00001.tif

PDF

Download

Open document

Transcript (if available)

Content INFORMATION TO USERS This manuscript has been reproduced from the microfilm m aster. UMI films the text directly from the original or copy submitted. T hus, some thesis and dissertation copies are in typewriter face, while others may be from any type of computer printer. The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleedthrough, substandard m erging, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send UMI a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note win indicate the deletion. Oversize materials (e.g., maps, drawings, charts) are reproduced by sectioning the original, beginning at the upper left-hand comer and continuing from left to right in equal sections with small overlaps. Each original is also photographed in one exposure and is included in reduced form at the back of the book. Photographs included in the original manuscript have been reproduced xerographically in this copy. Higher quality 6" x 9 " black and white photographic prints are available for any photographs or illustrations appearing in this copy for an additional charge. Contact UMI directly to order. A Bell & Howell Information Company 300 North Zeeb Road. Ann Arbor. M l 48106-1346 USA 313/761-4700 800/521-0600 MULTICHANNEL OPTICAL TRANSMITTER MODULES FOR GIGABITS/S PARALLEL DIGITAL DATA LINKS Sabeur Siala A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillm ent of the Requirements for the Degree DOCTOR OF PHILOSOPHY (Electrical Engineering) May 1995 Copyright 1995 Sabeur Siala UMI Number: 9621629 UMI Microform 9621629 Copyright 1996, by UMI Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. UMI 300 North Zeeb Road Ann Arbor, MI 48103 UNIVERSITY OF SOUTHERN CALIFORNIA THE GRADUATE SCHOOL UNIVERSITY PARK LOS ANGELES, CALIFORNIA 90007 This dissertation, written by ...................................... SABEUR SIALA..................................... under the direction of h.\§....... Dissertation Committee, and approved by all its members, has been presented to and accepted by The Graduate School, in partial fulfillment of re quirements for the degree of DOCTOR OF PHILOSOPHY D ean o f Graduate Studies DISSERTATION COMMITTEE Chairperson Dedicated To My Mother and My Father A ck now led gm ents Many people have greatly helped and supported me during the course of m y stay at USC, and I w ould like to thank them. My thesis advisor and supervisor, Prof. Richard N. Nottenburg has been of great support to me during my work with him. I have profited from his broad know ledge, extensive experimental skills, unlimited energy, and strong dedication to his students. I deeply thank him for his constant encouragement, superb guidance, and sincere care. I thank Prof. P. Daniel Dapkus and Prof. Murray Gershenzon for being on my dissertation committee and for their remarkable teaching. I thank Prof. Elsa Garmire for her guidance during my first two years at USC. A lso, I thank her and Prof. Alexander A. Sawchuk for being on my qualifying exam committee. I extend my thanks to Prof. Anthony F. J. Levi for the opportunity to work with him on some projects, particularly the work on optical ribbon fiber. I thank all my colleagues at the H igh-Speed Technology Laboratory (HSTL) for all their help and support. In particular, I would like to express m y deep appreciation and admiration to m y former colleague Prof. M. Govindarajan. His knowledge, skills, patience, and friendliness are very remarkable. A lso, I wish to thank Vinodkumar Ramakrishnan, Mike Chen, Sanjay Mansingh, A nish Shah, Natasha Polak, Sean P. W oyciehowsky, and Joseph T. Elliott. Finally, I w ish to acknowledge the great work done by the many undergraduate students at HSTL, w hom I thank deeply - Brent Chun, Benji DeBord, Jose