise (RIN) is a measure of the quality of laser diodes for broadband digital or analog systems. It is defined as P 2 RIN db / Hz. [1-4] 2 P The numerator is the mean square optical intensity fluctuation at a specified frequency and P is the average output power. Since the ratio of the optical powers squared is equivalent to the ratio of the detected electrical powers, this equation can be written as RIN N electrical db / Hz P ( electrical) [1-5] avg N electrical is the power-spectral density of the photocurrent at a specified frequency and P avg is the average power of the photocurrent. It can also be expressed as: RIN 2 N 4 R 2 1/ 2 2h 2q 16 ( ) ST H [1-6] P P 0 fiber where H(w) is the modulation transfer function of the laser, is the electrical frequency, R is the relaxation resonance frequency, is the damping factor, () ST is the modified Schawlow- Townes linewidth, N is the differential carrier lifetime,, h is the optical energy, P 0 is the optical power output from the laser, is the photodiode responsivity, and P fiber is the optical power coupled into the fiber. 47

65 The RIN was measured for a device with a single 400µm SOA and Dual SOAs (350µm) Fig 1-15a RIN for single 400m SOA Gain section bias varied [SOA 120mA nm] 48

66 Fig 1-15b RIN for Dual 350m SOA device [SOA1&2 100mA] As can be seen in Fig 1-15a and 1-15b, the RIN measured for devices with single and dual SOAs are very similar. In both cases the RIN exceeds - 140dB/Hz, a specification desired for commercial SGDBR lasers. The peak of the noise spectrum indicates a parasitic free means of measuring the modulation bandwidth of the directly modulated SGDBR. The actual bandwidth is limited by carrier transport[113]. 49

67 REFERENCES [100] Stubkjaer KE, Mikkelsen B, Durhuus T, Storkfelt N, Joergensen C, Jepsen K, Nielsen TN, Gliese U. Recent advances in semiconductor optical amplifiers and their applications. Fourth International Conference on Indium Phosphide and Related Materials (Cat. No.92CH3104-7). IEEE. 1992, pp [101] Eliseev PG, Vu Van Luc. Semiconductor optical amplifiers: multifunctional possibilities, photoresponse and phase shift properties. Pure & Applied Optics, vol.4, no.4, July 1995, pp [102] Joo-Heon Ahn, Kwang Ryong Oh, Jeong Soo Kim, Seung Won Lee, Hong Man Kim, Kwang Eui Pyun, Hyung Moo Park. Uniform and high coupling efficiency between InGaAsP-InP buried heterostructure optical amplifier and monolithically butt-coupled waveguide using reactive ion etching. IEEE Photonics Technology Letters, vol.8, no.2, Feb. 1996, pp [103] Donnelly JP, Walpole JN, Betts GE, Groves SH, Woodhouse JD, O'Donnell FJ, Missaggia LJ, Bailey RJ, Napoleone A. High-power 1.3- mu m InGaAsP-InP amplifiers with tapered gain regions. IEEE Photonics Technology Letters, vol.8, no.11, Nov. 1996, pp [104] Gilner L. Analysis of input power dynamic ranges in two types of expanded semiconductor optical amplifier gate switch arrays. IEEE Photonics Technology Letters, vol.8, no.4, April 1996, pp [105] Dagens B, Emery JY, Janz C. Multiwaveguide SOA for increased saturation power without noise penalty. Electronics Letters, vol.35, no.6, 18 March 1999, pp [106] Liu T, Obermann K, Petermann K, Girardin F, Guekos G. Effect of saturation caused by amplified spontaneous emission on semiconductor optical amplifier performance. Electronics Letters, vol.33, no.24, 20 Nov. 1997, pp [107] Jayaraman V, Chuang Z-M, Coldren LA. Theory, design, and performance of extended tuning range semiconductor lasers with sampled 50

68 gratings. IEEE Journal of Quantum Electronics, vol.29, no.6, June 1993, pp [108] Delorme F, Alibert G, Boulet P, Grosmaire S, Slempkes S, Ougazzaden A. High reliability of high-power and widely tunable mu m distributed Bragg reflector lasers for WDM applications. IEEE. IEEE Journal of Selected Topics in Quantum Electronics, vol.3, no.2, April 1997, pp [109] Delprat D, Ramdane A, Ougazzaden A, Carre M. Very simple approach for high performance tunable laser realisation. Electronics Letters, vol.32, no.22, 24 Oct. 1996, pp [110] Lin MS, Piccirilli AB, Twu Y, Dutta NK. Fabrication and gain measurements for buried facet optical amplifier. Electronics Letters, vol.25, no.20, 28 Sept. 1989, pp [111] Delorme F, Grosmaire S, Gloukhian A, Ougazzaden A. High power operation of widely tunable 1.55 mu m distributed Bragg reflector laser. Electronics Letters, vol.33, no.3, 30 Jan. 1997, pp [112] Sarlet G, Morthier G, Baets R. Control of widely tunable SSG-DBR lasers for dense wavelength division multiplexing. Journal of Lightwave Technology, vol.18, no.8, Aug. 2000, pp [113] Nagarajan R, Ishikawa M, Fukushima T, Geels RS, Bowers JE. High speed quantum-well lasers and carrier transport effects. IEEE Journal of Quantum Electronics, vol.28, no.10, Oct. 1992, pp [114] San-Liang Lee, Heimbuch ME, Cohen DA, Coldren LA, DenBaars SP. Integration of semiconductor laser amplifiers with sampled grating tunable lasers for WDM applications. IEEE. IEEE Journal of Selected Topics in Quantum Electronics, vol.3, no.2, April 1997, pp USA. [115] Coldren L.A., S.W. Corzine, Diode Lasers and Photonic Integrated Circuits, John Wiley & Sons Inc., 1995, pp [116] Jayaraman V, Heimbuch ME, Coldren LA, DenBaars SP. Widely tunable continuous-wave InGaAsP/InP sampled grating lasers. Electronics Letters, vol.30, no.18, 1 Sept. 1994, pp [117] Delorme F, Alibert G, Ougier C, Slempkes S, Nakajima H. Sampledgrating DBR lasers with 181 wavelengths over 44 nm and optimized power variation for WDM applications. OFC '98. Optical Fiber 51

69 Communication Conference and Exhibit. Technical Digest. Conference Edition OSA Technical Digest Series Vol.2 (IEEE Cat. No.98CH36177). Opt. Soc. America. 1998, pp [118] Sarlet G, Morthier G, Baets R. Wavelength and mode stabilization of widely tunable SG-DBR and SSG-DBR lasers. IEEE Photonics Technology Letters, vol.11, no.11, Nov. 1999, pp [119] Ishii H, Tanobe H, Kano F, Tohmori Y, Kondo Y, Yoshikuni Y. Quasicontinuous wavelength tuning in super-structure-grating (SSG) DBR lasers. IEEE Journal of Quantum Electronics, vol.32, no.3, March 1996, pp [120] Mason B, Barton J, Fish GA, Coldren LA, Denbaars SP. Design of sampled grating DBR lasers with integrated semiconductor optical amplifiers. IEEE Photonics Technology Letters, vol.12, no.7, July 2000, pp [121] San-Liang Lee, Tauber DA, Jayaraman V, Heimbuch ME, Coldren LA, Bowers JE. Dynamic responses of widely tunable sampled grating DBR lasers. IEEE Photonics Technology Letters, vol.8, no.12, Dec. 1996, pp [122] Jayaraman V, Mathur A, Coldren LA, Dapkus PD. Extended tuning range in sampled grating DBR lasers. IEEE Photonics Technology Letters, vol.5, no.5, May 1993, pp [123] Jayaraman V, Cohen DA, Coldren LA. Demonstration of broadband tunability in a semiconductor laser using sampled gratings. Applied Physics Letters, vol.60, no.19, 11 May 1992, pp USA. [124] Lovisa S, Bouche N, Helmers H, Heymes Y, Brillouet F, Gottesman Y, Rao K. Integrated laser Mach-Zehnder modulator on indium phosphide free of modulated-feedback. IEEE Photonics Technology Letters, vol.13, no.12, Dec. 2001, pp [125] Putz N, Adams DM, Rolland C, Moore R, Mallard R. Fabrication of an InP/GaInAsP based integrated gain-coupled DFB laser/m-z phase modulator for 10 Gb/sec fiber optic transmission. IPRM Eighth International Conference on Indium Phosphide and Related Materials (Cat. No.96CH35930). IEEE. 1996, pp [126] Nakagawa S, Fish G, Dahl A, Koh P, Schow C, Mack M, Wang L, Yu R. Phase noise of widely-tunable SG-DBR laser. Optical Fiber 52

70 Communications Conference (OFC). (Trends in Optics and Photonics Series Vol.86) Technical Digest (IEEE Cat. No.03CH37403). Opt. Soc. America. Part vol.2, 2003, pp vol.2. [127] Ackerman DA, Shtengel GE, Morton PA, Zhang LM, Johnson JE, Bethea CG, Ketelsen LJ-P. Identifying sources of residual reflections within integrated electroabsorption modulated laser cavities. Conference on Optical Fiber Communications. Technical Digest. Postconference Edition OSA Technical Digest Series. Vol.6 (IEEE Cat. No.97CH36049). Opt. Soc. America. 1997, pp [128] Skogen E.J. Quantum Well Intermixing for Wavelength-Agile Photonic Integrated Circuits UCSB Dissertation June 2003 [129] Simes R., G. A. Fish, P. Abraham, Y. A. Akulova, C. W. Coldren, M. Focht, E. M. Hall, M. C. Larson, H. Marchand, P. Kozodoy, A. Dahl, P.C. Koh, T. Strand, InP chip scale integration platform for high-perfomance tunable lasers, SPIE Photonics West, [130] Marshall WK, Crosignani B, Yariv A. Laser phase noise to intensity noise conversion by lowest-order group-velocity dispersion in optical fiber: exact theory. Optics Letters, vol.25, no.3, 1 Feb. 2000, pp [131] D. Derickson, Fiber Optics Test & Measurement, Upper Saddle River, NJ: Prentice Hall PTR, 1998, p

71 Chapter 2 MOCVD GROWTH AND FABRICATION In this chapter, we will look at some of the epitaxial layer growth considerations regarding the Metal Organic Chemical Vapor Deposition (MOCVD) growth of tunable transmitters. Growth consists of a base epitaxial structure which undergoes processing steps to etch off quantum wells and define gratings in certain areas. Then a regrowth is performed to grow an upper InP/InGaAs cladding over the whole structure. An optimized doping profile is critical for high performance in both the laser and modulator sections. The basis for characterization of the material is discussed with the use of x-ray, PL, and SIMs measurements. The fabrication steps are also outlined in Section 2.8. The MOCVD technique is used to grow high quality single crystal epitaxial films on substrates using metal-organic precursors that are transported to a heated substrate (615C 650C) with a carrier gas (H 2 ). The work here was done using a Thomas-Swan 2 horizontal flow rotating disk reactor which has very good deposition uniformity as shown in fig

72 Average 1540nm Fig 2-1 Micro-Photo-luminscence(PL) Intensity and Peak wavelength uniformity across base structure wafer S020215C MOCVD growth of In 1-x Ga x As 1-y P y quaternary is very important for fabrication of opto-electronic devices such as lasers, modulators, and detectors all fundamental in optical fiber communication systems. Typically this involves the highly toxic Arsine (ArH 3 ) and Phosphine (PH 3 ) precursors. At UCSB, tertiarybutylarsine (TBAs), and tertiarybutylphosphine (TBP) Group V precursors are used as they are considerably safer, and have been shown to be superior with respect to enabling very low III/V ratios and low growth temperatures as well as possess similar purity. For this work, the Group III precursors are trimethyindium(tmin) for Indium, trimethylgallium (TMGa) for Gallium. The material is doped using both disilane gas for silicon n-type layers, and DiethylZinc for Zinc p-type layers. 55

73 The basic reactions describing the growth of InP are: In( CH ) ( v) TBP( v) InP( s) CH TB [2.x] where during InGaAsP growth, the group III precursors decompose as follows: Ga( CH CH [2.x] 3 ) 3( v) Ga( CH 3) 2 3 Ga( CH CH [2.x] 3 ) 2 ( v) Ga( CH 3 ) 3 In( CH CH [2.x] 3 ) 3 ( v) In( CH 3 ) 2 3 In( CH In CH [2.x] 3 )( v) 3 Since the TMIn source is a solid source, two bubblers are placed in series to insure a constant source. For the liquid gallium source (TMGa), a dilution line is utilized to achieve the desired range of material concentrations. 2.1 SEMICONDUCTOR EPITAXIAL STRUCTURE Although SGDBR lasers have been fabricated using quantum well intermixing(qwi)[33] and butt-joint regrowth techniques, the offset Multiple Quantum Well (OMQW) approach has been a standard approach for the fabrication of photonic integrated circuits[ ]. This is due to the ease of removal of QWs in passive regions without significant index discontinuity or complicated regrowth processes. Next we will look at the quantum well design and epitaxial layer structures based on both conducting sulfur-doped and semiinsulating Fe-doped wafers. 56

74 2.2 QUANTUM WELL DESIGN The quantum wells should be designed to efficiently contain the carriers and not impede the transport across the structure. For well designed barriers, the thermionic emission time is much larger than the diffusion and capture times[312]. Thermionic emission and tunneling are competing processes and the faster one will dominate. For barriers less than 5nm, tunneling tends to dominate the carrier transport across the barrier. However, for wider barriers (8nm in this case), hole transport is associated with thermionic emission while electron transport is done mainly by tunneling. For sufficiently wide barriers both electrons and holes are dominated by thermionic emission[312]. The MQW structure relies on the barriers to be thick enough to provide 2-D carrier confinement in the quantum wells and prevent coupling of the quantum well states which leads to a broadening of the quantized energy levels[312]. It has been shown[225], that significant improvements in both differential gain and threshold current density can be achieved by compressively straining the quantum wells at least up to approximately 1%. In this work a MQW stack of 7 Quantum wells with 8nm barriers and 6.5nm wells were used as shown in fig 2-2. Without any strain compensation in the barriers and waveguide, one is limited to approximately 6 wells 6.5nm wide with 8.0nm barriers before reaching the critical thickness. In the current design, the quantum wells are 57

75 grown with a constant x composition for the barriers and quantum wells to allow fast switching between mass flow controller during the growth of the QWs and the barriers. Constant x growth also has the advantage that since the growth rate is governed by the group III flow rate, both the barrier and well grow at the same rate. In this structure, the waveguide and barrier are grown slightly tensile to compensate for the highly strained wells allowing for more wells to be grown before reaching the strain critical thickness. The target compositions for the structure are shown in table 2-1. Table 2-1 In 1-x Ga x As 1-y P y Compositions Material In[%] As[%] Pl peak Eg hh Perpendicular Strain Barrier (-0.3%) QW N/A N/A (1%) SCH Waveguide 1.4Q CONDUCTING SUBSTRATE BASE STRUCTURES Base structures for SGDBRs have been grown for some years now and the best growth conditions refined considerably. These growths use an indium pure component flow of 0.4sccm, with changes in the gallium flows to achieve the desired quaternary compositions as governed by the segregation coefficient K seg. The growth was held at a constant pressure of 350 Torr. However, the temperature of the growth is changed for the InP(615C) and 58

76 InGaAsP layers (650C) for best material quality. This is achieved using a three zone infrared lamp system with a center zone at 650C and adjoining regions at 630C to improve the temperature uniformity. As the passive regions have the QWs etched off, we have two different layer structures after regrowth. Fig. 2-2 shows layer structures in an active and passive section simultaneously. A 10nm un-intentionally doped (uid)-inp stop etch layer is grown on top of the waveguide in order to facilitate the QW wetetch. The base structure is grown with a Zn setback from the quantum wells that takes into account the subsequent regrowth diffusion(80nm). This also means that the regrowth on the active regions is on a doped material instead of a grown junction. Fig 2-2 Sulfur doped substrate epitaxial layer structure 59

77 2.4 GROWTH CHARACTERIZATION The different layers are characterized by determining the bandgap energy from both room temperature photoluminscence data and the lattice parameter a o from double crystal x-ray diffraction data (XRD)[102]. According to Vegard s law, we can relate the binary lattice constants to the unstrained In 1-x Ga x As 1-y P y quaternary lattice constant as follows[102]: a o = x 0.190y xy. [2-1] Fig. 2-3 Typical x-ray rocking curve for base structure S010406C 60

78 The lattice mismatch between layers can be calculated from x-ray by examining the difference between the substrate peak and the quaternary peaks as described by [204]: a a s tan cot [2.2] a a s cot cot [2.3] where corresponds to the angle between the (hkl) plane and the (001) reflection plane and is the bragg angle for the hkl reflection. This lattice mismatch is given by[205]: a a C C11 2C a o a o [2.4] where a o is the lattice constant, C 11 and C 12 are elastic stiffness constants. As many of the layers are strained in the multiquantum well structure, we must look also at the photoluminescence data as in fig

79 Fig 2-4 One dimensional Photoluminescence plot showing both the QW peak and WG peak An empirically modified version of Vegard s law expresses the band gap as a function of In 1-x Ga x As 1-y P y quaternary composition [102]: E g (295K) x y 0.33xy ( y) x(1 x) ev [2.5] ( x) y(1 y) 0.05 xy(1 x)(1 y) The layer compositions and thickness can be fit using a model using BEDE software. 62

80 2.5 SEMI-INSULATING SUBSTRATE GROWTH By growing a similar base structure to that discussed in the previous section on a Fe doped semi-insulating(si) substrate as seen in Fig. 2-5, this enables some of the series push-pull electrode structures discussed in chap µm 1E19 cm -3 Zn-pInGaAs 1.8µm 1E18 cm -3 Zn-p cladding InP 0.05µm setback uid InP 25nm 1.226Q uid-sch 7 uid 6.5nm QWs 8.0nm Barriers 10 nm uid-inp stop-etch layer 0.32µm 1.4Q InGaAsP waveguide 1.8µm 1e18 cm -3 Si InP Buffer 0.05µm n-ingaas contact layer 0.5µm 1e18 cm-3 Si InP n-buffer 100µm Fe-doped semi-i -nsulating substrate Fig. 2-5 Semi-insulating substrate based epitaxial layer structure Using a semi-insulating (SI) substrate lowers the RF dielectric losses and allows isolation between different n-contacts on the device. The n-contact layer is typically either a InGaAsP quaternary layer with emission bandgap in the Q range if it is required to be close to the waveguide particularly important if He implanting is required to isolate n-doped regions. For this work, an InGaAs layer was used for the best contact resistivity and placed 1.8m away from the waveguide so that there is only a very minimal overlap between 63