Franco, Ryan Kinter, Mark M endez, Anh N guyen, Jose Padilla, Jonathan Sadowsky, Jerome Ware. From the Compound Semiconductor Laboratory I thank Dr. Hanmin Zhao for his excellent work in grow ing the semiconductor lasers used in this work. From Prof. Levi’ s group, I thank Dr. N ew ton C. Frateschi, Ashok P. Kanjamala, and Verko DeKovic. I profoundly thank my friends in Los A ngeles for their unwavering support, particularly Karina Wojczakowska, Eugene Park, M oez Ayed, Robert Buchanan, Rahul Asthana, Don Fritts, and Jeff McLaughlin. I heartily thank my sisters and brothers - Sondes, Kalthoum, Fatima, Khaled, and Maher who, inspite of the distance, have been very supportive of me throughout the years. I wish to dedicate part of this work to the fond memories of my best childhood friend Mohamed Baccour and his family. Finally, no words can adequately describe the unconditional support and love from my parents, Bakhita and Rachid. To them I dedicate this work. Table of Contents Acknowledgm ents iii List of Figures vii List of Tables XV Abstract xvi I. Multichannel Optical Data Links 1.1 Introduction 1 1.2 Applications of Parallel Optical Data Links 2 1.3 Performance Requirements 4 1.4 Other Work in Optical Transmitter Arrays 5 1.5 Work Performed 9 References 11 II. Strained InGaAs/GaAs SQW Lasers 2.1 Introduction 15 2.2 Growth and Fabrication 17 2.3 DC Operation of InGaAs/GaAs Laser-Gate 19 2.4 Reliability of Strained InGaAs/GaAs Lasers 25 References 27 III. Small and Large-Signal Modulation 3.1 Introduction 32 3.2 Background on Small-signal Modulation of QW lasers 33 3.3 Small-Signal Modulation of Strained InGaAs/GaAs SQW Two-Terminal Lasers 42 3.4 Small-Signal Modulation of InGaAs/GaAs Laser-Gates 51 3.5 Large-Signal Digital Modulation of Two-Terminal InGaAs/GaAs Lasers 56 3.6 Large-Signal Digital Modulation of InGaAs/GaAs Laser-Gates 65 V 3.7 Design and Fabrication of Single-Channel Transmitter M odules 74 References 79 IV. SPICE Circuit Simulation of Optical Transmitters 4.1 Introduction 85 4.2 SPICE Circuit Modeling of 2-Terminal InGaAs Lasers 87 4.3 SPICE Circuit Modeling of InGaAs Laser-Gate 93 4.4 Future Work 98 References 100 V. Optical Ribbon Fiber for Parallel Data Links 5.1 Introduction 103 5.2 N oise in Multi-Mode Fiber Systems 105 5.5 Low Skew Optical Ribbon Fiber for Parallel Data Links 113 References 121 VI. Multi-Channel Transmitter Array Modules 6.1 Introduction 125 6.2 Four-Channel Transmitter Module Using Two- Terminal InGaAs/GaAs Laser Arrays 126 6.3 Four-Channel Transmitter Module Using InGaAs/GaAs Laser-Gate Arrays 129 6.4 Ten-Channel Transmitter Module Using Two- Terminal InP Laser Arrays 137 6.5 Conclusions 144 References 148 VII. Conclusions 151 References 154 VIII. Appendices 1. Appendix A 155 vi List o f Figures Fig. 2.1: Schematic drawing of a laser-gate. 20 Fig. 2.2: Gain versus carrier density for laser-gate. 20 Fig. 2.3: CW light output power versus anode current at various gate voltages. The dashed curve is for the case where the gate and anode are shorted together. 21 Fig. 2.4: Gate current (/,) versus gate voltage ( V ) for anode currents /,= 4, 6, 8,10,12, and 14 mA. 22 Fig. 2.5: Equivalent circuit o f the laser-gate (between the dashed lines) and the external supply circuit of the dc measurement. 23 Fig. 2.6: Light output power per facet versus gate voltage at various anode currents. 24 Fig. 3.1(a): Effect of gain compression on modulation response of QW lasers. 36 Fig. 3.1(b): Effect of gain compression on the resonance frequency of QW lasers. 36 Fig. 3.2: Simple equivalent circuit of laser parasitics. 39 Fig. 3.3(a): 3-dB bandwidth contours in the capacitance- inductance plane for R = 5 ft and Rs = 50 Q. 40 Fig. 3.3(b): 3-dB bandwidth contours in the capacitance- inductance plane for R = 14 ft and Rs = 50 41 Fig. 3.4: Experimental setup for laser small-signal m easurem ent. 42 VII Fig. 3.5: Normalized small-signal modulation response of a HR coated InGaAs laser for various light output powers. 44 Fig. 3.6: Measured 3-dB modulation bandwidth versus square root of incremental current ^Jl - /„,. MCEF is obtained by a linear curve fit at low power levels. 45 Fig. 3.7: Measured 3-dB modulation bandwidth versus square root of output power s[P^, with a linear curve fit of at low power levels. 46 Fig. 3.8(a): Intensity noise spectrum of laser biased at 6 mA showing a peak at the resonance frequency f r = 5.05 GHz. 48 Fig. 3.8(b): Intensity noise spectrum of laser biased at 12 mA showing a peak at the resonance frequency f r = 7.5 GHz. 48 Fig. 3.9: Laser resonance frequency as measured using the RIN technique (circles) and 6 dB electrical frequency as measured with Network Analyzer (filled circle), as a function of . Note that the ratio decreases with increasing 1 . 49 Fig. 3.10: dependence on (squares) indicating / the dominance of parasitics. Also, * s plotted for D = 3.45 GHz and ft„r = 7.0 GHz. 50 Fig. 3.11: Gain of laser-gate as a function of carrier density. 53 Fig. 3.12: Normalized small-signal modulation response of laser-gate for various optical output powers per facet. The anode current is 11 mA. 55 Fig. 3.13: Fig. 3.14: Fig. 3.15: Fig. 3.16: Fig. 3.17: Fig. 3.18: Fig. 3.19: Fig. 3.20: Fig. 3.21: Measured 3-dB modulation bandwidth f 3_i J B versus , with a linear curve fit of f 3_r lB at low pow er levels. 55 Optical transmitter consists o f a InGaAs 2-terminal laser and a common-base driver. 59 Turn-on delay of the transmitter as the low level of the input voltage is sw ept from 0 to 520 mV (/, from 0 to 3 mA) w ith a high level at 920 mV. 60 Eye diagrams 2-terminal laser transmitter obtained with an input voltage swing of 530 mV and a 2*5-1 NRZ PRBS at (a) 500 Mb/s, (b) 1 Gb/s, (c) 1.5 Gb/s, and (d) 2 G b/s. 61 Transmitter Eye-diagram and jitter distribution (at 1 G b /s 217-1 PRBS) for a modulating current high level of 4.2/,,, and low level of (a) % (/,*) and(b)/„,. 63 Bias-free modulation of 2-terminal laser at 500 M b /s (217-1 NRZ PRBS) for a current high level of (a) 4.2/„,, and (b) 7.5/,,,. 64 Circuit diagram of optical transmitter. 66 Transmitter's optical output power versus input voltage for anode current of 13.5 mA. 66 Eye-diagrams for a 2 17-1 NRZ PRBS data pattern after transmission through 16 m of MM optical fiber. The input signal swing is 0.8 V at a data rate of (top to bottom) 500 M b/s, 1 Gb/s, 1.5 Gb/s, and 2 Gb/s. 68 ix Fig. 3.22: Fig. 3.23: Fig. 3.24: Fig. 3.25: Fig. 3.26: Fig. 3.27: Fig. 3.28: Fig. 3.29: Fig. 3.30: Fig. 3.31: Fig. 4.1: Experimental setup for the BER measurement. 69 BER versus input signal swing at the transmitter measured with a 27-l NRZ PRBS data pattern at 1 Gb/s. 70 Phase margin after transmission through 16 m of MM fiber versus input voltage swing at a BER of lO'11 for a 1 G b/s, 27-l NRZ PRBS data pattern. 71 Phase margin after transmission through 16 m of MM fiber versus data rate at a BER of 10'1* for an input voltage swing of 0.8 V (27-l NRZ PRBS data pattern). 72 Laser-gate use in an actual data link transmitter. 73 Photograph of the completed ceramic board of an optical transmitter with a 2-terminal laser. 75 Photograph showing the different parts of a transmitter module incorporating a 2- terminal InGaAs laser. 76 Photograph of the completed ceramic board of an optical transmitter with InGaAs laser-gate. 77 Photograph showing the different parts of a transmitter module incorporating an InGaAs laser-gate. 78 Photograph of the completed transmitter m odule. 79 Methods for simulating the behavior of semiconductor lasers. 86 Fig. 4.2: Equivalent circuit m odel of a semiconductor laser. 