81 the optical mode and this highly absorptive layer. The use of an InGaAs layer for the n-contact has been shown to improve the extinction ratio since substrate modes are absorbed to a greater degree[220]. 2.6 REGROWTH The regrowth quality over gratings is highly dependent on surface preparation, grating etch depth 9 and regrowth conditions. Low damage RIE (at low power 22Watts) and proper growth conditions for grating regrowth are very important to reduce defects at the growth interface[228]. Gratings are defined using a holographic setup 10 and etched in sampled regions in each mirror section using a Reactive Ion Etch (RIE) system as shown in fig 2-8a&b. Fig. 2-8a Atomic Force Microscope (AFM) image of grating burst before grating regrowth Fig 2-8b Field Emission Scanning Electron Microscope (FESEM) image of dry-etched gratings before regrowth 9 Typically 75nm 10 See Beck Mason Dissertation [224] 64

82 Previously fabricated gratings used a relatively high-power etch (500V) that reduced the photoluminescence intensity by a factor of 100. By reducing the RIE power and using a 1min sulfuric acid dip to remove residual organics, this photoluminescence intensity can be improved to approximately 34% when compared with non-grating regions[228]. Grating regrowth quality is superior if the growth in initiated immediately as the growth temperature is reached (615C) so that the gratings do not decompose. The growth cycle was modified similar to the process shown in [212], to remove the bake step and reduce the initial growth rate (1/16 growth rate) as well as use both a phosphorus (1Torr) and arsenic (0.07Torr) overpressure during the ramp-up to growth that appears to reduce the exposed InGaAsP desorption and increase the oxide/h 2 O desorption resulting in improved growth over quaternary material surfaces. Fig 2-9a Regrowth over gratings Poor surface preparation Fig 2-9b Regrowth over gratings with good surface preparation and TBA with heatup A more extensive look at the photoluminescence data under different regrowth conditions is published in [228]. When done successfully, it is difficult to see in 65

83 an optical microscope where the grating bursts are after the regrowth is performed as shown in Fig Zn DOPING OF InP and InGaAsP The device doping profile needs to take into consideration the requirements of the laser and that of the modulator section. The laser requires high Zn doping in the cladding to improve the injection efficiency and contact resistivity, however, this needs to taper off near the waveguide to prevent excessive free carrier losses. Doping the waveguide with silicon helps the transport time in the laser regions. However, control of the doping profile is critical particularly in the modulator region where it governs both the high speed and efficiency performance 11. The doping profile used in the devices shown in this dissertation is shown in section 3.1. The capacitance of the structure is improved with a large intrinsic region and the bandwidth will be less influenced by the reverse bias, however, this is a tradeoff with transport properties through the structure and the efficiency as there is less electric field overlap with the optical mode. As Zn is used as the p-type dopant in this work, understanding of the diffusion mechanisms is important. 11 See Chapter 3 for details 66

84 Due to the high diffusion constant of Zinc, control of the p-doping profile is particularly crucial. Not only does excessive P-doping of the waveguide add considerable absorption loss this also adversely affects the modulator efficiency. It is thought that the following two mechanisms account for the interactions of Zn diffusion. Frank-Turnbull mechanism m Zni VIn Zns ( m 1) h [2.6] An interstitial Zn finds it s way to an Indium vacancy site and forms a substitutional Zn atom and a hole. M is the charge state (+2). Kick-out mechanism A charged interstitial Zn atom kicks out an Indium atom from its lattice site to form a substitutional acceptor via the reaction m Zni VIn Zns I in ( m 1) h [2.7] 67

85 For this reason the doping incorporation is highly dependent on the III/V ratio and phosphorus overpressure essentially governing the number of vacancies in the material. Combination of the interstitial Zn with phosphorus vacancies often yields neutral complexes for this reason we have high Zn incorporation but it is not all electrically active. Zinc diffusion in InP thought to be dominated by highly mobile Zn interstitals that are in equilibrium with substitutional Zn. The diffusivity of the interstitials is on the order of 1 million times larger for the interstitials. However, there are far more substitutional Zn impurities in comparison with interstitials. The incorporation and activation of p-type dopant Zn are elevated on the <311>B and <110> planes. The n-type dopants are suppressed on these planes[101]. The Zn doping profile is shown for two different device runs using a Secondary Ion Mass Spectroscopy (SIMS) technique. Run #1 and run #2 have vastly different laser and modulator properties. The first case (fig.1.1a) has very high bandwidth due to the large intrinsic region but poor tuning characteristics due to added resistance in the tuning regions that heat considerably when current is injected. The second case (fig 1.1B) has Zn doping extending into the waveguide that although gives better carrier transport, has higher free carrier losses. 68

86 Figure 2.6A&B Secondary Ion Mass Spectroscopy for run#1 and run #2 SIMS work by Charles & Evans As shown in fig. 2.6b, if the waveguide is doped p-type by Zn diffusion during the regrowth we find that the effective area of the PIN junction is enhanced considerably. During growth, since Zn readily diffuses and the quaternary is more easily doped than InP, Zn tends to segregate at the hetro-interface between the InP and InGaAsP waveguide. This highly conductive layer acts as the top side of the capacitor raising the capacitance as high as 5-10x for the device as illustrated in fig 2-7a. This problem can be mitigated by dry-etching off this highly conductive layer 100nm in the InGaAsP waveguide of the modulator sections to improve the bandwidth as shown in fig 2-7b. 69

87 Fig. 2-7a Zn doped InGaAsP waveguide layer after dry/wet etch of shallow-ridge waveguide structure Fig 2-7b Passivation etch removes Zn-doped region in the modulator sections This etch is done by RIE at low power to minimize roughness of the waveguide and the scattering losses which would result. Etching the waveguide makes the ridge more confining, however can make the device more susceptible to reflections. 2.8 TRANSMITTER FABRICATION Fabrication of the tunable transmitters involves the steps as shown in table 2-2. More detailed procedures are shown in Appendix C-Process. Integration of a semiconductor optical amplifier does not add any additional steps as the SOA region structure is the same as the gain section. However, the integration of a high-speed modulator does add a few more steps related to the low k dielectric and topside n-metal contacts. Although topside n-metal contacts are not strictly necessary on S-doped substrates, they do allow for direct radio frequency (RF) probing of the devices without the necessity of a well 70

88 designed RF carrier or influences of wirebonds/ribbon bonds. Because of this, Ni-AuGe-Ni-Au n-metal was deposited on both types of substrates. Also, backside Ti/Pt/Au n-metal was deposited on the thinned wafers to facilitate soldering which has much better heat conduction than thermally conducting epoxy. Table 2-2 Fabrication Steps Necessary Desirable Not necessary STEPS BASIC SGDBR Description S-Doped Substrate SGDBR + Modulator SI- Substrate SGDBR + Modulator Active/Passive Remove QWs in passive regions Sampled Grating Bursts and Gratings holographic gratings defined Regrowth Upper cladding is grown over gratings and active regions Dry/Wet Ridge Define ridge Etch Ridge waveguide structure Passivation Etch Important for high speed modulation BCB/Dielectric Deposition Low k dielectric to reduce capacitance N-Metal Etch Etch down to buried InGaAs layer N-Metal Ni Ge-Au-Ni Au Deposition Contacts to topside 71

89 p-ingaas Etch Remove the InGaAs conductive region between sections Isolation p+ Isolate the P-regions Implant between sections Modulator via Open through the BCB and SiNxOy layers for p-contact Laser Contact Open through the via SiNxOy for the p- contact P-Metal TiPtAu Metal Deposited Thinning Required for cleaving Backside Metal N-contact or soldering layer Total Steps 10 13/15 14/15 The addition of the high speed modulator adds a few more process steps (see total steps for each type) however the yield of devices is not compromised excessively. Detailed process development for high speed modulators is detailed in chapter 3. 72

90 REFERENCES [200] Chu SNG, Logan RA, Geva M, Ha NT. Concentration dependent Zn diffusion in InP during metalorganic vapor phase epitaxy. Journal of Applied Physics, vol.78, no.5, 1 Sept. 1995, pp [201] Berger P.R., S.N.G. Chu, R.A. Logan, Erin Byrne, D. Coblentz, James Lee Ill, Nhan T. Ha, N. K. Dutta, Substrate orientation effects on dopant incorporation in InP grown by metalorganic chemical vapor deposition J. Appl. Phys. 73(8), 15, April [202] Flemish J. R., H. Shen, K.A. Jones, M. Dutta, Determination of the composition of strained InGaAsP layers on InP substrates using photoreflectance and double-crystal x-ray diffractometry, J. Appl. Phys 70 (4), 15 August, [203] Li E.H., Material parameters of InGaAsP and InAlGaAs systems for use in quantum well structures at low and room temperatures, Physica E 5 (2000) [204] Matsui J., K. Onabe, T. Kamejima, I. Hayashi, Lattice Mismatch Study of LPE-Grown InGaPAs on (001)-InP Using X-Ray Double-Crystal Diffraction, J. Electrochemical Society. Vol 126, no4, April 1979, pg [205] Asai H., K. Oe., Energy band-gap shift with elastic strain in GaxIn1-xP epitaxial layers on (001) GaAs substrates, J. Appl. Phys. 54(4) April [206] Swaminathan V., C. L. Reynolds Jr., M. Geva, Effect of Zn on the electro-optical characteristics of metalorganic chemical vapour deposition grown 1.3um InGaAsP/InP lasers Electron. Lett. Vol 32., No.7, Mar. 28, 1996 [207] Chen C -H, U. M. Gosele, T.Y.Tan, Dopant diffusion and segregation in semiconductor heterostructures: Part 1. Zn and Be in III-V compound superlattices Applied Physics A, [208] Camargo Silva M.T., J. E. Zucker, L. R. Carrion, C. H. Joyner, A. G. Dentai, Growth Optimization for p-n Junction placement in the integration of Heterojunction Bipolar Transistors and Quantum Well Modulators on InP IEEE. J. Quantum Electron. Vol.6, No.1, Jan.,

91 [209] Hornstra J., W. J. Bartels, Determination of the lattice constant of epitaxial layers of III-V Compounds J. Crystal Growth 44(1978) [210] Bensaada A., A. Chennouf, R.W. Cochrane, J.T. Graham, R. Leonelli, R.A.Masut, Misfit strain, relaxation, and bandgap shift in GaxIn1-xP/InP epitaxial layers J. Appl. Phys. 75 (6) 15 March [211] Van Geelen A, T.M.F. de Smet, T. van Dongen, W.M.E.M van Gils, Zinc doping of InP by metal organic vapour phase diffusion (MOVPD), J. Crystal Growth 195(1998) [212] Franke D., H. Roehle, Highly reproducible and defect-free MOVPE overgrowth of InGaAsP-based DFB gratings J. Crystal Growth, 170(1997) [213] Belenky G.L., C.L. Reynolds Jr., D.V. Donetsky, G. E. Shtengel, M.S. Hybertson, M.A. Alam, G. A. Baraff, R.K. Smith, R. F. Kazarinov, J. Winn, L.E. Smith, Role of p-doping Profile and Regrowth on the Static Characteristics of 1.3um MQW InGaAsP-InP Lasers: Experiment and Modeling IEEE. J. Quantum Elect., Vol. 35, No. 10, Oct [214] Grinberg A.A., M. A. Alam, S. K. Sputz, Modeling of the photoluminescence in Multi-quantum well Heterostructure Laser wafers, IEEE. J. Quantum Elect. Vol. 35, No. 1, Jan [215] Schroeter-Janssen H, Roehle H, Franke D, Bochnia R, Harde P, Grote N. Comparison of MOVPE-based Zn diffusion into InGaAsP/InP using H/sub 2/ and N/sub 2/ carrier gas. Elsevier. Journal of Crystal Growth, vol.221, Dec. 2000, pp [216] Mei XB, Loi KK, Chang WSC, Tu CW. Improved electroabsorption properties in 1.3 mu m MQW waveguide modulators by a modified doping profile. Elsevier. Journal of Crystal Growth, vol , May 1997, pp [217] Dong-Ning Wang, Venables D, Waltemyer D, Lentz J. Investigation of p-n junction and dopant profiles in InP-based laser by low voltage SEM. Conference Proceedings International Conference on Indium Phosphide and Related Materials (Cat. No.00CH37107). IEEE. 2000, pp [218] Swaminathan V, Reynolds CL Jr, Geva M. Zn diffusion behavior in InGaAsP/InP capped mesa buried heterostructures. Applied Physics Letters, vol.66, no.20, 15 May 1995, pp

92 [219] Cheng-Yu Tai, Seiler J, Geva M. Modeling of Zn diffusion in InP/InGaAs materials during MOVPE growth. Conference Proceedings. Eleventh International Conference on Indium Phosphide and Related Materials (IPRM'99) (Cat. No.99CH36362). IEEE. 1999, pp [220] Rolland C., G. Mak, W. Bardyszewski, D. Yevick, Improved Extinction Ratio of Waveguide Electroabsorption Optical Modulators Induced by and InGaAs Absorbing layer, J.of Lightwave Tech., Vol. 10, No.12, [221] Masanovic M.L., V. Lal, J. S. Barton, E. J. Skogen, L. A. Coldren, and D. J. Blumenthal, "Monolithically integrated Mach-Zehnder interferometer wavelength converter and widely tunable laser in InP," IEEE Photonics Technology Letters, vol. 15, pp , [222] Mason B, Barton J, Fish GA, Coldren LA, Denbaars SP. Design of sampled grating DBR lasers with integrated semiconductor optical amplifiers. IEEE Photonics Technology Letters, vol.12, no.7, July 2000, pp [223] Fish G., InGaAsP/InP based photonic integrated circuits for optical switching, Dissertation [224] Mason B., InP Based Photonic Integrated Circuits, Dissertation, [225] Silver M., E. P. O Reilly, Optimization of long wavelength InGaAsP strained quantum well lasers, IEEE Journal of Quanum Electronics, vol. 31, pp ,

93 Chapter 3 Lumped Modulator Designs A number of different authors have fabricated lumped Mach-Zehnder modulators, mostly using MQW structures[300-4]. For this work Franz-Kelydsh bulk quaternary waveguides were used instead of MQWs as they tend to have less wavelength dependence and lower optical propagation loss 12. DBR[338-9], DFB[306-8] and now SGDBR[309] laser integrated Mach-Zehnder modulators have been explored as well. In this chapter, we will look at the factors that influence the modulation efficiency, optical bandwidth, and insertion losses. Fig 3.1 Lumped electrode Mach-Zehnder Modulator 12 Fiber to fiber losses 31-40dB reported in [303] 76

94 3.1 DEVICE EFFICIENCY Mach-Zehnder modulators rely on a change in the electric field and carrier density to change the absorption and phase in one branch of the MZ modulator. In this work, a PN junction is formed in the first 80nm of the waveguide by diffusion of Zn during the regrowth 13. This gives an electric field profile as given in fig. 3-2 for a waveguide doping level of 3e17cm -3 Si. Fig. 3-2 Electric field profile and indexing structure for waveguide as a function of bias for device doping structure The device efficiency would be improved further by moving the Zn diffusion front into the middle of the waveguide. However, this would mean more free 13 Fig 2.6a&b 77

95 carrier absorption due to the Zn overlap with the optical mode, and a deeper passivation etch as outlined previously in fig. 2-7a&b. Unfortunately, this optimization approach also maximizes the Franz-Keldysh absorption. If less wavelength dependence is desired, it is best to use the LEO effect along with carrier based effects with a design that improves the overlap factor (Γ) by using a lower doping level in the waveguide effectively spreading out the electric field profile. Ideally, the modulator biases will be as low as possible to lower the propagation loss through the device. In this work, we have a doping profile as shown in Fig 3-3. This doping profile was chosen as a compromise between having reasonable bandwidth and the high efficiency associated with a PN junction and it is compatible with typical SGDBR processing. Fig 3-3 Doping profile for device structure in passive regions 78

96 The semi-insulating substrate based devices have a similar structure in the waveguide with a 0.1µm 1E18 Si doped InGaAs contact layer, and 0.5µm 1E18 Si-doped n-inp layer instead of the Sulfur doped substrate shown in fig The efficiency can be measured directly by analyzing the DC extinction curves. 3.2 DC EXTINCTION CURVES The output power intensity as a function of voltage can be derived knowing the optical field in the two branches of the Mach-Zehnder. The light power intensity is given by: 2 1, V2 ) E( V1, 2 ) I( V V [3.1] The optical field from the output of a Mach Zehnder is given by[301]: E V1, V2 ) Ei SRinSR ( V1 ) ( V2 ) exp j ( V1 ) L exp j 2 2 ( out 2 ( V ) L [3.2] where E i E o ( 1 SRin )(1 SRout ) [3.3] where V 1 and V 2 are voltages on the two MZ branches, is the change in optical absorption, is related to the change in index where 2n the power splitting ratio of the input MMI. SR in = (P 1 /P 2 ) in.. L is the length of the modulator. SR in is 79

97 Fiber-coupled output power was measured for a 300m long modulator as a function of input optical power as shown in fig. 3.3 and fit with the model from equation [3.2]. Fig 3.3 Fiber coupled output power at =1555nm for three different input powers (4.9mW, 11mW, 15.6mW) corresponding to SOA = (20mA, 60mA, 120mA) Using the expression for optical field from equation 3.3, along with the change in absorption, one can obtain the refractive index change. The Franz Keldysh absorption can be modeled with the following model. 80