91 Fig. 4.3: Temperature-independent subcircuit to model the gain function in Eqn.(4.5). 92 Fig. 4.4: Simulated and measured output light per facet versus injection current of an InGaAs/GaAs laser. Inset: Simulated pulse response of the laser. 92 Fig. 4.5: Equivalent circuit m odel of InGaAs laser-gate. 96 Fig. 4.6: Simulated (solid) and measured (dashed) output light per facet versus anode current for input voltages vl > ( = 0.8,1.6, and 2.8 V. 97 Fig. 4.7: Simulated eye-diagram of the three-terminal laser transmitter for input voltage swing of 800 mV, input signal rise and fall times of 100 ps, and a 2^-1 NRZ PRBS pattern. 98 Fig. 5.1: Optical Spectrum of a 2-terminal InGaAs/GaAs laser modulated from slightly above threshold at 500 M b/s with ...1010... input pattern, resulting in peak X ~ 955.1, mean X ~ 955.85 nm, FWHM ~ 2.97 nm, m ode spacing AA ~ 0.4 nm (Af ~ 131.53 GHz), a ~ 1.26 nm. 108 Fig. 5.2(a): Optical Spectrum of a 3-terminal InGaAs/GaAs laser modulated at 500 M b/s w ith ...1010... input pattern for Ia = 13 mA, resulting in peak X = 950.02 nm, mean X = 952.05 nm, FWHM = 8.96 nm, and c r = 3.8 nm. 110 Fig. 5.2(b): Optical Spectrum of a 3-terminal InGaAs/GaAs laser modulated at 1 G b /s with ...1010... input pattern for Ia = 13 mA, resulting in peak X = 948.71, mean X = 951.83 nm, FWHM = 9.77 nm, and a = 4.15 nm. 110 Fig. 5.3: Fig. 5.4: Fig. 5.5: Fig. 5.6: Fig. 5.7: Fig. 5.8: Fig. 6.1: Fig. 6.2: Eye-diagrams obtained at 1 G b /s with 217 NRZ PRWS for input voltage swing of 1 V, w hen the fiber is (a) not disturbance, (b) bent to a radius of 5 mm, (c) bent to a radius of 1 mm, (d) shaken.at random. 112 Absolute coupling efficiency versus longitudinal laser-to-fiber separation distance of an as-cleaved optical ribbon fiber, with the lateral and transverse positions of the fiber at maximum coupling efficiency. 115 Relative coupling efficiency and interchannel crosstalk of the ribbon fiber as a function of the longitudinal laser-to-fiber separation distance. 116 Relative coupling efficiency and interchannel crosstalk of the ribbon fiber as a function of the lateral offset of the fiber with respect to the laser active region. 117 Illustration of the optical ribbon fiber MT connector (After [22]). 118 Eye-diagram and BER versus clock-to-data delay of parallel synchronous digital transmission over 102 m of low skew MM ribbon fiber at 622 M b /s with 27-l NRZ PRBS. 120 Photograph of the completed ceramic board for the four-channel transmitter with two- terminal InGaAs laser array. 127 Eye-diagrams of the four channel transmitter • • XII Fig. 6.3: Fig. 6.4: Fig. 6.5: Fig. 6.6: Fig. 6.7: Fig. 6.8: Fig. 6.9: with two-terminal InGaAs laser array. The eye- diagrams were measured one at a time with an input signal at lG b /s 2*7 NRZ PRWS and low and high voltage levels of 0.9 V and 1.4 V, respectively. 128 Photograph of the completed ceramic substrate of the four-channel laser-gate transmitter m odule. 130 Eye-diagrams of the four-channel laser-gate transmitter at 1 G b/s with NRZ 2*7-1 PRWS for input voltage swing of 800 mV and fiber length of 200 m. 132 BER versus input signal swing measured with 1 G b /s NRZ 2*5-1 PRBS after synchronous transmission through 16 m of MM fiber. 133 Phase margin versus data rate for 800 mV input signal swing for synchronous transmission with 27-l NRZ PRBS. 134 Overall interchannel crosstalk versus data rate (filled circles) of the four-channel laser-gate transmitter array. Lower curve (diamonds) represents the measured feed-through from gate to anode of a single laser-gate. 135 Electrical feed-through between the gate and anode of a laser-gate in the case of no bypass capacitance (diamonds), 82 pF capacitor (circles), and 82 pF capacitor in parallel with 1 nF capacitor (triangles). 136 Photograph of the completed ceramic board of the 10-channel transmitter module. The laser driver array circuitry is implemented using discrete components. 139 xiii Fig. 6.10: Photograph of the ten-channel 1.3 pm transmitter array module. 140 Fig. 6.11: Bias-free modulation at 500 M b /s for input pattern of 217 PRWS and high-level current through the laser of 19 mA. 141 Fig. 6.12: Eye-diagram obtained at 1 G b /s, with 2*7 2 3 (/„,= 2.5mA). 142 NRZ PRWS for Ih = Ihw = - /„ and = 19mA Fig. 6.13: Eye-diagram (top) obtained at 1 G b/s, with 217 NRZ PRWS for Ihw - 1 „ = 2.5 mA and high =19m A, and corresponding timing jitter of cr = 26 ps. 142 Fig. 6.14: Transmitter duty-cycle versus the ratio of laser bias to laser threshold ( ) at 500 M b /s and lG b /s. " 143 Fig. 6.15: Eye-diagrams of some of the channels of our 1.3 pm laser transmitter array module at 1 G b/s/channel for an ECL input voltage from -1.7V to -0.9 V and a 217 NRZ PRWS pattern. 145 List of Tables Table 1.1: Comparison of reported multi-channel optical transmitters for parallel data links. Table 2.1: Reported results on the reliability of strained InGaAs/GaAs QW lasers. Table 4.1: Definitions and units of various symbols used in the derivation. Table 4.2: Definition and units of variables ABSTRACT We have designed, implemented, and evaluated multichannel optical transmitter m odules operating at 1 G b it/s/ch an nel for parallel digital data links. We have used two and three-terminal (laser-gate) edge- em itting strained InGaAs/GaAs single quantum well laser arrays em itting at a w avelength of 980 nm , as well as edge-em itting two- term inal strained InG aA sP/InP m ultiple quantum laser arrays em itting at 1.3 pm. We have compared the large-signal digital m odulation performance of these different laser technologies. We have demonstrated the advantages of InGaAs/GaAs laser-gates such as efficien t on -off m odulation, elim ination o f driver circuitry, com patibility w ith the em itter-coupled logic fam ily, and high temperature stability. Synchronous D C-coupled transm ission over 200 m of multimode optical fiber is shown using our 4-channel laser- gate transmitter module. Bit-error-rates of ~ 10‘13 and phase margin of 311° have been measured at 1 G b/s for an input voltage sw ing of 800 mV . In addition, a large-signal SPICE equivalent circuit m odels for two and three-terminal InGaAs/GaAs lasers based on the monom ode rate equations is d eveloped. We have also reported a 12-channel m ultim ode optical ribbon fiber with a very low interchannel skew of 1.25 p s/m . We have demonstrated that synchronous transmission of 622 M Bytes/s over a distance of ~ 1 km can be achieved w ith this very low skew optical ribbon fiber. CHAPTER I M ultichannel Optical Data Links 1.1 Introduction In response to an ever increasing demand for higher interconnection density, bandw idth, and distance, optical data links have been investigated as a viable solution to data transfer bottleneck in advanced switching and computer systems [l]-[4]. Similarly to long-haul (> 100 km) telecom m unication system s, the developm ent of these short- distance (1 m - 100 m) optical data links are motivated by the high bandwidth and low-dispersion properties of optical fibers. However, the requirements of optical data links within and among computer and sw itching system s are different from those of the well-established telecommunications systems [3]-[4]. In addition to parallelism, dense packaging, and low manufacturing cost, optical data links have to operate at lower bit-error-rate with smaller power dissipation [3]. In this thesis, we present high-speed parallel optical transmitter array modules 1 for data links with transmission distances up to 200 m. This increased transmission distance opens the door for these parallel transmitters to be used in future high-speed hybrid local loops for broad band digital services to the home [5]-[6]. 