98 3.3 FRANZ-KELDYSH ABSORPTION Franz-Keldysh absorption is caused by is the tilting of the bands during reverse bias which in turn increases the tunneling probability as the electric field increases. The phenomenon of absorption is caused by the presence of photon assisted inter-band tunneling due to the tilted energy band diagram. The absorption of Franz-Kelydsh based modulators [311] is given by the following relation[346]: ( h, E ) A j j E 1/ 3 dai( z) dz j j A ( i j 2 ) 2 [3.4] where E is the electric field in V/cm. Ai are airy functions j B ( E h ) E j g 2 / 3 [3.5] A j 3.55x nh m j 4 / 3 [3.6] The sum of j is over both the light and heavy hole bands where h is plank s constant, is the effective mass, and is the angular frequency, m is the mass of an electron, n is the refractive index, and Eg is the bandgap. Even in the absence of applied field, the built-in electric field contributes to a static absorption coefficient is given by the Urbach tail expressed by[312]: The internal field is given by: o g i A exp ( E hv) / F [3.7] F ( V V ) / d [3.8] i bi i 81

99 where V bi is the built-in field, V is the applied voltage, and d i is the intrinsic region width. The Franz-Keldysh absorption is shown for waveguide compositions close to the bandgap ( ) in fig. 3-5 as calculated from equation 3-2. Quat = 1.4Q Quat = 1.41Q Quat = 1.42Q Absorption (cm-1) Absorption (cm-1) Wavelength (nm) Quat = 1.43Q Wavelength (nm) Quat = 1.44Q Wavelength (nm) Quat = 1.45Q 0kV/cm 50kV/cm 100kV/cm 150k/cm 200kV/cm Wavelength (nm) Wavelength (nm) Wavelength (nm) Fig. 3-5 Total Franz-Keldysh Absorption as a function of wavelength for different waveguide compositions and Electric fields. As can be seen in Fig 3-5, considerable wavelength dependence in the absorption will result from using a waveguide composition bandgap closer to the operating wavelength. Verification of this model was performed by comparison with the absorption as measured with photocurrent as the following relation: 82

100 1 ln 1 L I pc ( V ) qpin [3.9] where is the modal absorption, L is the length of the modulator, I pc (V) is the photocurrent generated, P in is the input optical power. And hw/q is the phonon energy in ev. Fig 3.6 Modal absorption data and FKE model above for a 300m long modulator As can be seen in fig 3.6, the Franz-Keldysh model fits the absorption data well as shown for three different input powers (P in = 4.9mW, 11mW, and 15.6mW). 83

101 The refractive index can be changed with two different classes of effects - Field effects consisting of the linear electro-optic effect (LEO or Pockels effect) and the quadratic (electrorefractive or Kerr) effect, and Carrier based effects such as carrier induced bandgap shrinkage, heating induced bandgap shrinkage, Plasma loading, and the Burstein shift (Band-filling effect). These effects will be outlined below: 3.4 ELECTRIC FIELD EFFECTS The change in index due to electric field effects is given by the following equation[305]: nbulk no [ r( E Eo ) s( )( E Eo ) ( reo s( ) Eo )] [3-10] 2 n o is the index without field applied Ґ -is the overlap confinement factor E is the electric field r is the linear coefficient s is the quadratic coefficient LINEAR ELECTROOPTIC EFFECT (LEO) (Pockels effect) This effect is due to the biaxial birefringence induced by the presence of an electric field. It is polarization dependent where the effect is positive for TE 84

102 light if the light propagates in the [-110] direction and negative if the light propagates in the [110] direction. 3 n r41e n LEO [3.11] 2 The linear electro-optic coefficient has been measured for InGaAsP quaternaries in [344] and can be extrapolated for different compositions as was done in [345] using the coefficients in table 3.1. Table 3.1 Fitting Parameters for various materials from [345] 1 Material A0 B0 C0 D0 E0 F0 InP GaP GaAs InAs P44 r41 For the wavelength range, and quaternary compositions of interest, the r 41 coefficient is plotted in fig

103 Fig 3.7 Linear electro-optic coefficient extrapolated as a function of waveguide composition and wavelength from [345] and [344] QUADRATIC ELECTRO-OPTIC EFFECT OR KERR EFFECT (QEO)/ELECTROREFRACTIVE EFFECT This effect is a third order nonlinear optical process and unlike the LEO effect, is polarization independent. In order to determine the index change as a function of absorption change one uses the Kramers-Kronig transform[316]. hc n( E) 1 2 P 2 0 ( ) 2 E' E 2 de' [3.12] where the Principle is given by P 0 lim a 0 Ea 0 Ea 86

104 0 hc 1 ( ) n(, E ) P de' [3.13] field 2 2 ' E de' The resulting Kerr effect turns out to be fit well by a quadratic relation with respect to electric field. The change in refractive index is related to the square of the electric field as expressed by: n ker r n 3 R ker r 2 E 2 [3.14] As the composition of the waveguide approaches that of the operating wavelength, the quadratic dependence (due to electro-refractive effect) becomes more pronounced where the Kerr coefficient has been empirically approximately as[347]: R r 1 V 15 2 ker.5x10 exp( 8.85 E) cm / 2 [3.15] where E is the difference between the photon energy of the guided light and the quaternary material gap energy. 87

105 Fig 3.8 The Kerr Coeffient as a function of wavelength and waveguide composition The Kerr coefficient as a function of wavelength and waveguide composition wavelength are shown in fig 3.8. as extrapolated from the Adachi model [344]. 3.5 CARRIER BASED EFFECTS As the waveguide is depleted of carriers under reverse bias, the index changes due to a few effects. As shown later, the carrier based effects are significant for doped waveguides and important for high modulation efficiency. PLASMA EFFECT The Plasma effect is due to intraband free carrier absorption in both the valence and conduction bands. Free carrier absorption in p-type material 88

106 (most important) consists of mostly intraband and interband absorption of holes[2]. The free-carrier plasma reduces the index of refraction of the material. The following formula gives the change in index for free holes and free electrons as a function of doping. n plasma ro 2n 2 N me P m h [3.16] where the hole effective mass is given by: m h m m 3/ 2 hh 1/ 2 hh m m 3/ 2 lh 1/ 2 lh [3.17] where r o = 2.82E-13 cm, N is the electron density, P is the hole density, m e is the electron mass, m h is the hole effective mass, n is the refractive index, and is the wavelength of light[349]. 89

107 Figure 3.9 Refractive index change due to Plasma effect as a function of Donor concentration, and waveguide composition[349] Assuming N-doped waveguide As can be seen from Figure 3.9, the plasma effect is significant for higher(>1e16) waveguide doping concentrations. This relationship can be approximated as[347]: 21 n plasma 3.63x10 N [3.18] The relationship is fairly independent of waveguide composition and wavelength. N is the doping level (n-type) and this approximation is at 1.52um. 90

108 BAND-FILLING EFFECT / BURSTEIN SHIFT This effect is brought about by the removal of carriers in the depletion region and the resulting reduction of absorption of the region. This effect, also known as the Burstein-Moss effect [347], has been described as bandfilling of the conduction band in n-type semiconductors. Because of this bandfilling, subsequent valence band electrons require greater energy to be excited into the conduction band resulting in less absorption. This effect is interdependent with carrier induced bandgap shrinkage effects. When the device is reverse biased, the index reduces as bandemptying occurs. Assuming a parabolic band structure, the bandfilling-induced change in absorption is given by: ( N, P, E) C hh EE g [ f v ( E ah ) f c ( E bh ) 1] C lh EE g [ f v ( E al ) f c ( E bl ) 1] o ( E) [3.19] where N, and P are the carrier densities, and E the photon energy, E g is the bandgap energy, f v f c is the Fermi-dirac probability functions for the valence and conduction bands respectively, Chh and Clh are fitting constants for the light holes and heavy holes. [337] The change in index due to the bandfilling can be evaluated from the Kramer s Kronig relations mentioned earlier however, one also needs to take into account the carrier induced bandgap shrinkage particularly in the 0.9e17 to 91

109 4e17 n-type doping range. The bandfilling effect has been empirically determined as a function of composition for 1.52m in [347]. CARRIER INDUCED BANDGAP SHRINKAGE Shrinkage of the waveguide bandgap occurs due to two major mechanisms. Firstly, as the PN junction in the waveguide is depleted out, there is a change in the bandgap due to the change in carriers based on the screening of electrons which lower the energy of the conduction band edge and raise the valence band. This shrinkage has been modeled analytically for InGaAsP materials in [349]: An Bn * * / 3 Eg * 1 no / n m r e [3.20] where A = 1.04E3, B = 2.8E-7, is -0.19, r is the relative dielectric constant Additionally, due to heating in the modulator particularly noticeable with such large photocurrent in an integrated device, the bandgap shrinks as well, as it is fairly sensitive to temperature. 92

110 3.6 TEMPERATURE INDUCED BANDGAP SHRINKAGE The bandgap of the material will shrink with increases in temperature. This has been expressed with the Varshni equations for unstrained materials: where Alpha is 4.9E-4 ev/k Beta = 327K. E g T ( T ) Eg (0K) [3.21] T The change in bandgap energy with temperature has also been extrapolated for binary data at 300K for lattice matched quaternary material and expressed as[350]: de g dt 1x10 4 ( y 0.61y 2 ) [3.22] The change in bandgap for the tensile strained modulator structure was measured using a micro-photoluminescence setup as a function of temperature as shown in fig

111 =0.432meV/K Fig 3-10 Temperature dependence of the waveguide composition emission wavelength The slope of the waveguide composition vs temperature is shown in fig This is a little higher than the slope shown in the literature[351] at 1.3Q of 0.333meV/K, probably due to the strain in the material. As can be seen in fig 3-10, the material is highly temperature sensitive and linear with respect to temperature. As the modulator heats up with high optical powers, the bandgap shrinks and the effective waveguide Q can change from 1.4Q to 1.435Q from 20-70C. Heat crosstalk is an important issue in integrated devices as the laser benefits from low temperatures with higher gain and lower optical loss, and the modulator benefits from the higher efficiencies at higher temperatures. The rise in temperature with bias can be evaluated with the following model: T P d Z t [3.23] where P d is the power dissipated, and the thermal impedance (Z t ) is given by: 94

112 ln(4h / w) Z t [3.24] L where h is the height of the substrate, w is the width of the device and L is the length. is the thermal conductivity[350]. The thermal resistivity of InGaAsP has been given by [350] as: y 39.42y 2 T K cm / W [3.25] The refractive index of InGaAsP as a function of composition, and temperature has been extended from a model given by Adachi, and fitted to experimental data and given by: n r A( y) f ( z) 1 2 E g ( T ) E g ( T ) o 3 / 2 f ( z o 1 ) B( y) o T ( T 300) [3.26] where A(y) = y [3.27] B(y) = y [3.28] 2 1 z 1 z f ( z) [3.29] 2 z E z [3.30] E (T ) g z o E g E ( T ) o [3.31] Although InGaAsP data is difficult to come by, for InP near 300K T x10 o 1/K [3.32] 95

113 As one can see, the temperature dependence stems from the bandgap and the high frequency dielectric constant terms. 3.7 ACCUMULATION OF EFFECTS The accumulation of the aforementioned field effects and carrier effects are plotted for a 300m long device with three different input powers (4.9mW, 11mW, and 15.6mW) under reverse bias. Fig. 3.11a Change in refractive index as a function of voltage with input optical power 4.9mW T = 16C = 1555nm 96

114 Fig. 3.11b Change in refractive index as a function of voltage with input optical power 11mW T = 16C, = 1555nm Fig. 3.11c Change in refractive index as a function of voltage with input optical power 15.6mW T = 16C 1555nm 97

115 The refractive index change was extracted from the absorption curves (fig. 3.6) and the output power dc extinction curves (fig. 3.3) using equation 3.2 and shown as DATA on each graph in fig. 3-11a-c. The total change is also plotted for each case accounting for all of the effects which fit very well the observed change in index. A number of conclusions can be made from these plots. First of all, the dominant effect is clearly the bandfilling effect or in this case bandemptying due to the n-doped waveguide. The plasma, linear and quadratic effects all have fairly similar contributions given the doping profile that was used. The rf change in index is the total change minus the heating portion as under RF modulation, the device will not heat up much. This RF line appears to line up well with the RF V pi data observed in the next chapter in this case 4V. From the change in phase due to a change in index, one can determine the modulator arm length required to achieve a pi phase shift. 2L n [3-33] Modulator Length to achieve pi shift. L 2 n [3-34] 98

116 For a 300m modulator, the index change to achieve a pi shift for 1550 nm is approximately is 0.26%. As can be seen from the previous plots, the index and absorption are strongly dependent on the optical power at DC as the modulator is heated and experiences bandgap shrinkage. Under RF modulation, this efficiency is not very power dependent. As the heating due to the photocurrent absorption changes the refractive in the same direction as the other effects, at DC this gives the appearance that the efficiency is better than it is at RF. Obviously, DC extinction is not a very good indicator of RF performance. 3.8 HIGH SPEED DESIGN Capacitance and carrier lifetime govern the maximum bandwidth possible for a modulator. For a lumped modulator with an open termination port, the smallsignal modulation response is given by: S [3.35] 1 jwrc assuming that the microwave attenuation is low[5]. Typically, there are three approaches to achieving high speed operation: low impedance matching 14, reducing the capacitance and distributing the modulation region 15. The 14 See Chapter 4 Termination Section 15 with T sections as shown in Chapter 4 99

117 capacitance can be decreased by either increasing the intrinsic region in the waveguide 16, lowering the pad capacitance with low k dielectrics, or decreasing the waveguide area[314,320] 17. An accurate account of the capacitance in the structure needs to take into account the junction capacitance [C j ], parallel plate capacitance [C pp ] of the interconnect region and the fringing capacitance[c f ] for the geometry as shown in the side-view of the modulator ridge in Fig Fig 3-12 Modulator end-view with different contributions of capacitance 16 Reducing the modulator efficiency and reducing the optical loss 17 Potentially increasing optical loss and or drive voltage 100

118 Next we will look at the minimization of these capacitances separately. In both lumped and traveling wave devices, one would like to reduce the capacitance per unit length. 3.5 JUNCTION CAPACITANCE MINIMIZATION The PN junction capacitance is minimized by using a short device with a narrow ridge. As shown in Fig. 3-13, the junction capacitance improves for wider intrinsic region widths and lower doping levels. The material exhibits less free carrier absorption with low doping particularly Zn. Structures with large intrinsic regions do not provide high electric fields so Fig 3-13 Capacitance per unit length [pf/m] for various doping structures - 2m ridge 101

119 clearly there is a tradeoff between capacitance as improved with a PIN structure and efficiency with a PN junction. Since the devices here use a lowdoped PN junction 18, the bandwidth varies considerably with bias as seen in the variance in fig 3-13 of the capacitance with bias. As an illustration of this, the small-signal modulation response is shown for a 200µm long lumped MZ at various biases in fig Fig Small-Signal normalized modulation response at 1555nm for a 200um long electrode device. 18 3e17 Si as in Fig 3-9b 102

120 The waveguide is depleted out as the bias induced electric field increases in the waveguide changing the capacitance and bandwidth as shown in fig In this work, the ridges were tapered down to 2.5µm (effectively 2.2µm) in the modulator regions. This did not seem to adversely affect the insertion loss of the modulators much as was shown in Chap. 1. Below 2µm wide, one would expect propagation losses to increase markedly due to light scattering. 3.6 PARASITIC CAPACITANCE MINIMIZATION There are a number of different innovative materials that can be used for providing a low-dielectric constant dielectric layer in the modulator section as shown in table

121 Table 3-1 Dielectric Materials Material Dielectric Constant Nanoporous silica Fluorinated organic polymers Fluorinated amorphous carbon Non-fluorinated organic polymers Cyclotene Benzocyclobutene (BCB) 2.65 SILK (Dow) 2.65 Non-fluorinated polymers Inorganic polymers Phase separated hybrids Poly-imides Fluorinated HDPCVD SiO 2 Fluorinated PECVD SiO Thermal SiO Plasma deposited SiO Thermal silicon nitride Si 3 N Plasma silicon nitride Si-N-H Application techniques, vary from LPCVD, PECVD, sputtering, to deposition of low-k liquids by simple spin coating and multiple baking techniques, similar to photoresist processing. These materials are helpful for a number of reasons. First of all, the parasitic capacitance in the modulator is reduced due to the low dielectric constant which is important for high speed. Also, the dielectrics are useful for planarization over rough topographies on InP wafers particularly with n-topside contacts. 104

122 Although there are a number of low-k electronic material candidates for electronic device designs such as oxide-based materials that can handle temperatures as high as 600 C, Cyclotene BCB was chosen for fabrication as the dielectric material has not only a low dielectric constant (2.65) but is easily cleaved and easily applied. Fig 3-15a PhotoBCB planarized ridges Fig. 3-15b Dry-etchable BCB Although dry-etchable BCB tends to have superior planarity over photodefinable varieties (see fig 3-15ab) the latter choice avoids excessive overetches of the BCB that are necessary in order to remove BCB residuals fully from the surface as shown in Fig The shelf life of Photo-BCB is not very long however at room temperature 19, so freezing it is a necessity. 19 approximately 1 week 105

123 BCB scum voids Fig BCB residuals after etch The reactive ion etcher (RIE) tends to leave a BCB residue scum on the surface with the etch conditions that were used consisting of 20% CF 4 / 80% O 2 with either 250V (W) or 350V (W) conditions as recommended by Dow. Going to a lower CF 4 percentage gives better selectivity between the BCB and Silicon oxy-nitride layers however is more susceptible to oxide scum and the etch rate decreases dramatically. BCB Fig 3-17 Cyclotene 4024 PhotoBCB defined in only the modulator regions. 106