1.2 A pplications o f Parallel O ptical Data Links Some of the applications of fiber-optic data links are (i) computer-to- com puter, (ii) board-to-board intrashelf, and (iii) chip-to-chip interconnections. Each of these applications requires different parameters such as bandwidth, power, circuit complexity, and cost [3]. Computer-to-computer interconnects provide communication between central processing units (CPUs) and from CPUs to peripherals. For this category, optical fiber data links are currently being successfully used in high-performance com puter system s [2], [7]-[8]. A t this interconnection level, optical data links are mostly implemented in a serial half-duplex [7] or duplex [8] form and used in several formats, such as the Fiber Distributed Data Interface (FDDI) [9], Synchronous Digital Hierarchy (SDH)/Synchronous Optical Network (SONET) [10], High-Performance Parallel Interface (HIPPI) [11], and the Fibre Channel [12]. 2 The transmission distance for board-to-board interconnects is approximately in the 1-50 m range, making it difficult som etim es to distinguish from computer-to-computer interconnects. Fiber-optic data link products for board-to-board interconnections are used in large sw itching system s, such as the AT&T 5ESS-2000 switch [13]. Other optical backplane interconnects have been dem onstrated in the laboratory using either optical fiber [14]-[15] or free-space as the transmission m edium [16]. In the former case, parallelism has been achieved through multiplexing or the use of multi-fiber cables. The use of optical data links for chip-to-chip interconnection is still in its early experimental stage [17]. With the rapid advance in m ulti-chip m odule (MCM) packaging techniques [18], it is projected that in the intra-substrate level, electrical interconnections w ill be adequate for present and near-future computing and switching systems [3]- In this thesis, we focus on parallel optical fiber data links with transmission distances in the range of 1-200 m and data rates up to 1 G b /s per channel. Such high-speed parallel data links envisage com puters networked at the m icroprocessor bus level and the peripheral input/output level. These parallel data links can also be used in real-time video signal transmission over distances longer than those possible with copper interconnects (> 30 m). Another application 3 may be in future high-speed local hybrid loops linking switching terminals to curbside distribution boxes for broad band digital services to the home. 1.3 Perform ance R equirem ents There are two approaches to realizing optical computer-to-computer interconnections (i) optical serial link based on tim e d ivision multiplexing (TDM) technique, and (ii) optical parallel data link. Serial optical data links have many disadvantages such as fixed bit rate, delay and added com plexity in the parallel-to-serial (P /S) and serial-to- parallel (S/P ) conversion, and potentially higher cost [19]. In one parallel approach, data is transmitted unstaggered in phase together with clocks and parity. Synchronization signals may be sent over som e of the channels, eliminating the need for tim ing recovery circuitry. The requirements for fiber-optic parallel data links can be summarized as [3], [19]: (i) High bandwidth per channel- typically 1 G b/s/channel without data encoding. (ii) Low interchannel timing skew- no more than half a bit period minus the maximum rise and fall times. (iii) Low bit-error-rate (BER) - less than 10‘14. 