124 Due to adhesion problems and device heat dissipation issues BCB was defined to be only under the modulator section pads. This was defined using a photolithographic stepper tool and the rest developed off using a puddle emersion developer DS-2100 avoiding a 5µm BCB etch. It was found that adhesion of the pads during wirebonding was not acceptable on the first device run due to the BCB being etched under the pads which had excessive roughness and resulted in delamination during wedgebonding. Using a different approach only etching a via to the ridge and leaving the area under the pad unetched with a sandwiching layer of SiN y O x on top of the BCB proved superior not only as a thicker dielectric leaving lower parasitic capacitance but very good adhesion for wirebonding. See process Appendix C. Photo-BCB does not have very good definition resolution as can be seen in fig. 17-a with very sloped sidewalls, however it is sufficient for this application. 3.7 FRINGING CAPACITANCE Using the basic geometry given in Fig. 3-12, one can calculate the parallelplate capacitance C pp of the interconnect segment. However, in interconnect lines where the wire thickness (t) is comparable in magnitude to the groundplane distance (h), fringing electric fields significantly increase the total parasitic capacitance (fig. 3-1). It has been shown [315] that the influence of fringing fields increases with the decreasing (w/h) ratio, and that the fringing-field 107

125 capacitance can be as much as times larger than the parallel-plate capacitance. It was mentioned earlier that the sub-micron fabrication technologies allow the width of the metal lines to be decreased somewhat, yet the thickness of the line must be preserved in order to ensure structural integrity. This situation, which involves narrow metal lines with a considerable vertical thickness, is especially vulnerable to fringing field effects. A set of simple formulas [315] can be used to estimate the capacitance of the interconnect structures in which fringing fields complicate the parasitic capacitance calculation. The following two cases are considered for two different ranges of line width (w) ln 2 2 t h t h t h h t w C for 2 t w [3.36] ln (1 t h t h t h h t h w C for 2 t w [3.37] 108

126 where t, h and w are the dimensions as shown in Fig These formulas permit the accurate approximation of the parasitic capacitance values to within 10% error, even for very small values of (t/h). The other contribution of capacitance is attributed to the parasitic capacitance of the contact pad. This contribution was measured in Fig 4-3b to be approximately 0.2pF. Figure 3-18 shows the parasitic capacitance as a function of dielectric thickness for different dielectrics and modulator lengths. Parasitic Pad Capacitance (pf) BCB 100um device BCB 200um device BCB 300um device SiNx 100um device SiNx 200um device SiNx 300um device Dielectric Thickness (µm) Fig Pad Capacitance for different dielectrics and pad sizes w/fringing fields 3.8 MULTI-MODE INTERFERENCE DESIGN Another very important element to the Mach-Zehnder design is that of the MMI splitters and combiners[ ]. General MMI theory states that the 109

127 shortest 1x2 splitter requires a length of 3/8L pi where the beat length of the two lowest order modes is given by[328]: L 4 n w eff eff [3-38] 3 o 2 where n eff is the effective index of the mode, is the wavelength, and w eff is the equivalent width of the MMI Fig. 3-19a Electric Field profile for the optimized MMI design showing imaging into the two MZ branches (Waveguide 1550nm) Fig. 3-19b MMI with curved waveguides (Height = 9um Length = 85um, taper = 20um) Using Beamprop, an MMI design was optimized with a center wavelength of 1550nm as shown in fig 3-19ab. MMIs have broad optical bandwidth 20 [328] much wider than the tuning range of the SGDBRs here (38nm). The length of the MMI becomes very long for wide widths due to the quadratic dependence so it is imperative to minimize the width as much as possible. A 9 µm wide MMI was chosen to that the gap between the waveguides could be resolved with the stepper as shown in Fig Note also the high angle sidewall in 20 close to 100nm for 1dB bandwidth 110

128 this gap due to the crystal orientation during the ridge wet etch. This sidewall is not likely to adversely affect reflections in the device in fact it gives a more gradual index discontinuity. Fig 3-20 Gap between waveguides approximately 1µm Curved waveguides were used to extend the separation distance to 16µm as shown in Fig 3-21 to minimize the propagation distance. The ridge was defined using a dry etch/wet etch process where approximately 1µm of material is RIE etched with Methane/Hydrogen/Oxygen with a subsequent 3:1 H 3 PO 4 :HCl wetetch to remove the rest of the InP on top of the waveguide. As the radius of curvature is low, the sidewall roughness appears to be low as shown in fig

129 Fig Sidewall roughness on curved waveguides and MMI taper 3.9 PHASE SHIFTER A phase shifter electrode was integrated in one branch of the MZ in order to facilitate changing the phase for different wavelengths. It is best to design the waveguide structure to achieve a pi-phase shift without bias. Pi-shifted modulators have been fabricated with one length a multiple of 0.241µm 21 longer than the other. The devices in this dissertation utilize a pi-shifted configuration however this is accomplished using one ridge slightly wider (0.2µm) in the curved waveguide regions than the other to achieve the pi shift 22 [300]. Unlike the RF sections, the phase section can be forward biased, which gives close to 5x the index shift as reverse bias as illustrated in fig for 1550nm 22 This is easier due to fabrication tolerances 112

130 Fig Fiber-coupled power for device #1 as a function of bias on phase section in both forward and reverse bias As this device needs to operate over the full C-Band in which the pi-shift will change with wavelength, designs allowed for the use of a forward biased electrode to achieve the pi-phase shift. This requires fairly good control over waveguide widths/thicknesses/compositions in order to achieve from run-torun. By biasing this electrode however, it induces a significant amount of loss in that waveguide as shown in fig Ideally the device is forward biased slightly as very little current is required ~2mA to achieve the desired phase. 113

131 Fig Normalized Optical Loss vs. wavelength and bias for 100µm long phase electrode The loss was measured with Device #1 23 where the laser sections are forward biased, and the SOA is reverse biased to measure the optical power that makes it through the phase section as a function of bias on the phase electrode ST GENERATION DESIGN The initial design involved the integration of a SGDBR with a passive Mach- Zehnder modulator as demonstrated in fig see Generation 2 designs

132 Fig 3-24 Integrated SGDBR- Mach Zehnder modulator One branch of the MZ modulator was meant for DC biasing to change the phase for each wavelength, and the second for RF modulation (Pad #2). Devices were fabricated with parameters as shown in Table 3-2. These modulators uses two identical 3dB MMI splitters/combiners that are 98µm long as described in section 3.8. Table st Generation Devices Mach Zehnder Lengths 550,750,950 Waveguide offset 40um Width1 2um Width2 2.1 to 2.2 to achieve pi shift Curve length 185 Curve width 20um Trench 15um MMI length 98um MMI width 9um 115

133 Although these devices were fairly long and suffered from high capacitance due to the problem outlined in fig. 2-7b, the DC extinction and chirp 24 characteristics looked promising as shown in fig Output Power (dbm) Vbias +0.6V +0.3V 0V Bias -1V Bias -2V Bias -3V Arm #1 DC Bias Voltage Fig µm long electrode at λ = 1535nm ND GENERATION DESIGNS A number of different improvements were made to the 2 nd generation devices to improve performance. First, SOAs were integrated before the MZ and inside the MZ modulator to mitigate the 4-5dB insertion losses. Additionally, the gap between the two waveguides was reduced from 37um to 16um which allowed for shorter curved waveguide sections. A 2x2 MMI was placed at the output to 24 As will be shown in Chap 5 116

134 guide away off-state light in a controllable way as shown in fig The output was curved and flared as well as a front passive detector electrode placed on the output waveguide to reduce reflections. Also, two RF electrodes were placed on each device so that push-pull modulation could be possible. In addition, considerably shorter electrodes were employed to improve the high speed performance. Laser Input Output 1x2 splitter 2x2 combiner Fig Ridge waveguide structure illustrating the 1x2 and 2x2 MMIs with curved waveguides and output flares The first three devices have Dual SOAs as mentioned in Chapter 1. Device 7 and 8 have electrodes at the rear of the modulator for rear resistive termination as will be elaborated in chapter 4 and

135 The first 8 designs use lumped electrodes and are shown for reference: Table 3-3 Lumped electrode MZ devices Total device length = 3200µm SOA # Config MZ Electrode Length SOA Length 1 Dual Dual Dual Single Single Single Single Single

136 REFERENCES [300] Yu J, Rolland C, Yevick D, Somani A, Bradshaw S. Phase-engineered III-V MQW Mach-Zehnder modulators. IEEE Photonics Technology Letters, vol.8, no.8, Aug. 1996, pp [301] Lawetz C, Cartledge JC, Rolland C, Yu J. Modulation characteristics of semiconductor Mach-Zehnder optical modulators. Journal of Lightwave Technology, vol.15, no.4, April 1997, pp [302] Bardyszewski W, Yevick D, Yong Liu, Rolland C, Bradshaw S. Theoretical and experimental analysis of Mach-Zehnder quantum-well modulators. Journal of Applied Physics, vol.80, no.2, 15 July 1996, pp [303] Fetterman M, Chao C-P, Forrest SR. Fabrication and analysis of highcontrast InGaAsP-InP Mach-Zehnder modulators for use at mu m wavelength. IEEE Photonics Technology Letters, vol.8, no.1, Jan. 1996, pp [304] Cites JS, Ashley PR. High-performance Mach-Zehnder modulators in multiple quantum well GaAs/AlGaAs. Journal of Lightwave Technology, vol.12, no.7, July 1994, pp [305] T. Hausken, High Performance Double Heterostructure and Quantum Well Waveguide Phase Modulators in GaAs/AlGaAs, P.h.D Dissertation [306] Xun Li, Huang W-P, Adams DM, Rolland C, Makino T. Modeling and design of a DFB laser integrated with a Mach-Zehnder modulator. [Journal Paper] IEEE Journal of Quantum Electronics, vol.34, no.10, Oct. 1998, pp [307] Adams DM, Rolland C, Puetz N, Moore RS, Shepherd FR, Kim HB, Bradshaw S. Mach-Zehnder modulator integrated with a gain-coupled DFB laser for 10 Gbit/s, 100 km NDSF transmission at 1.55 um. Electronics Letters, vol.32, no.5, 29 Feb. 1996, pp [308] Adams DM, Rolland C, Fekecs A, McGhan D, Somani A, Bradshaw S, Poirier M, Dupont E, Cremer E, Anderson K um transmission at

137 Gbit/s over 1102 km of NDSF using discrete and monolithically integrated InGaAsP-InP Mach-Zehnder modulator and DFB laser. Electronics Letters, vol.34, no.8, 16 April 1998, pp [309] Barton JS, Skogen EJ, Masanovic ML, DenBaars SP, Coldren LA., Tailorable chirp using integrated Mach-Zehnder modulators with tunable sampled grating distributed Bragg reflector lasers IEEE 18th International Semiconductor Laser Conference. Conference Digest (Cat. No.02CH37390). IEEE. 2002, pp [310] Chin MK. Optical confinement factor in waveguide electroabsorption modulators: design and measurement. IEEE Photonics Technology Letters, vol.6, no.8, Aug. 1994, pp [311] Chin MK. Effect of electroabsorption on electrorefractive intensity modulators. IEEE Photonics Technology Letters, vol.4, no.6, June 1992, pp [312] J. D. Dow and D. Redfield, Towards a unified theory of Urbach s rule and exponential absorption edge, Phys. Rev. B 5, (1972) [313] Mason B, Fish GA, DenBaars SP, Coldren LA. Widely tunable sampled grating DBR laser with integrated electroabsorption modulator. IEEE Photonics Technology Letters, vol.11, no.6, June 1999, pp [314] Satzke K, Baums D, Cebulla U, Haisch H, Kaiser D, Lach E, Kuhn E, Weber J, Weinmann R, Wiedemann P, Zielinski E. Ultrahigh-bandwidth (42 GHz) polarisation-independent ridge waveguide electroabsorption modulator based on tensile strained InGaAsP MQW. Electronics Letters, vol.31, no.23, 9 Nov. 1995, pp [315] Yuan CP, Trick TN. A simple formula for the estimation of the capacitance of two-dimensional interconnects in VLSI circuits. IEEE Electron Device Letters, vol.edl-3, no.12, Dec. 1982, pp [316] V. Swaminathan, A. T. Macrander, Materials Aspects of GaAs and InP Based Structures Prentice Hall Advanced Reference Series, pg 355. [317] Penninckx D, Delansay Ph. Comparison of the propagation performance over standard dispersive fiber between InP-based pi -phase-shifted and symmetrical Mach-Zehnder modulators. IEEE Photonics Technology Letters, vol.9, no.9, Sept. 1997, pp

138 [318] Hishimura Nishimura S, Inoue H, Sano H, Ishida K. Electrooptic effects in an InGaAs/InAlAs multiquantum well structure. IEEE Photonics Technology Letters, vol.4, no.10, Oct. 1992, pp [319] Mak G, Rolland C, Fox KE, Blaauw C. High-speed bulk InGaAsP-InP electroabsorption modulators with bandwidth in excess of 20 GHz. IEEE Photonics Technology Letters, vol.2, no.10, Oct. 1990, pp [320] Ido T, Tanaka S, Suzuki M, Koizumi M, Sano H, Inoue H. Ultra-highspeed multiple-quantum-well electro-absorption optical modulators with integrated waveguides. Journal of Lightwave Technology, vol.14, no.9, Sept. 1996, pp [321] Delansay P, Penninckx D, Boucherez E, Fortin C, Le Gouezigou O. InP/GaInAsP phase-shifted Mach-Zehnder modulator for wavelengthindependent ( nm) performance in 10 Gbit/s transmission over dispersive fiber. Conference on Optical Fiber Communications. Technical Digest. Postconference Edition OSA Technical Digest Series. Vol.6 (IEEE Cat. No.97CH36049). Opt. Soc. America. 1997, pp [322] Rolland C. InGaAsP-based Mach-Zehnder modulators for high-speed transmission systems. OFC '98. Optical Fiber Communication Conference and Exhibit. Technical Digest. Conference Edition OSA Technical Digest Series Vol.2 (IEEE Cat. No.98CH36177). Opt. Soc. America. 1998, pp [323] Soda H, Furutsu M, Sato K, Okazaki N, Yamazaki S, Nishimoto H, Ishikawa H. High-power and high-speed semi-insulating BH structure monolithic electroabsorption modulator/dfb laser light source. Electronics Letters, vol.26, no.1, 4 Jan. 1990, pp [324] Jackel JL, Perlmutter P, Johnson J. High-speed low-voltage modulation with a nonsymmetric Mach-Zehnder interferometer. Journal of Lightwave Technology, vol.7, no.6, June 1989, pp [325] Rolland C, Mak G, Prosyk KL, Maritan CM, Puetz N. High speed and low loss, bulk electroabsorption waveguide modulators at 1.3 mu m. IEEE Photonics Technology Letters, vol.3, no.10, Oct. 1991, pp [326] Pennings ECM, van Roijen R, van Stralen MJN, de Waard PJ, Koumans RGMP, Verbeck BH. Reflection properties of multimode interference devices. IEEE Photonics Technology Letters, vol.6, no.6, June 1994, pp

139 [327] Soldano LB, Pennings ECM. Optical multi-mode interference devices based on self-imaging: principles and applications. Journal of Lightwave Technology, vol.13, no.4, April 1995, pp [328] Besse PA, Bachmann M, Melchior H, Soldano LB, Smit MK. Optical bandwidth and fabrication tolerances of multimode interference couplers. Journal of Lightwave Technology, vol.12, no.6, June 1994, pp [329] Erasme D, Spiekman LH, Herben CGP, Smit MK, Groen FH. Experimental assessment of the reflection of passive multimode interference couplers. IEEE Photonics Technology Letters, vol.9, no.12, Dec. 1997, pp [330] Leuthold J, Joyner CW. Multimode interference couplers with tunable power splitting ratios. Journal of Lightwave Technology, vol.19, no.5, May 2001, pp [331] Rolland C, Adams DM, Yevick D, Hermansson B. Optimization of strongly guiding semiconductor rib waveguide Y-junctions. IEEE Photonics Technology Letters, vol.2, no.6, June 1990, pp [332] Cartledge JC, Rolland C, Lemerle S, Solheim A. Theoretical performance of 10 Gb/s lightwave systems using a III-V semiconductor Mach-Zehnder modulator. IEEE Photonics Technology Letters, vol.6, no.2, Feb. 1994, pp [333] Janz CF, Keyworth BP, Allegretto W, Macdonald RI, Fallahi M, Hillier G, Rolland C. Mach-Zehnder switch using an ultra-compact directional coupler in a strongly-confining rib structure. IEEE Photonics Technology Letters, vol.6, no.8, Aug. 1994, pp [334] Rolland C, Moore RS, Shepherd F, Hillier G. 10 Gbit/s, 1.56 mu m multiquantum well InP/InGaAsP Mach-Zehnder optical modulator. Electronics Letters, vol.29, no.5, 4 March 1993, pp [335] Rolland C, Adams DM, Yevick D, Hermansson B. Optimization of strongly guiding semiconductor rib waveguide Y-junctions. IEEE Photonics Technology Letters, vol.2, no.6, June 1990, pp [336] Rolland C, Mak G, Fox KE, Adams DM, Springthorpe AJ, Yevick D, Hermansson B. Analysis of strongly guiding rib waveguide S-bends: theory and experiment. Electronics Letters, vol.25, no.18, 31 Aug. 1989, pp

140 [337] Bennett BR, Soref RA, Del Alamo JA. Carrier-induced change in refractive index of InP, GaAs and InGaAsP. IEEE Journal of Quantum Electronics, vol.26, no.1, Jan. 1990, pp [338] Zucker JE, Jones KL, Newkirk MA, Gnall RP, Miller BI, Young MG, Koren U, Burrus CA, Tell B. Quantum well interferometric modulator monolithically integrated with 1.55 um tunable distributed Bragg reflector laser. Electronics Letters, vol.28, no.20, 24 Sept. 1992, pp [339] ] Djurisic AB, Rakic AD, Kwok PCK, Li EH, Majewski ML. Modeling the optical constants of GaP, InP, and InAs. Journal of Applied Physics, vol.85, no.7, 1 April 1999, pp [340] Cahill LW, Payne FP. Optical switches based on the generalized Mach- Zehnder interferometer Digest of the LEOS Summer Topical Meetings. Electronic-Enhanced Optics. Optical Sensing in Semiconductor Manufacturing. Electro-Optics in Space. Broadband Optical Networks (Cat. No.00TH8497). IEEE. 2000, pp.iv [341] Brindel P, Leclerc O, Duchet C, Goix M, Grard E, Maunand E, Desurvire E. 20 Gbit/s optically regenerated transmission over 40 mm based on polarisation-independent, push-pull InP Mach-Zehnder modulator. [Conference Paper] 24th European Conference on Optical Communication. ECOC '98 (IEEE Cat. No.98TH8398). Telefonica. Part vol.1, 1998, pp vol.1. [342] O. Leclerc, P. Brindel, D. Rouvillain, E. Pincermin, B. Dany, E. Desurvire, C. Duchet, E. Boucheriz, S. Bourchoule, 40Gbit/s polarizationindependent, push-pull InP Mach-Zehnder modulator for all-optical regeneration PD35-1 [343] O. Leclerc, C. Duchet, P. Brindel, M. Goix, E. Grard, E. Maunand, E. Desurvire, Polarization-independent InP push-pull Mach-Zehnder modulator for 20Gbit/s soliton regeneration, Electronic Lett. Vol. 34, No. 10, May 14, [344] Bach HG, Krauser J, Nolting HP, Logan RA, Reinhart FK. Electro-optical light modulation in InGaAsP/InP double heterostructure diodes. Applied Physics Letters, vol.42, no.8, 15 April 1983, pp USA. [345] Adachi S, Oe K. Linear electro-optic effects in zincblende-type semiconductors: key properties of InGaAsP relevant to device design. Journal of Applied Physics, vol.56, no.1, 1 July 1984, pp

141 [346] Alping A, Coldren LA. Electrorefraction in GaAs and InGaAsP and its application to phase modulators. Journal of Applied Physics, vol.61, no.7, 1 April 1987, pp [347] Vinchant J-F, Cavailles JA, Erman M, Jarry P, Renaud M. InP/GaInAsP guided-wave phase modulators based on carrier-induced effects: theory and experiment. Journal of Lightwave Technology, vol.10, no.1, Jan. 1992, pp [348] Fiedler F, Schlachetzki A. Optical parameters of InP-based waveguides. Solid-State Electronics, vol.30, no.1, Jan. 1987, pp [349] Botteldooren D, Baets R. Influence of band-gap shrinkage on the carrierinduced refractive index change in InGaAsP. Applied Physics Letters, vol.54, no.20, 15 May 1989, pp [350] Van Eck TE, Walpita LM, Chang WSC, Wieder HH. Franz-Keldysh electrorefraction and electroabsorption in bulk InP and GaAs. Applied Physics Letters, vol.48, no.7, 17 Feb. 1986, pp USA. [351] Temkin H, Keramidas VG, Pollack MA, Wagner WR. Temperature dependence of photoluminescence of n-ingaasp. Journal of Applied Physics, vol.52, no.3, pt.1, March 1981, pp [352] Adachi S. Refractive indices of III-V compounds: key properties of InGaAsP relevant to device design. Journal of Applied Physics, vol.53, no.8, Aug. 1982, pp [353] Kingston R.H., Electroabsorption in GaInAsP, Appl. Phys. Lett. 34(11), 1 June [354] Dutta NK, Olsson NA. Electroabsorption in InGaAsP-InP double heterostructures. Electronics Letters, vol.20, no.15, 19 July 1984, pp UK. 124

142 Chapter 4 Series Push-Pull Modulator Designs As first demonstrated by Walker[7,455] and later Spickermann[456] in the GaAs/AlGaAs material system, if a RF signal is applied across the two MZ electrodes thereby connecting the diodes in series one achieves superior performance in terms of optical bandwidth and zero-chirp performance with a single RF input. This chapter will first look at the design concept and results from lumped device designs with input side termination. Next, end-terminated CPS transmission line electrode designs that aim to match the characteristic impedance are explored with respect to transmission line design and device characteristics. MZ#1 p-contact MZ#2 p-contact SiN x O y n-contact BCB InGaAs contact layer RF SIGNAL MZ #1 MZ #2 Vdc1 Vdc2 R SI InP substrate Figure 4-1 Series push-pull bias configuration 125

143 4.1 LUMPED SERIES PUSH-PULL BANDWIDTH The series push-pull biasing scheme and cross-section of the modulator are shown in fig The use of a SI substrate lowers the parasitic capacitance under the electrode pads and enables series push-pull (SPP) operation. Ideally, there would be full isolation between the n-contact region in the modulator and that of the rest of the chip. In the devices presented here, the n- InP region was etched down to the SI-substrate close to the ridge and the p- InP region was proton implanted. This leaves a narrow region below the shallow ridge that is not effectively isolated. The resistance between the n- contact to the modulator and that of the rest of the transmitter typically measured approximately 150 ohms. Lumped series push-pull devices were fabricated and tested using special 75m pitch CPS picoprobes with integrated 50 ohm parallel resistors. In this case, the 50 ohm termination is on the front end of the device. Later transmission line based devices use a rear termination. The 3dB optical bandwidth of three different devices with 200m, 250m and 300m electrode lengths are shown in fig. 4-2 and compared to the one-sided modulation. The probe configuration is shown in fig. 4-3 for the single side and SPP configurations. 126

144 GND SIGNAL GND SIGNAL Fig. 4-2 Biasing for single side and series push pull Note that using the series push-pull electrode structure almost doubles the optical bandwidth the loss due to parasitics to ground on one branch of the MZ which equates to approximately the capacitance from the n-contact to the backside of the substrate (100m thick). The substrate is metalized on the backside to facilitate soldering to a carrier for good heat conduction. Alternatively the device could be either flip-chipped without backside metalization or epoxied to the carrier thereby removing the metallization and reducing the parasitic capacitance at the expense of reduced thermal conductivity. Clearly a smaller n-contact region would be beneficial. One can see in fig 4-3b the capacitance per unit length (1225pF/m and 690pF/m for single side and SPP respectively) as the slope and pad capacitance as the y-intercept for both single side and SPP modulation. 127

145 1 f3 db RC Fig. 4-3a Comparison of single side and series push-pull 3dB optical small signal bandwidth Fig 4-3b Capacitance vs electrode length for single side and SPP configuration DC -3V Bias Due to the photocurrent generated in the devices the impedance is reduced considerably. This means that with a 50ohm parallel resistor, the single side configuration has approximately 45.5 ohms and the SPP configuration yields 47.6 ohms. The effect of added bandwidth is also evident in the back-to-back eye diagrams for the comparison of single side to SPP modulation as shown for a 250m long electrode device at 10 Gbit/s with a PRBS in fig m single side lumped 250m SPP Fig. 4-4 Back to back Eyes comparing single side and series push-pull operation with 10dB extinction. Both at using -2V DC bias with 1.5V Vpp. OC-192 with PRBS 128

146 4.2 DUAL RF SERIES PUSH-PULL DEVICES As mentioned before, the Dual RF series push-pull devices take advantage of the improved bandwidth of the SPP electrode structure and reduced voltage by having two of them. Figure 4-5 shows the device layout and the parasitic conduction path between the two n-contact regions. Ideally this path would be cut or reduced by He implanting in between the sets of electrodes. In the current layer structure this would be difficult as the n-inp and n-ingaas layers are approximately 2.3µm thick as well as the ridge on top (2µm) which is difficult to achieve without very high implant energies. To do this a quaternary contact would need to be placed much closer to the waveguide 0.5µm. DATA GND GND N-contact DATA N-contact Fig 4-5 Dual RF Series push-pull 4 electrode structure Due to the finite conductivity of the n-layer, the conduction path prefers the closest GND and even without He implantation, the device operates well at 10Gbit/s. 129

147 Fig 4-6 device layout for Dual RF SPP electrode devices As can be seen in fig. 4-7, the swing improves considerably with Dual RF sources. The bandwidth is compromised a bit due to the lack of isolation between the two n-contacts however not excessively as illustrated in the back-to-back eyes for single SPP and Dual SPP as shown in fig Both SPP Risetime: 72ps Falltime: 56ps One SPP Risetime: 65ps Falltime: 52ps Fig. 4-7 Optical signal levels for single and dual SPP operation and back-to-back eye diagrams for each with V pp = 2V with 10Gbit/s PRBS signal 130

148 4.3 TRAVELING WAVE MODULATORS Numerous groups have demonstrated discrete high-speed modulators and DFB integrated devices utilizing traveling-wave electrode structures [4,7]. Although lumped electrodes can provide fairly good performance with respect to bandwidth, careful design of the transmission line will provide superior return loss (S 11 ) if the loaded transmission line is designed to match the driver [25 or 50 ohms] and/or superior bandwidth if the microwave index is matched. An assortment of different transmission line structures have been pursued for traveling wave electrodes. Microstrip, Coplanar waveguide (CPW), and Coplanar strip (CPS) transmission lines are most often employed. Microstrip, although simple is sometimes regarded as disadvantageous due to inaccessible ground planes, difficulties in shunt connections between the strip and ground, limitations on the substrate thickness and exhibit more radiation with thick substrates. In the case of CPW lines the impedance is mostly defined by the lateral dimensions and the substrate thickness is not as important. CPW localizes the electric field reducing spurious coupling, radiation and dispersion[402]. Unfortunately, both the even and odd modes can exist in CPW which this odd mode can be suppressed with air bridges.[408] Additionally, parallel plate modes are supported (microstrip modes) between the CPW and the ground plane on the bottom which is a cause of energy leakage from the CPW 131

149 [408]. As a general rule the thickness of the substrate must be > 2(2G + W) in order to suppress the microstrip modes. In order to match the characteristic impedance of the source, the on-chip loaded transmission lines require fairly large unloaded characteristic impedances. Although CPW can easily be made to match 50ohms, capacitively loaded lines require much larger unloaded characteristic impedances to yield 50 ohms loaded and these high values cannot be realized in CPW easily with the current doping restraints of the integration platform. CPW designed for index matching yields poor characteristic impedance matching. However, Coplanar Stripline (CPS) which has a range of possible Z o values twice that of CPW works well for the matching region. CPS was chosen for this reason, and the compactness of the transmission lines suitable for further integration such as in a photocurrentdriven wavelength converter. However, it is more difficult to make a 50 ohm unloaded section (for the feedthroughs) without very narrow gaps and wide pads leading to higher microwave attenuation. The feedthroughs were designed at a linear taper as this was found to be the best approach in [462]. Also, the phase difference between the two lines will affect the matching ability at higher frequencies. Ideally the lines should be excited with equal length feedthrough lines. The design was chosen to have an input line at 30 degrees to allow probing away 132

150 from the optical waveguide but minimize the phase difference between the two electrodes. 4.4 TRAVELING WAVE MATCHING The design of Traveling-Wave (TW) modulators is based on the matching of the optical and electrical wave velocities. As has been pointed out[454], it is the group index that should be matched not the phase velocity. In the case of LiNbO 3 modulators, the electrical wave (n eff 4.225) propagates slower than the optical wave(n eff 2.138)[303,421,424]. To perform matching, one can use a buffer layer, phase reversal, or a shielding plane to decrease the microwave effective index of the line[422,423]. Alternatively, one can increase the electrode thickness, decreasing the effective index further [421]. GaAs and InP, modulators can have electrical waves that propagate faster than the optical wave. In order to match the index, often either capacitive coupling or inductive coupling approaches are employed. According to work done by Spickermann et al.[461], the inductively coupled slow wave structures have higher attenuation loss for a given gap width and are harder to model than capacitively-coupled devices. LiNbO 3 devices do not have a PN structure and do not load the line substantially similarly to devices such as demonstrated by Spickermann that rely on the electric field between the electrodes to change the index which usually is far less efficient than the use of a PN structure. The devices in this dissertation use PN junctions to increase the electric field 133

151 overlap with the optical mode which leads to a very large capacitance per unit length resulting in a similar situation as the LiNbO 3 where the microwave index is much higher than the optical index. For a capacitively loaded transmission line, the optimum loading capacitance is given by Walker: C loading n 2 opto cz o n n 2 cpw opto [4.1] where c is the speed of light, n opto is the optical group index, n cpw is the electrical index and Z o is the characteristic impedance However, in order to fabricate high performance SGDBRs, the doping required typically results in capacitance per unit lengths in the range of 2000pF/m to 2500pF/m for a 3m wide ridge. The junction capacitance/length of the device due to the PN or PIN region is considerably larger (x10) than the capacitance/length of the coplanar line. The result of this is the line is highly capacitively coupled which both slows the electrical wave and reduces the characteristic impedance considerably. First, the optical group index of the modulator section needs to be assessed. The effective index and group index are shown in fig. 4.8 for various waveguide compositions. 134

152 Effective Index Q 1.35Q 1.4Q 1.45Q Group Index Wavelength (nm) 1.3Q 1.35Q 1.4Q 1.45Q Wavelength (nm) Fig. 4-8 Effective Index and Group index for different waveguide compositions. Assuming a structure where the waveguide has been etched off halfway. One can see that not only is the group index significantly higher for waveguide compositions at approx but the dispersion increases as the operating wavelength approaches that of the band-edge. One will obtain a superior velocity match at the lower wavelengths and higher waveguide Q as loaded transmission lines tend to slow the electrical wave excessively. Matching over a wide wavelength range becomes more difficult as the waveguide composition Q increases. 135

153 Next we should consider the group index of the microwave signal. The electrical signal does not have as much dispersion as the optical signal and is often approximated as just the phase velocity. The dispersion has been curve fitted from spectral domain data and is given by [415]. n eff ( f ) eff ( f ) q r1 (1 af b q ) [4.2] f TE c 4h1 r1 1 [4.3] where: f = frequency; F = f/f TE the cutoff frequency for the lowest-order TE mode S ( u log v) W a 10 [4.4] u q q 2 v q q 2 q = log(s/h 1 ) h 1 thickness of substrate b = 1.8 q = effective permittivity at the quasi-static limit. 136

154 4.5 TRANSMISSION LINE MODEL The CPS transmission line in these series push-pull devices can be modeled as a distributed circuit model along the device as shown in fig Often, the characteristic impedance of a transmission line is approximated for low microwave loss in equation 4.5. Z o Z L lossless cps [4.5] Y Ccps n c ZY c lossless cps cps [4.6] L C However, the devices here experience microwave losses due to a number of sources as outlined in section 4.7 and the transmission line model fits the data best if the capacitive and inductive loading are accounted for in the model. The transmission lines in the device are loaded by the depletion capacitance from each ridge C PN1 and C PN2 in the ridge as is shown in fig. 4-9 as well as a small amount of inductance due to the T structures. The capacitance shown in fig. 4-9 is composed of the PN junction capacitance (significant), the CPS metallization capacitance and the parasitic capacitance. 137

155 C cps C PN G C PN Fig. 4-9 Device cross-section equivalent circuit for smooth CPS This can be expressed as a distributed circuit model as shown in fig Z R L cps L T Y G PN C PN G n C cps G PN2 C PN2 C Para L T Transmission line Equivalent circuit for T-electrode CPS line Fig Transmission line distributed equivalent circuit. G n is conductance in n-cladding region, C para is parasitic capacitance to ground, L T is the T-inductance, G PN is the conductance due to the photocurrent in the ridge C pn is the depletion capacitance 138

156 The equivalent circuit model for the smooth CPS line devices is the same as given in fig 4-10 except without the inductance contribution of the Ts (L T = 0). Given the equivalent circuit model in fig. 4-10, the characteristic impedance of the smooth CPS transmission line can be expressed as: L cps j R Z [4.7] p PN PN n n cps smooth C C C j G G C j Y [4.8] p PN PN n n cps cps smooth osmooth C C C j G G C j L j R Y Z Z ) ( [4.9] The T structures have some additional inductance as shown in the equivalent circuit in fig p PN PN PN PN T n cpst T C C j G C j G L j G C j Y [4.10] p PN PN PN PN T n cpst cpst T ot C C j G C j G L j G C j L j R Y Z Z ) ( [4.11] 139

157 4.6 CHARACTERISTIC IMPEDANCE COMPARISON The CPS lines used in this dissertation were modeled using ADS software. As the lines are considerably capacitively loaded, this means we need to design a transmission line that has a much larger impedance unloaded in order to obtain 50ohms loaded. Figure 4.11 shows the unloaded characteristic impedance for two CPS structures, one with 50 m Ts and one with a smooth CPS line 16m apart with 8m wide strips Unloaded Characteristic Impedance for smooth CPS and 50m T structures from devices as shown in table m thick Au As can be seen in fig. 4-12a, narrow lines increase the characteristic impedance by increasing the inductance at the expense of microwave loss. 140

158 A much higher characteristic impedance is possible with the Ts as shown in fig 4-12b for a given electrode width. The width of the T electrodes was chosen at 8µm as a compromise with microwave loss shown in fig 4-12b as design G. Characteristic Impedance um spacing 1000pF/m loading Frequency (GHz) Loaded Characteristic Impedance Design G 5um DesignG 8um DesignG 15um Frequency (GHz) Fig 4-12a Characteristic Impedance for different CPS line widths given a 16um spacing Fig 4-12b Characteristic impedance for T-section electrodes vs electrode width From S parameters and the resulting [ABCD] matrix, the characteristic impedance was extracted for different biases for device #9. After testing the characteristic impedance of the different devices it was clear that they don t fit the characteristics shown at low frequencies in fig 4-12b. After analyzing the expected conductance in the structure, it was obvious that the n-epilayer conductance for this structure is considerably higher than that of previous structures done on lower doped or dielectric substrates. 141

159 Based on Hall measurements: The conductivity of the n-layer between the ridges is given by InP( n) q n n = (1.6E-19 C)(1800cm 2 /Vs)(1E18 1/cm 3 ) = 288 S/cm (0.032S/cm in Spickermann) where the Conductance is Area GInP( n) 2 InP( ) = S (compare with 0.01S in Spickermann) Length Where the Length is 16µm; Area = (2.3µm*314µm) for Device #9 Data was taken comparing the characteristic impedance of T structures, smooth CPS lines and lumped rear terminated electrode devices. The fit from the model shown in equations 4.9 and 4.11 are also shown assuming for the Ts the capacitance per unit length of the transmission line is C T = 2.737e-11F/m, Inductance per unit length is L T = 1.736e-6 H/m and for the smooth CPS C s = 4.602e-11F/m and L s = e-7H/m with R pn = 500ohms, R = 0.2 ohms, C pn = 0.5pF, C para = 0.5pF, L T = 5e