4 (iv) Low interchannel crosstalk - less than -35 dB at 1 Gb/s. (v) Low power dissipation - less than 125 mW per channel. (vi) Compact, inexpensive, and reliable receiver and transmitter m odules. (vii) Ease of packaging. The above performance requirements summary clearly shows that the requirements of short-haul parallel optical data links are different than those of long-haul telecommunication links. In the latter systems, data is normally encoded to narrow the signal band, and the clock is recovered using a phase-locked loop. In addition, the received optical pow er is normally less than -20 dBm. Moreover, costly high-speed devices such as distributed feedback (DFB) lasers and expensive opto- isolators are required. 1.4 Other W ork In O ptical Transm itter Arrays A variety of light em itting devices have been investigated for im plem enting transmitter array m odules in parallel optical data links. The choice of devise technology depends on the bandwidth-distance product needed, performance requirements, and cost. Light-emitting diodes (LEDs) are incoherent light sources that have attractive features 5 such as low fabrication cost, good temperature stability, and high reliability. However, LEDs have limited m odulation bandwidth and suffer from a very low overall efficiency (electrical power into LED to optical power collected by a photodetector through a fiber) of typically 10"4 to 10"5 [20]. This low overall efficiency is due to tw o reasons (i) LEDs emit light in all directions, resulting in a very inefficient optical coupling, and (ii) their quantum efficiency is between only 1 and 10%. The overall efficiency can be increased by a factor o f 100 w ithout increasing the overall heat dissipation by using a laser diode (LD), instead. Modulation speed in the GHz regime and more relaxed gain requirements on the receiver are gained through the use of LDs. Edge- emitting laser (EEL) arrays technology is currently more mature and reliable than that of surface-em itting laser (SEL) arrays. Uniform, reliable high-speed 10-wide EEL arrays are commercially available [21]. Vertical-cavity surface-emitting laser (VCSEL) technology has evolved quite rapidly during the past few years, and is expected to be a future contender to the w ell-established EEL technology in data link applications [22], Distributed-feedback (DFB) lasers are being used in long-haul telecom m unication system s and are unlikely to be a contender in parallel optical data links due to their com plexity, nonuniformity, and high cost. The fundamental ways of modulating laser arrays are (i) direct current drive (with or without pre-bias), (ii) external cavity active 6 modulation , and (iii) intra-cavity modulation. Lasers directly driven from a logic family (such the em itter-coupled logic (ECL) family) w ithout pre-bias are desirable for digital modulation of laser arrays because it elim inates the need for bias m onitoring and feedback control, w hile permits high-contrast modulation. Digital modulation of low threshold laser arrays driven directly with ECL-voltage levels have been dem onstrated [23]. Research on external cavity active modulators, such as the symmetric self-electrooptic effect device (S- SEED) [24], is still underway to improve their performance in optical m odulation intensity, power dissipation, signal functionality, and optical loss budget. Intra-cavity m odulation is possible w ith three- term inal lasers w ith m onolithically integrated electroabsorption modulator (or gate). The gate section is voltage-modulated while the gain section (or anode) is kept at a constant current. In the rest of this thesis, we w ill refer to this device as laser-gate. Highly-efficient truly ON-OFF modulation at G b/s from a voltage source a laser-gate array have been reported [3]. Optical ribbon fiber is the natural m edium for high-speed parallel optical data links, with its attractive properties such as low loss, low dispersion, low interchannel skew, compactness, and immunity to electrom agnetic interference (EMI). Silica single