160 Fig 4.13 Lower 4 lines extracted from Device #7 (single side 300m long electrode) Middle 4 lines extracted from Device #18 (Smooth CPS 500m long electrode) Top 4 lines extracted from Device #7 (CPS Ts 250m long electrode) As can be seen in fig 4-13, the characteristic impedance improves for higher reverse biases on the electrodes where the depletion region is increased and the capacitance/unit length decreases. Also, clearly one can see a large benefit of using a T electrode over the smooth CPS lines in terms of better characteristic impedance matching as it is much closer to 50 ohms. 143

161 4.7 RF LOSS MECHANISMS High frequency losses stem from three different mechanisms: 1. Dielectric losses 2. Ohmic/conductor losses 3. Radiation loss These losses can be minimized using a number of approaches such as the use of deep trenches between electrodes or thick dielectric layers below the electrodes separating them from the substrate. By careful design of the electrodes, minimization of longitudinal substrate currents may also reduce the overall microwave attenuation[435]. Most work is done on Semi-insulating InP and GaAs where the bulk of the electrical attenuation comes from the conductor and radiation losses at least below 20GHz [450]. However, typical SGDBR design is performed on n-inp conducting substrates with lossy InGaAs contact layers. In this case, the dielectric losses are very high and the lines become highly dispersive. Also, the capacitance between the two lines increases dramatically - effectively loading the line and dropping the characteristic impedance significantly. Dielectric loss is given by the following relationship: D q r eff tan g (Np/m) *27.3 for db/lamda [4.12] 144

162 For a doped semiconductor the loss tangent can be expressed as[450]: 2f "( f ) '( f ) tan ( f ) [4.13] 2f '( f ) "( f ) where and are the real and imaginary parts of the complex dielectric permittivity and and are the respective parts of the complex conductivity. Taking the conductivity from the Drude model, we have = s /(1-j2f m ) where the conductivity can be extracted from Hall measurements. The attenuation drops linearly with increasing metal thickness up to the point where the metal depth is 3x skin depth. As the dimensions of the transmission line increase, the attenuation decreases. There seems to be an optimum w/d point of approximately 0.40 for InP with 0.25um of gold. If w = 80um that corresponds to d = [405] In order to match the velocities of the electrical and optical waves, one can manipulate a few parameters electrode thickness, coplanar gap width, and distributed capacitance along the line. The electrode thickness highly affects the microwave loss in the structure as shown in fig

163 Loss (db/cm) 5um Loss (db/cm) 10um Loss (db/cm) 20um 12 Microwave Index nload (5um) nload (10um) nload (20um) 10 8 Loss (db/cm) Electrode Thickness (um) Fig Microwave index and loss for loaded CPS line [2000pF/m loading] with different electrode widths varying from 5-20 microns Clearly an electrode thickness exceeding 2µm is preferable to reduce both the microwave index and loss. For the work shown here, the p-metal thickness is approximately 1.5m. The loaded-microwave index drops significantly with electrode thickness although as we have shown before, we would like This does not take into account the change in effective index when the area between the center conductor and ground are removed or BCB is placed below the contacts. Although thickening the electrode improves the index match, the characteristic impedance is reduced. In order to match the characteristic impedance and index simultaneously, the capacitance per unit length of the line must be reduced. 146

164 Table 4-1 Material Relative Permittivity Loss Tangent (tan ) Typical Resistivity InP n 12.4,12.6 5x E-4 ohm-cm InP SI E7 ohm-cm GaAs x10-4 High resistivity Si x10-4 at 30GHz 4000ohm-cm Standard Si x10-3 at 30GHz 1000ohm-cm BCB 2.65 Aluminum Nitride 8.9 Conductor loss can be estimated from the unloaded Q factor Q u w Z o f 2 [4.14] Good up to 2 GHz above 2 GHz one must keep in mind the dielectric attenuation is mostly dependent upon the substrate thickness. The conductor attenuation coefficient is minimized at a particular w/s. This conductor attenuation decreases with dielectric constant. The dispersion is lower for smaller waveguide dimensions. The coplanar waveguide design takes into account the dielectric that the lines reside, the thickness of the metal layer to achieve a 50 ohm line. As the lines are deposited on a multilayer dielectric not just on the InP surface, one needs to take into consideration the effective dielectric constant that insues. This can be calculated analytically using the conformal mapping technique.[407] 147

165 As discussed earlier, conductor losses are reduced by using wider and thicker electrodes. However, the characteristic impedance is improved by going to thinner and narrower electrodes. A compromise was made using 8m wide electrodes. The gap between the two ridges was designed to be fairly close (16m) to shorten the curved waveguides and reduce propagation losses. LiNbO 3 traveling wave modulators usually use CPW transmission lines as without careful attention to the electrode gap widths have experienced large RF losses in CPS structures [453]. It has been found [453] that leakage of the CPS modes into substrate modes may occur at fairly low frequencies (11 and 22GHz). This leakage is due to the geometry of the device (gaps 0.5mm to 1mm on substrates mm thick) The electrical attenuation was measured for different biases without bias on the laser or SOA. The loss was extracted from the ABCD parameters [see appendix] after measuring the S parameters of the device. These values compare closely with other similar EAM devices that report losses in the range 15-20dB/mm at 40GHz. The microwave loss results with bias are shown in fig As can be seen, the microwave loss decreases considerably with 148

166 reverse bias. This has been attributed to loss due to undepleted material in the PN junctions 25. Fig 4-15 Microwave loss as a function of frequency and bias from Device #9 As the PN junction depletes out, the loss becomes dominated by ohmic losses due to the skin depth in the electrodes. 25 Spickermann Dissertation pp

167 4.8 CPS T-ELECTRODE DEVICES Much work has been done to velocity match traveling wave Mach Zehnder structures using T-sections that both increase the characteristic Impedance and the length thereby reducing the capacitance per unit length and providing better matching that results in higher bandwidth. SGDBR Laser MZ Phase electrode Modulator n-contact GeAuNiAu n-contact Semiconductor Optical Amplifier Fig T-Electrode SPP-MZ-SOA-SGDBR Transmitter Layout The device layout of these devices is shown in fig above. 150

168 By distributing the capacitance using fins, one can lower the capacitance per unit length at will. However, as the InP/InGaAsP material has considerable optical loss, and the mismatch becomes greater between the optical and electrical waves at longer lengths, the Ts designed in this work did not lengthen the device much. The periodicity of the tabs is related to the cutoff frequency for a given phase velocity and width[304]. f cutoff v 2d phase [4.15] where d is the spacing of the fins (period) and v phase is the phase velocity. These Ts are 50µm long with 10µm spacing between as shown in fig Fig 4-17 TW electrode structure with 50µm Ts with 10µm gaps 151

169 This approach is only practical when the capacitance per unit length is already small. For highly capacitively loaded lines the device length that is required in order to improve the bandwidth is very large leading to excessive microwave and optical insertion losses. In this work a few different CPS transmission line electrode designs were explored as shown in table 4-2. Table 4-2 Transmission Line based electrode MZ devices using Dual RF Series Push-pull drive Total # SOA Config SOA Length Active electrode length Electrode Length (um) Electrode Width T length (number) 9 Single (5) 10 Dual (8) 11 Dual (10) 12 Dual (12) 16 Single (8) 17 Single N/A 18 Single N/A 19 Single (5) 20 Single (10) 21 Single (12) 152

170 4.9 TRAVELING WAVE BANDWIDTH The bandwidth of a traveling wave modulator is governed by the difference in the optical and electrical waves and the overlap factor of these two modes, the frequency dependent attenuation along the device, the termination impedance, and the length of the device. Accounting for both the attenuation and the optical-electrical matching and assuming that the device is terminated with the characteristic impedance the bandwidth can be approximated as[4]: B( f ) e sinh / 2 l 2 l l 2 l sin l 2 where 2f [ n no ] [4.16] c It is clear from the previous equation that both the attenuation and index matching are very important to achieve high bandwidth. Although simple, the above equation does not take into account mismatches in the characteristic impedance which is important as it is difficult to reach 50 ohms with such high loading capacitance. The small-signal modulation response S 21 can be modeled accounting for the opto-electrical velocity mismatch, microwave attenuation, and impedance mismatch[5]. 153

171 2 21 ) ( 1 ) ( exp( ) 2 exp( ) ( 1 ) ( exp ) 2 exp( 1 L j j L L j L j L L T S L S L [4.17] where the amplitude transmission into the modulator is : T, The reflection coefficients at the source and the load are given by: 1 S ) ( ) ( m s m s S Z Z Z Z [4.18] ) ( ) ( m L m L L Z Z Z Z [4.19] Z m is the characteristic impedance of the modulator, and Z s and Z L are the impedances of the source and load respectively. The propagation constant is given as: n 2 [4.20] and the optical Beta coefficient is: o n o 2 [4.21] where n o is the optical group index, and is the optical wavelength 4.10 MEASURED BANDWIDTH Taking the fit data from the characteristic impedance for device 9 as shown in fig. 4.16, the attenuation coefficient, and microwave index, and assuming the 154

172 optical group index is 3.92 at 1555nm, the model expressed in equation [4.11] fits the experimental data well. Fig 4-17 shows the small-signal frequency response of two devices, (9 and 16) which have 250µm and 400µm long electrodes. Both are terminated at the rear with 50 ohms. Fig Small-signal frequency response for two T-electrode devices at 1555nm -3V bias for a 250m electrode device (device #9) and 400m device (device #10) The bandwidth was also explored with a low matching resistor. This was done with a resistor ladder structure that was fabricated on the carrier adjacent to the device as shown in fig

173 Phase electrode RF Signal Input Termination Resistor n-contact bias Fig Transmission line electrode configuration Device 17 The S 21 response with a 35 ohm termination resistor with wirebonds as in fig 4-19, is shown in fig This is compared with the bandwidth of a lumped electrode device with the same length and the 50 ohm terminated data. As one might expect, the bandwidth is extended out to close to 40GHz and some peaking is observed due to the reflections/standing wave along the electrode structure. 156

174 Fig 4.20 Small-signal response for Device #9 as a function of termination resistor As can be seen in fig 4.20, the data fits the traveling wave bandwidth model very well for the 50ohm termination. In this case, a 50ohm probe was used at the end of the device with a 50ohm termination. The 35ohm termination data was obtained using wirebonds to the carrier in which a resistor ladder was used. One sees a little more peaking than the model predicts due to the wirebond impedance discontinuity which leads to a reflection at the end of the device. Note also that there is very little difference in the bandwidth observed between the counter-propagating and co-propagating bias configurations. Most likely the traveling wave effect would be more noticeable for longer 157

175 devices with low termination resistors as has been seen in the literature, however, this could not be done due to the restraints of the resistor ladder. REFERENCES [400] Jaeger NAF, Rahmatian F, James R, Berolo E. An AlGaAs/GaAs Mach- Zehnder modulator using slow-wave coplanar electrodes. Conference Proceedings. IEEE Canadian Conference on Electrical and Computer Engineering (Cat. No.98TH8341). IEEE. Part vol.2, 1998, pp vol.2. [401] Spickermann R, Dagli N. Experimental analysis of millimeter wave coplanar waveguide slow wave structures on GaAs. IEEE Transactions on Microwave Theory & Techniques, vol.42, no.10, Oct. 1994, pp [402] Monolithic Microwave Integrated Circuits Technology and Design, Ravender Goyal Pg 224 [403] Jaeger, N.A.F, F. Rahmatian, H. Kato, R. James, E. Berolo, Z. K. F. Lee, Velocity Matched Electrodes for Compound Semiconductor Travelling wave electro-optic modulators: Experimental Results, IEEE Microwave and guided wave letters, Vol. 6., No. 2. Feb [404] Ponchak G.E., M. Matloubian, L.P.B. Katehi, A Measurement-Based Design Equation for the Attenuation of MMIC-Compatable Coplanar Waveguides, IEEE. Trans. On Microwave Theory and Techn. Vol 47, No.2, Feb, 99 [405] Haydl WH, Braunstein J, Kitazawa T, Schlechtweg M, Tasker P, Eastman LF. Attenuation of millimeterwave coplanar lines on gallium arsenide and indium phosphide over the range 1-60 GHz IEEE MTT-S International Microwave Symposium Digest (Cat. No.92CH3141-9). IEEE. 1992, pp vol.1. [406] Khazaeil HR, James R, Berolo E, Rahmatian F, Jaeger NAF, Ghannouchi F. Novel coplanar-strip slow-wave structure for ultrawide-bandwidth 158

176 electro-optic modulators. Microwave & Optical Technology Letters, vol.15, no.5, 5 Aug. 1997, pp [407] Bedair SS, Wolff I. Fast, accurate and simple approximate analytic formulas for calculating the parameters of supported coplanar waveguides for (M)MIC's. IEEE Transactions on Microwave Theory & Techniques, vol.40, no.1, Jan. 1992, pp [408] Jackson RW. Mode conversion at discontinuities in finite-width conductor-backed coplanar waveguide. IEEE Transactions on Microwave Theory & Techniques, vol.37, no.10, Oct. 1989, pp [409] Hilberg W. From approximations to exact relations for characteristic impedances. IEEE Transactions on Microwave Theory & Techniques, vol.mtt-17, no.5, May 1969, pp [410] Ponchak GE, Matloubian M, Katehi LPB. A measurement-based design equation for the attenuation of MMIC-compatible coplanar waveguides. IEEE Transactions on Microwave Theory & Techniques, vol.47, no.2, Feb. 1999, pp [411] Yip P.C.L, High-Frequency Circuit Design and Measurement, Chapman and Hall, pg [412] Giboney K.S., M. J. W. Rodwell, J. E. Bowers, Travelling-Wave Photodetector Theory, IEEE. Trans. on Microwave theory and tech., Vol 45, No.8., August, [413] Zhang S.Z., Traveling wave Electro-absorption modulators UCSB Dissertation 1996 [414] Ludwig R., P. Bretchko, RF Circuit Design Theory and Applications, Prentice Hall, [415] Simons R.N., Coplanar Waveguide Circuits, Components and Systems, Wiley Series in Microwave and Optical Engineering, [416] Giboney K.S., Travelling-Wave Photodetectors, IEEE. Photonics Tech. Letts., vol. 4 No. 12, Dec [417] Heinrich W. Quasi-TEM description of MMIC coplanar lines including conductor-loss effects. IEEE Transactions on Microwave Theory & Techniques, vol.41, no.1, Jan. 1993, pp

177 [418] Collins R.E., Foundations for Microwave Engineering Second Ed. IEEE press series on Electromagnetic wave theory Donald G. Dudley editor, [419] Kiziloglu K, Dagli N, Matthaei GL, Long SI. Experimental analysis of transmission line parameters in high-speed GaAs digital circuit interconnects. IEEE Transactions on Microwave Theory & Techniques, vol.39, no.8, Aug. 1991, pp [420] Dinesen P.G., A. Bjarklev, T. Rasmussen, C. Lester, Microwave properties of a traveling wave modulator in polymers employing a microwave buffer layer for velocity matching, Electronic Letters, Vol. 31, No. 2, 19 th, Jan [421] Jin H, Belanger M, Jakubczyk Z. General analysis of electrodes in integrated-optics electrooptic devices. IEEE Journal of Quantum Electronics, vol.27, no.2, Feb. 1991, pp [422] Zhang X., T. Miyoshi, Optimum Design of Coplanar Waveguide for LiNbO3 Optical Modulator, IEEE. Trans. on microwave Theory and Tech. Vol. 43., No.3, March [423] Boyu W., X. Guangjun, J. Xiaomin, Travelling wave electrode optimization for high speed electro-optic modulators using the Fourier Series Method, IEE. Proc.-Optoelectron, Vol. 141, No.6, Dec [424] Gopalakrishnan G.K., W. K. Burns, R. W. McElhanon, C. H. Bulmer, A. S. Greenblatt, Performance of Modeling of BroadBand LiNbO3 Traveling wave Optical Intensity Modulators, J. Lightwave tech., vol.12, No.10, Oct [425] Cho W., Y. Lim, Y. Choi, Bandwidth Limitation Factors and Linear Characteristics of Travelling-wave CPW Electro-absorption Modulators, MWP 99 Digest. [426] Hong I., H. Park, Quasi-static analysis of traveling wave electrodes with asymmetrical finite thickness for electro-optic modulators, IEE. Proc. Optoelectron. Vol No.2, April [427] Svacina J. Special types of coplanar transmission lines suitable up to MM-wave bands. Electrical Performance of Electronic Packaging (Cat. No.97TH8318). IEEE. 1997, pp

178 [428] Takuchi H., High Speed Electroabsorption modulators with Travelling- Wave Electrodes, OFC [429] Liou JC, Lau KM. Analysis of slow-wave transmission lines on multilayered semiconductor structures including conductor loss. IEEE Transactions on Microwave Theory & Techniques, vol.41, no.5, May 1993, pp [430] Kawano K., M. Kohtoku, M. Ueki, T. Ito, S. Kondoh, Y. Noguchi, Y. Hasumi, Electrode electroabsorption (TW-EA) modulator with bandwidth over 50 GHz and driving voltage less than 2V, Electronic Lett. Vol. No [431] Rao HG, Liu QZ, Kellner AL, Jiang XS, Lau SS, Yu PKL. Microwave characteristics of coplanar waveguides on helium-implanted epitaxial p- InP. Applied Physics Letters, vol.66, no.15, 10 April 1995, pp [432] Mesa FL, Cano G, Medina F, Marques R, Horno M. On the quasi-tem and full-wave approaches applied to coplanar multistrip on lossy dielectric layered media. IEEE Transactions on Microwave Theory & Techniques, vol.40, no.3, March 1992, pp [433] Baker-Jarvis J, Vanzura EJ, Kissick WA. Improved technique for determining complex permittivity with the transmission/reflection method. IEEE Transactions on Microwave Theory & Techniques, vol.38, no.8, Aug. 1990, pp [434] Kim, I., M. R.T. Tan, S-Y Wang, Analysis of a New Microwave Low-loss and Velocity Matched III-V Transmission Line for Traveling wave Electrooptic Modulators J. Lightwave Tech. Vol. 8, No.5, May [435] Berg H, Gevorgian S. The effect of longitudinal substrate currents on the losses in silicon substrate coplanar-strip and coplanar waveguides. International Journal of RF & Microwave Computer-Aided Engineering, vol.10, no.5, Sept. 2000, pp [436] Zhang SZ, Yi-Jen Chiu, Abraham P, Bowers JE. Polarization-insensitive multiple-quantum-well traveling-wave electroabsorption modulators with 18 GHz bandwidth and 1.2 V driving voltage at 1.55 um. International Topical Meeting on Microwave Photonics. Technical Digest (including High Speed Photonics Components Workshop) (Cat. No.98EX181). IEEE. 1998, pp

179 [437] Koshiba M., Y. Tsuji, Design and modeling of microwave photonics devices, Optical and Quantum Electronics, 30(1998) pp [438] Goverdhanam K, Simons RN, Katehi LPB. Coplanar stripline propagation characteristics and bandpass filter. IEEE Microwave & Guided Wave Letters, vol.7, no.8, Aug. 1997, pp [439] Goverdhanam K, Simons RN, Katehi LPB. Coplanar stripline components for high-frequency applications. IEEE Transactions on Microwave Theory & Techniques, vol.45, no.10, pt.1, Oct. 1997, pp [440] Groteluschen Enno, S. L. Dutta, and S. Zaage, Quasi-analytical Analysis of the Broadband Properties of Multiconducter Transmission Lines on Semiconducting Substrates IEEE. Trans. on Components Packaging and Manufacturing Tech. Part B., Vol. 17, No. 3, Aug [441] Simons RN, Dib NI, Katehi LPB. Coplanar stripline to microstrip transition. Electronics Letters, vol.31, no.20, 28 Sept. 1995, pp [442] Goverdhanam K, Simons RN, Dib N, Katehi LPB. Coplanar stripline components for high frequency applications IEEE MTT-S International Microwave Symposium Digest (Cat. No.96CH35915). IEEE. Part vol.2, 1996, pp vol.2. [443] Durr,W., U. Erben, A. Schuppen, H. Dietrich, H. Schumacher, Investigation of Microstrip and Coplanar Transmission Lines on Lossy Silicon Substrates Without Backside Metalization, IEEE. Trans. of Microwave Theory and Tech., Vol. 46, No. 5, May [444] Morl L., D. Hoffmann, K. Matzen, C. Bornholdt, G. G. Mekonnen, F. Reier, Travelling wave electrodes for 50 GHz Operation of Opto- Electronic devices based on InP 11 th International Conf. on Indium Phosphide and Related Materials May [445] Walker RG, Edge C. A folded-path GaAs travelling-wave modulator for phased array receivers. International Topical Meeting on Microwave Photonics MWP 2000 (Cat. No.00EX430). IEEE. 2000, pp [446] Ding XZ, Lu Taijing, Ong CK, Tan BTG. Effects of impurity concentration on dielectric loss in Zn-doped InP at microwave X-band frequencies. Journal of Applied Physics, vol.75, no.11, 1 June 1994, pp

180 [447] Jae-Kyung Wee, Young-June Park, Hong-Shick Min, Dae-Hyung Cho, Man-Ho Seung, Hun-Sug Park. Modeling the substrate effect in interconnect line characteristics of high-speed VLSI circuits. IEEE Transactions on Microwave Theory & Techniques, vol.46, no.10, pt.1, Oct. 1998, pp [448] Fuster JM, Marti J, Candelas P. Modeling Mach-Zehnder LiNbO/sub 3/ external modulators in microwave optical systems. Microwave & Optical Technology Letters, vol.30, no.2, 20 July 2001, pp [449] Tan MRT, Kim I, Chang J, Wang SY. Velocity matching of III-V travellingwave electro-optic modulator structures. Electronics Letters, vol.26, no.1, 4 Jan. 1990, pp [450] Joo-Hiuk Son, Hsi-Huai Wang, Whitaker JF, Mourou GA. Picosecond pulse propagation on coplanar striplines fabricated on lossy semiconductor substrates: modeling and experiments. IEEE Transactions on Microwave Theory & Techniques, vol.41, no.9, Sept. 1993, pp [451] Haydl WH. Properties of meander coplanar transmission lines. IEEE Microwave & Guided Wave Letters, vol.2, no.11, Nov. 1992, pp [452] Holloway CL, Kuester EF. A quasi-closed form expression for the conductor loss of CPW lines, with an investigation of edge shape effects. IEEE Transactions on Microwave Theory & Techniques, vol.43, no.12, pt.1, Dec. 1995, pp [453] Gopalakrishnan GK, Burns WK, Bulmer CH. Electrical loss mechanisms in travelling wave LiNbO/sub 3/ optical modulators. Electronics Letters, vol.28, no.2, 16 Jan. 1992, pp [454] Spickermann R, Sakamoto SR, Dagli N. In traveling wave modulators which velocity to match? Conference Proceedings. LEOS '96 9th Annual Meeting. IEEE Lasers and Electro-Optics Society 1996 Annual Meeting (Cat. No.96CH35895). IEEE. Part vol.2, 1996, pp.97-8 vol.2.. [455] Walker RG. Broadband (6 GHz) GaAs/AlGaAs electro-optic modulator with low drive power. Applied Physics Letters, vol.54, no.17, 24 April 1989, pp [456] Spickermann R., Sakamoto, S.R., Peters, M. G., Dagli, N. GaAs/AlGaAs traveling wave electro-optic modulator with an electrical bandwidth >40GHz Electron. Lett., vol. 32, no. 12., June 1996 pp

181 [457] Edwards T.C., M.B. Steer, Foundations of Interconnect and Microstrip Design, Third Edition, John Wiley and Sons LTD, New York. [458] Simons RN, Dib NI, Katehi LPB. Modeling of coplanar stripline discontinuities. IEEE Transactions on Microwave Theory & Techniques, vol.44, no.5, May 1996, pp [459] Lei Zhu, Ke Wu. Field-extracted lumped-element models of coplanar stripline circuits and discontinuities for accurate radiofrequency design and optimization. IEEE Transactions on Microwave Theory & Techniques, vol.50, no.4, April 2002, pp [460] Jager D, Kremer R, Stohr A. Travelling-wave optoelectronic devices for microwave applications IEEE MTT-S International Microwave Symposium Digest (Cat. No.95CH3577-4). IEEE. Part vol.1, 1995, pp vol.1. [461] Spickermann R. High speed Gallium Arsenide/Aluminium Gallium Arsenide traveling wave electo-optic modulators UCSB Dissertation

182 Chapter 5 COMPARISON OF DEVICE DESIGNS This chapter compares the performance of the devices outlined in the previous two chapters with respect to modulation efficiency, bandwidth and chirp characteristics. This work was done on the test setup as shown in fig. 5-1 with devices soldered down to RF Aluminum Nitride carriers. Fig. 5-1 Test setup and RF carrier 165

183 A number of authors have proposed figures of merit for modulators of both the EAM and Mach-Zehnder varieties in Table 5.1. These merits consider the electrical bandwidth, drive power requirements, low return loss, optical insertion loss, and suitability for integration with lasers. Wavelength and temperature dependence are also a consideration. The contrast ratio is important for large signal modulation. Often Mach-Zehnder and EAM devices, are compared as V pi with V 10dB. This is not usually a fair comparison however as the MZ modulator does not need to driven so hard. Table 5.1- Figures of Merit Based on R. G. Walker [7] 2R 50 R fo V Modulation Bandwidth and generator voltage. GHz-um/V R is shunt resistance M.K.Chin[9] T ; o ; 2 E Change in Transmission Optimizing the device for the largest Bandwidth requires the consideration of a V Drive number of parameters as in Table 5.2. The bandwidth is a function of the capacitance(length), S11(Z o ), and Termination/Resistance. The drive voltage is a function of the Insertion loss/power, Doping(Length), Waveguide composition, and termination. 166

184 Table 5.2 Main Design Parameters Bandwidth Doping Device Length Device Width Passivation Dielectric constant Waveguide[Q] Bandwidth Insertion Loss Drive Voltage Wavelength Dependence N/A N/A N/A N/A N/A N/A for TW For analog applications, the modulation depth is low and RF gain is of utmost concern, the modulator maximum phase shift per unit length per volt is a fairly good metric. V a L [3] One can also compare devices simply as the change in transmission divided by the voltage to make this change (db/v).[8] This value changes quite a bit over the modulation extinction curve, however could be compared at V pi /2. Also, MZ modulators often are compared with phase shift efficiency and the chirp parameter. 167

185 Phase shift efficiency neff 2 [5.1] VL V neff ( real) 2 Chirp Parameter chirp [5.2] n ( imag) L eff For InGaAs MQW structures efficiencies of approximately 12 deg/vmm have been reported [3]. In a real digital communications application, a device would need to exceed a set of specifications typically for output power, drive voltage, RF extinction, and chirp parameter. It would be best to compare different devices with respect to a particular output power and extinction (10dB RF). This is difficult to compare. As in this case as each design has a different output power and none of them meet a 10dBm fiber coupled output power specification. The output power has a large influence on the required drive voltage as illustrated in fig

186 Fig 5-2 Reduction in drive voltage for high output power. Assumes 0dBm max power required with -20dBm off state power For a given output power specification (in this case 0dBm), if the transmitter can output more power than necessary, the drive voltage can be significantly lower than V pi. In fact, for a doubling of output power either by increasing the gain of the SOAs/SGDBR or reducing the insertion loss in the modulator, the drive voltage is only ½ of the V pi value. This point is rarely considered in figures of merit. As the laser is integrated and cannot be simply swapped out for a higher power laser, high output power is very important. Because of this, V pi overestimates the actual drive voltage requirements in some cases considerably. However, it does give a metric to compare different 169

187 modulator designs more independently from the output power so DC extinction was measured for each device to base a comparison. 5.1 MODULATION EFFICIENCY The different devices are compared with respect to V pi as a function of input power into each branch of the MZ (measured from photocurrent as seen in Chapter 3) and wavelength. As can be seen in fig. 5-1A&B, the drive voltage is highly dependent on optical power. As the optical power increases, photon induced carrier recombination occurs in the modulator electrode region which generates heat and the localized waveguide bandgap shrinks giving enhanced absorption and index change. 170

188 Fig. 5.3 Comparison of different modulator designs with respect to V pi as a function of optical power into modulator branch. λ = 1554nm SOA = 100mA Devices with Dual SOAs have more optical power incident to the modulator giving both more heating in the modulator and more optical power due to the improvement in the saturation power. The lowest V pi was exhibited for a 500µm device with Ts and Dual SOAs Device

189 Clearly this is a non-linear effect as the shorter devices which heat up more - are more efficient modulators when normalized with respect to length. Also, the lowest values were for T electrode devices with Dual SOAs which yielded considerably less photocurrent at the peak. The extinction curves were measured for different wavelengths as illustrated in fig Fig. 5-4 DC extinction curve for a 200µm long lumped MZ for three different wavelengths 172

190 Figure 5-4 shows the wavelength dependence of the DC extinction curves for a 200um long device. As there is more gain for centered wavelengths, the curve peak is higher at 1564nm. Note that the residual power at 0V is highly dependent on the coupling of the fiber (angle and distance) so some of the variance is due to coupling and some is due to the different extinction due to phase bias differences. In Fig 5-5, the DC extinction V pi was measured for different optical powers going into the modulator at different wavelengths. Fig 5-5 Vpi for 200µm long device as a function of optical input power to one branch and wavelength 173

191 The SOA bias was varied from 20mA to 150mA. The logarithmic optical power dependence is considerably larger than the wavelength dependence of V pi indicating a large change in temperature within the device since the optical power is high in this integrated device. 5.2 RF EXTINCTION OF DEVICES The RF extinction was measured at 10Gbit/s with a PRBS as a function of wavelength for various peak-to-peak voltage outputs with the 0 level at the null of the modulator characteristic. Although the DC extinction for each device has been demonstrated in the previous section, under RF modulation not as much heating takes place and photocurrent competition occurs making the modulation not as efficient. Also, in order to have high speed, a parallel resistor is used in which less current flows through the device. As can be seen in fig 5-6ab, there is much more wavelength dependence with a singleended drive. 174

192 Fig 5-6a 250µm Single ended drive RF extinction at 10Gbit/s PRBS for various wavelengths Fig. 5-6b 250µm lumped series push-pull with 50 ohms in parallel 175

193 As the operation of the device is limited by the worst channel, there is a factor of two improvement by using series-push-pull for a wide wavelength range. As there is more photocurrent generated at lower wavelengths even though there is also more index shift and absorption the drive is opposing this photocurrent, and the RF extinction tends to be superior for the longer wavelengths. In the series push-pull case, one branch opposes the photocurrent and one is in the same direction giving less wavelength sensitivity. Clearly the devices are temperature sensitive. Taking the change in index as a function of voltage from fig. 3-6 one can estimate the drive voltage for different optical powers as shown in fig Fig 5.7 RF Vpi at 10Gbit/s model taken from DC extinction data in fig 3-6 for different optical powers. 176

194 The modeled data is compared to piecewise measured RF Voltages using 10Gbit/s BERT with maximum of V pp of 2V. As can be seen on the plots, at RF the modulator does not heat up as much and the voltage is higher than one would expect from looking at DC characteristics alone. Also shown on the plot is a comparison of V pi with different terminations 27. As one would expect, unterminated the drive voltage is much lower due to reflections off the end of the stub and more interaction with the modulator. Also, using a low termination of 25 ohms degrades the V pi but it is not linear related to the bandwidth enhancement of low termination. 27 Performed on device #20 177

195 5.3 BANDWIDTH COMPARISON The main trends of the bandwidth are shown in fig Extrapolated curves go through the single side modulation, and the SPP modulation from data in fig. 4-3 with 50 ohm front side termination. Fig 5-8 Small-signal optical response for different modulator designs. 50ohm termination Devices with Ts and end-50ohm termination have even better performance due to the improved characteristic impedance mismatch and traveling wave design. Using low termination resistors would enhance the bandwidth further as was discussed in Chap

196 5.4 CHIRP MEASUREMENTS Although direct modulation of the SGDBR offers a simple solution as a transmitter, as noted before, the chirp parameter is positive and ranges from approximately 3-9[330] over the C-Band. Chirp in Externally Modulated Lasers(EML) is caused by the electro-optic effect in the modulator, electrical crosstalk between the laser sections and the modulator and from reflected optical power[505]. Residual feedback from the output facet induces chirp and relaxation oscillations resulting in a lower transmission distance (between repeaters). In order to provide an adequate transmission distance particularly at higher bit rates, one desires a chirp parameter that is slightly negative usually in the range of 0 to -1 for 10Gbit/s operation[515] as was shown in the introduction. 5.5 CHIRP MEASUREMENT TECHNIQUES Chirp can be measured with a few different techniques. One of the easiest ways is the Gated-Delayed Self-Heterodyne (GDSH) technique. This setup consists of a modulated laser with a gated sinusoidal signal where the output is connected to a fiber interferometer similar to that of the linewidth measurement in Chap. 1. One arm has a delay of approximately 3.5µs. Both signals are combined and measured using an RF spectrum analyzer. 179

197 For a Mach-Zehnder modulator, the chirp can be expressed from the intensity and phase of the output signal. Where cos( i E I [5.3] cos cos sin sin tan [5.4] ) )sin( ' ' ( ) ) cos( ' ' ( ' ' [5.5] b b V t V V V V V t V V V V V V )sin ( )sin ( )sin ( )cos ( [5.6] so for the small-signal regime[516] ( 2 ) V V V V [5.7] However, the self-heterodyne method only gives the magnitude of the chirp parameter. Alternatively, one can measure the magnitude and sign of the chirp parameter using a network analyzer as discussed in [514,520] where the resonant frequencies of the fiber frequency response as measured from the network analyzer[514]. ) arctan( u D c L f u [5.8] where is the chirp parameter, is the wavelength, u is the number of the minimum in the response, D is the Dispersion parameter of the fiber, L is the length of the fiber 180

198 2-5 Chirp Parameter Alpha 1525nm Alpha 1545nm (Power (dbm) 1525nm Power (dbm) 1545nm Arm #1 DC Bias (V) Fig. 5-9 Chirp Parameter as a function of DC extinction curve for pi-shifted case with 550µm device (1.38um waveguide) Generation1 device Output Power (dbm) As can seen by fig. 5-9, with a one-sided modulation scheme with a pi-shift configuration, the device exhibits negative chirp. In the small-signal regime, the chirp parameter can be expressed as[541]: d 2I [5.9] di where I is the intensity and is phase. This means that at the maximum in the output power curve, we have high chirp since the change in intensity is minimal. It is the chirp in the on state that affects the transmission performance and this is often refered to as the 3dB rule[515]. When the device is off the chirp parameter is slightly positive and increasingly negative as the modulation depth is increased close to the maximum output power. A push-pull scheme lowers this negative chirp so that the modulation depth can 181

199 be increased further. This was measured for a 250µm long device as shown in fig Fig 5-10 small-signal chirp parameter for inverting and non-inverting operation for parallel pushpull and single-sided drive Although examining the small-signal chirp parameter is illustrative of the dynamic device operation, performance under large signal operation is preferred. A time-resolved or dynamic measurement of the chirp parameter can be achieved using a setup utilizing an optical filter which is either a Fabry- Perot etalon, waveguide grating router, Mach-Zehnder interferometer[542] or Optical Spectrum Analyzer monochromator and high-speed oscilloscope[505]. By using this setup one can see dynamically how the frequency shifts on the rising and falling edge during modulation at 10Gbit/s. 182

200 Fig 5-11 Dynamic chirp measurement demonstrating how a chirp parameter of -0.7 can be achieved by adjusting the gain in the two branches of the MZ performed at 10 Gbit/s The alpha parameter can be changed by varying the power in the two branches either with gain using SOAs or loss with the passive sections. It is fairly easy to achieve 0 chirp using a SPP electrode structure, however if negative chirp is desired, it is easier to use a single sided drive modulation as shown in fig

201 5.7 LINEARIZATION OF MODULATORS For analog modulation, a high degree of linearization is desired to maintain the dynamic range of a photonic link. Mach-Zehnder modulators generally exhibit nonlinear transfer characteristics. Fiber optic RF links require sufficient linearity for applications such as satellite communications, radar, CATV, and others. A number of different techniques have been employed to achieve linearity using external modulators. Approaches have included the use of directional coupler modulators[524], Mach Zehnder modulators[525], and Electro-absorption modulators[526] and combinations of these in dual parallel and series configurations[5]. Alternatively, electrical linearization can be performed using optical negative feedback, phase-shift modulation, feedforward or pre-distortion techniques although an all optical scheme avoids complicated electrical circuits and their frequency responses. The combination of multiple modulators allows for modification of the nonlinear response of either to cancel out the nonlinearity of the total response. A series combination will decrease the bandwidth. The null in the third-derivative occurs at V pi for a Mach-Zehnder and at a bias with significant modulation efficiency for a Franz-Keldysh EAM[2]. This leads to the increase in SFDR without sacrificing power for high gain links. 184

202 For narrowband applications (< 1 octave) the Spur free dynamic range is governed by only the odd order inter-modulation terms- however for broadband applications all harmonics and orders of inter-modulation distortion must be considered. [2]. Spur-free dynamic range (SFDR) is defined as the range of input powers over which the output power at the carrier frequency is above the noise floor while the third-order distortion products remain below the noise floor. Fig 5.12 Fundamental signals (500MHz) and 3 rd order distortion signals When a two-tone RF signal is applied, the bias voltage is given by: V V m (cos t cos ) [5.10] b 1 o 1 2t where m o is the modulation depth and V b is the dc bias voltage [5]. 185

203 The inter-modulation distortion is highly dependent on the bias conditions as shown in fig Fundamental Second harmonic Third harmonic Optical Power Detected Power (dbm) Reverse bias (V) Fig Detected average optical power and RF power of fundamental and distortion products for 0 dbm modulation power. The narrowband SFDR was measured for device #5, at 500MHz with a optimized phase section bias. 186

204 = nm f=0.5ghz IIP3=25.2dBm Output RF Power, dbm fundamental 3rd-order distortion SFDR=112dB-Hz 2/ Input RF Power, dbm Fig Spur-free dynamic range measured on device #5 SOA = 100mA Gain = 100mA Phase = 1.2mA Modulator bias = -1V These results demonstrate that fairly good linearity can be obtained for optimized bias points. 187

205 REFERENCES [500] Saavedra AA, Rigole P-J, Goobar E, Schatz R, Nilsson S. Amplitude and frequency modulation characteristics of widely tunable GCSR lasers. IEEE Photonics Technology Letters, vol.10, no.10, Oct. 1998, pp [501] D. A. Neamen, Semiconductor Physics and Devices Basic Principles Second edition Irwin, Chicago [502] Nishimoto H, Okiyama T, Kuwata N, Arai Y, Miyauchi A, Touge T. New method analyzing eye patterns and its application to high-speed optical transmission system. Journal of Lightwave Technology, vol.6, no.5, May 1988, pp [503] Cartledge JC, Debregeas H, Rolland C. Dispersion compensation for 10 Gb/s lightwave systems based on a semiconductor Mach-Zehnder modulator. IEEE Photonics Technology Letters, vol.7, no.2, Feb. 1995, pp [504 Adams DM, Rolland C, Fekecs A, McGhan D, Somani A, Bradshaw S, Poirier M, Dupont E, Cremer E, Anderson K um transmission at 2.5 Gbit/s over 1102 km of NDSF using discrete and monolithically integrated InGaAsP-InP Mach-Zehnder modulator and DFB laser. Electronics Letters, vol.34, no.8, 16 April 1998, pp [505] Agilent 86146B-DPC Time Resolved Chirp Application User s Guide [506] Saavedra AA, Rigole P-J, Goobar E, Schatz R, Nilsson S. Relative intensity noise and linewidth measurements of a widely tunable GCSR laser. IEEE Photonics Technology Letters, vol.10, no.4, April 1998, pp [507] Henry CH. Theory of the linewidth of semiconductor lasers. IEEE Journal of Quantum Electronics, vol.qe-18, no.2, Feb. 1982, pp [508] Arnaud J. Natural linewidth of semiconductor lasers. IEE Proceedings-J Optoelectronics, vol.134, no.1, Feb. 1987, pp.2-6. [509] Osinski M, Buus J. Linewidth broadening factor in semiconductor lasersan overview. IEEE Journal of Quantum Electronics, vol.qe-23, no.1, Jan. 1987, pp

206 [510] Harder C., K. Vahala, A. Yariv, Measurement of the Linewidth Enhancement Factor of semiconductor lasers, Appl. Phys. Lett., vol 42, no. 15, pp , Feb [511] M. Aoki, K. Uom, T. Tsuchiya, S. Sasaki, M. Okai, N. Chinone, DFB Laser, IEEE. J. Quantum Electron., Vol. 27, no.2, pp.1782, Feb [512] Amann M-C, Schimpe R. Excess linewidth broadening in wavelengthtunable laser diodes. Electronics Letters, vol.26, no.5, 1 March 1990, pp [513] Wang SJ, Dutta NK. Intermodulation and harmonic distortion in GaInAsP distributed feedback lasers. Electronics Letters, vol.25, no.13, 22 June 1989, pp [514] Devaux F, Sorel Y, Kerdiles JF. Simple measurement of fiber dispersion and of chirp parameter of intensity modulated light emitter. Journal of Lightwave Technology, vol.11, no.12, Dec. 1993, pp [515] Dorgeuille F, Devaux F. On the transmission performances and the chirp parameter of a multiple-quantum-well electroabsorption modulator. IEEE Journal of Quantum Electronics, vol.30, no.11, Nov. 1994, pp [516] Hoon Kim, Gnauck AH. Chirp characteristics of dual-drive. Mach- Zehnder modulator with a finite DC extinction ratio. IEEE Photonics Technology Letters, vol.14, no.3, March 2002, pp [517] Sung Kee Kim, Mizuhara O, Park YK, Tzeng LD, Kim YS, Jichai Jeong. Theoretical and experimental study of 10 Gb/s transmission performance using 1.55 um LiNbO/sub 3/-based transmitters with adjustable extinction ratio and chirp. Journal of Lightwave Technology, vol.17, no.8, Aug. 1999, pp [518] Devaux F, Sorel Y, Kerdiles JF. Chirp measurement and transmission experiment at 10 Gbit/s with Wannier-Stark modulator. Electronics Letters, vol.29, no.9, 29 April 1993, pp [519] Fells JAJ, White IH, Gibbon MA, Penty RV, Thompson GHB, Wright AP, Saunders RA, Armistead CJ. Controlling the chirp in electroabsorption modulators under digital modulation. Electronics Letters, vol.30, no.24, 24 Nov. 1994, pp

207 [520] Srinivasan RC, Cartledge JC. On using fiber transfer functions to characterize laser chirp and fiber dispersion. IEEE Photonics Technology Letters, vol.7, no.11, Nov. 1995, pp [521] Zadok A, Shalom H, Tur M, Cornwell WD, Andonovic I. Spectral shift and broadening of DFB lasers under direct modulation. IEEE Photonics Technology Letters, vol.10, no.12, Dec. 1998, pp [522] Koch TL, Corvini PJ. Semiconductor laser chirping-induced dispersive distortion in high-bit-rate optical fiber communications systems. IEEE International Conference on Communications '88: Digital Technology - Spanning the Universe. Conference Record (Cat. No.88CH2538-7). IEEE. 1988, pp vol.2. [523] Cummings UV, Bridges WB. Bandwidth of linearized electrooptic modulators. Journal of Lightwave Technology, vol.16, no.8, Aug. 1998, pp [524] Cox CH III, Betts GE, Johnson LM. An analytic and experimental comparison of direct and external modulation in analog fiber-optic links. IEEE Transactions on Microwave Theory & Techniques, vol.38, no.5, May 1990, pp [525] Welstand RB, Sun CK, Liu YZ, Pappert SK, Zhu JT, Chen JM, Yu PKL. High dynamic range in electroabsorption modulator for analog links. SPIE-Int. Soc. Opt. Eng. Proceedings of Spie - the International Society for Optical Engineering, vol.2560, 1995, pp USA. [526] Farwell ML, Lin Z-Q, Wooten E, Chang WSC. An electrooptic intensity modulator with improved linearity. IEEE Photonics Technology Letters, vol.3, no.9, Sept. 1991, pp USA. [527] Lam JF, Tangonan GL. A novel optical modulator system with enhanced linearization properties. IEEE Photonics Technology Letters, vol.3, no.12, Dec. 1991, pp [528] Sung-Il Sohn, Sang-Kook Han. Linear optical modulation in a serially cascaded electroabsorption modulator. Microwave & Optical Technology Letters, vol.27, no.6, 20 Dec. 2000, pp [529] Cummings UV, Bridges WB. Effects of velocity mismatch and transit time on linearized electro-optic modulators. Conference Proceedings. LEOS '96 9th Annual Meeting. IEEE Lasers and Electro-Optics Society

208 Annual Meeting (Cat. No.96CH35895). IEEE. Part vol.2, 1996, pp vol.2.. [530] Bridges WB, Cummings UV, Schaffner JH. Linearized modulators for analog photonic links. International Topical Meeting on Microwave Photonics. MWP '96 Technical Digest. Satellite Workshop (Cat. No.96TH8153). IEEE. 1996, pp [531] Price AJ, Pierre L, Uhel R, Havard V. 210 km repeaterless 10 Gb/s transmission experiment through nondispersion-shifted fiber using partial response scheme. IEEE Photonics Technology Letters, vol.7, no.10, Oct. 1995, pp [532] Devaux F. Optimum prechirping conditions of externally modulated lasers for transmission on standard fibre. IEE Proceedings Optoelectronics, vol.141, no.6, Dec. 1994, pp [533] Cartledge JC. Combining self-phase modulation and optimum modulation conditions to improve the performance of 10-Gb/s transmission systems using MQW Mach-Zehnder modulators. Journal of Lightwave Technology, vol.18, no.5, May 2000, pp [534] Jackson MK, Smith VM, Hallam WJ, Maycock JC. Optically linearized modulators: chirp control for low-distortion analog transmission. Journal of Lightwave Technology, vol.15, no.8, Aug. 1997, pp [535] Cartledge JC. Optimizing the bias and modulation voltages of MQW Mach-Zehnder modulators for 10 Gb/s transmission on nondispersion shifted fiber. Journal of Lightwave Technology, vol.17, no.7, July 1999, pp [536] Laliew C, Lovseth SW, Xiaobo Zhang, Gopinath A. A linearized optical directional-coupler modulator at 1.3 um. Journal of Lightwave Technology, vol.18, no.9, Sept. 2000, pp [537] Elrefaie AF, Wagner RE, Atlas DA, Daut DG. Chromatic dispersion limitations in coherent lightwave transmission systems. Journal of Lightwave Technology, vol.6, no.5, May 1988, pp [538] Simons RN, Goverdhanam LK, Katehi LPB. Novel low loss wide-band multi-port integrated circuit technology for RF/microwave applications Topical Meeting on Silicon Monolithic Integrated Circuits in RF Systems. Digest of Papers (IEEE Cat. No.01EX496). IEEE. 2001, pp

209 [539] Nakamoto H, Akiyama Y, Matsuda M, Yamaji K, Kotaki Y, Hajikano Y, Kubota M, Machida H, Hamano H. 10-Gbit/s single mode fibre transmission using a negative-chirp modulator integrated DFB-LD transmitter and APD receiver. 11th International Conference on Integrated Optics and Optical Fibre Communications 23rd European Conference on Optical Communications IOOC-ECOC 97 (Conf. Publ. No.448). IEE. Part vol.1, 1997, pp.7-10 vol.1. [540] Fells JAJ, Gibbon MA, White IH, Thompson GHB, Penty RV, Armistead CJ, Kimber EM, Moule DJ, Thrush EJ. Transmission beyond the dispersion limit using a negative chirp electroabsorption modulator. Electronics Letters, vol.30, no.14, 7 July 1994, pp [541] Koyama F, Oga K. Frequency chirping in external modulators. Journal of Lightwave Technology, vol.6, no.1, Jan. 1988, pp [542] Laverdiere C, Fekecs A, Tetu M. A new method for measuring timeresolved frequency chirp of high bit rate sources. IEEE Photonics Technology Letters, vol.15, no.3, March 2003, pp [543] Shinhee Won, Jaehoon Lee, Yonghoon Kim, Seongha Kim, Jichai Jeong. Performance limits of 10-Gb/s optical duobinary transmissions using reduced bandwidth single-arm Mach-Zehnder modulators considering residual chirp and dc-bias offset. IEEE Photonics Technology Letters, vol.15, no.3, March 2003, pp [544] Welstand RB, Zhu JT, Chen WX, Clawson AR, Yu PKL, Pappert SA. Combined Franz-Keldysh and quantum-confined Stark effect waveguide modulator for analog signal transmission. Journal of Lightwave Technology, vol.17, no.3, March 1999, pp [545] Welstand RB, Sun CK, Pappert SA, Liu YZ, Chen JM, Zhu JT, Kellner AL, Yu PKL. Enhanced linear dynamic range property of Franz-Keldysh effect waveguide modulator. IEEE Photonics Technology Letters, vol.7, no.7, July 1995, 192

210 Chapter 6 CONCLUSIONS AND FUTURE WORK Although the device performance shown here rivals some of the best discrete components, there are always improvements that could be done to enhance the performance further. Higher output power will improve the transmitter characteristics markedly as not only will the heating in the modulator make it more efficient, but the required drive voltage will be reduced considerably. A less conservative SGDBR design would give higher output power and more efficiency in the modulator. Alternatively, using longer SOAs or enhancing the saturation power by widening the ridges would improve the output power. Using a QWI integration platform would improve the gain of the material with centered wells which improves the laser at the expense of the SOA saturation characteristics. Using the quantum well-intermixing platform would allow for more than 2 bandgaps - providing low loss waveguides in the passive regions and higher efficiency in the modulator electrode region. Although much better matching of Zo than many CPW TW EAMs, neither the characteristic impedance or the microwave index are completely matched and this can be improved further by reducing the capacitance of the PN junction as the parasitics are quite low using BCB and a SI substrate. Using a 193

211 slightly lower doped waveguide would give a better confinement factor as the depletion region would move out and overlap more with the optical mode. Also, integration of the termination resistor would make testing easier and remove the electrical reflections inherent in wirebonds/ribbonbonds. Based on V pi RF and bandwidth measurements, rear termination seems to be superior due to the reduced microwave reflections at the input. This along with a low resistance termination gives an enhancement in the S 21 characteristics without reducing the drive voltage as much. More modeling of the microwave properties of these devices and loaded transmission lines that give the desired index and characteristic impedance is needed. As can be seen in fig. 6-1, the current power management shows that we have approximately 20mW output from the SGDBR untuned which is amplified approximately 5-7dB depending on the SOA length then is attenuated around 5dB. This means that the SOA compensates mostly for the insertion loss of the modulator. Unfortunately, with the lensed fiber this 20mW quickly becomes 5mW fiber coupled. The integration of a mode converter would make this design more in-line with typical supplier requirements of 10mW fiber coupled output power. 194

212 SGDBR 20mW SOA +5dB 60mW -5dB 20mW 5mW coupled fiber Fig. 6-1 Power management through the device. Going to a shorter electrode will have a prohibitively high drive voltage >8V RF. This doesn t seem to be a good option unless the device is coupled with more output power or a higher Q waveguide used to make the device more efficient although it is more wavelength dependent. Another possibility is the use of shallow MQW in the waveguide which also most likely will need to be designed carefully to prevent excessive wavelength dependence and optical losses. Although not desirable in terms of complexity, a butt-joint regrowth is always possible to reduce the capacitance in the modulator. This is always a compromise with efficiency however if the layers are not doped highly the propagation loss could be reduced so that the device could be longer. In the foreseeable future, tunable Mach-Zehdner based transmitters could be found useful in more complex photonic integrated circuits such as photocurrent driven wavelength converters. 195

213 6.1 WAVELENGTH CONVERTERS Tunable wavelength converters represent a novel class of highly sophisticated photonic integrated circuits that are crucial in the function of future optical networks. They allow for the manipulation of wavelengths in WDM optical switches, routers and add/drop multiplexers. Many different implementation of non-tunable wavelength converters have been proposed using cross phase modulation(xpm) in SOAs and fiber[2,3], and cross absorption modulation(xam) of SOAs [1], Many of these architectures have been demonstrated to perform the significant feature of digital signal regeneration including improvements in extinction ratio, signal to noise ratio, pulse width etc. More recently, monolithically-integrated tunable all-optical wavelength converters (TAO-WC)[4] have been demonstrated and have shown promise to allow for the conversion of one wavelength to another without requiring the signal to pass through electronics. One further extension of the work in this thesis is in the integration of the Mach-Zehnder-SOA-SGDBR with a photodetector to provide wavelength conversion over a wide tuning range[717]. 196

214 Semiconductor Optical Amplifier SGDBR Laser Mach-Zehnder Modulator 2 1 Franz-Keldysh MZ optical waveguide Fig Device layout for the 1 st demonstration of an OEIC wavelength converter using the Mach-Zehnder-SOA-SGDBR transmitter and Franz-Keldysh detector. The first implementation of wavelength converters used a Franz-Keldysh detector that does not involve quantum wells. This type of detector gives a fairly linear response. Improvements on the design of wavelength converters will focus on decreasing the optical input requirements with the integration of SOAs before the detector and investigation of the linearity and efficiency of using integrated active quantum-well detectors. 197

215 Fig Photocurrent Generated in the Franz-Kelydsh detector as a function of optical power Nonetheless, first results using this configuration seem promising as the extinction ratio is sufficient to provide <2dB power penalty over the tuning range of the transmitter[617]. 198