mode (SM) optical fiber has the advantages of very high bandwidth-distance product and immunity from modal noise. However, SM fiber suffers from stringent 7 optical alignment tolerance and low optical coupling efficiency. Silica m ulti-m ode (MM) fibers, on the other hand, offer a more relaxed optical alignment tolerances (by a factor o f 10) and higher optical coupling efficiency, but have lower bandwidth-distance product than SM fibers. The bandwidth-distance product of graded-index (GRIN) MM fibers is about three times larger than that of a similar size step- index MM fiber. MM plastic fibers are attractive due to their low cost and ease of manufacturing, but more research is needed to improve their loss, skew, and bandwidth-distance performance. It should be A uthors [Ref.] Year #chan Fiber Light Source, X Distance (m) Data Rate (M b /s/c h ) Total Skew(ps) K aede etal. 1261 1990 12 MM LED, 1.3 um 1000 14 6000 Yamanaka eta l. |14| 1991 5 MM EEL,825nm _ 1000 N ordin etal. 131 1992 12 MM EEL, 1.3 um _ 1000 Lew is et al. [271 1992 32 MM SEL, 850nm . 500 1500 Shim izu atal. |28| 1992 4 MM EEL, 1.3 um 26 2000 210 O ta et al. 1291 1992 9 MM LED, 1.3 um 500 NaKahori et al. [19] 1992 12 MM LED, 1.3 um 100 150 1600 Takai et al. (301 1994 8 SM EEL, 1.3 um 100 200 2000 P arker et at. [151 1992 6 MM DFB,1.55um _ 700 Arm iento et al. (311 1992 4 SM EEL, 1.3 um _ 1000 _ K arstensen et al.[231 1994 12 MM EEL, 850 nm 10 1000 200 Horim ntsu eta l. [321 1994 4 SM EEL, 1.55 um 400 1200 _ B ursky [331 1994 10 MM SEL, - 30 150 T ab le 1.1: C om parison o f reported m ulti-channel optical transm itters for parallel data links. 8 noted here that for synchronous parallel optical links, SM fiber loses its intrinsically higher bandwidth-distance advantage when inter channel skew of the ribbon becomes the limiting factor. Table 1.1 show s a summary of some optical transmitter arrays for parallel data links reported in literature. Most of the transmitter arrays in Table 1.1 were obtained by using hybrid packaging technology, rather than opto-electronic integrated circuit (OEIC) technology [25]. It should be noted that low interchannel skew is a very important requirement in synchronous parallel optical data links. The transmitter, receiver, and transmission medium contribute to the total skew of a parallel data link. Inter-channel skew due to variations in propagation delays among the different channels of the ribbon fiber can be the major contributor to the total skew in parallel data transm ission over relatively long distances (> 100 m) [26]. 1.5 W ork Performed We now give a short summary of the present work described in this thesis. The main goal was to design and implement high-speed m ulti channel transmitter array m odules that m eet the perform ance 9 requirements outlined earlier for synchronous digital optical parallel data links. To meet that goal, the plan was to: (i) Experimentally study the dc performance and operational characteristics of very-low threshold single quantum well (SQW) strained InGaAs/GaAs laser-gate, and compare that to similar 2- terminal lasers. (ii) Design, implement, and evaluate single-channel transmitter modules using two and three-terminal lasers. (iii) Design, implement multi-channel transmitter array modules using two and three-terminal lasers. This would lend itself to evaluating and comparing the performance of these transmitter arrays. (iv) Develop a SPICE equivalent circuit model of two and three- terminal SQW strained InGaAs/GaAs lasers that would enable the inclusion of driver circuitry in the simulation of the whole transmitter. The thesis is organized as follows: in Chapter II, after a brief introduction to the InGaAs/GaAs laser technology, we present a detailed study of the dc operation of SQW strained InGaAs/GaAs laser- gates. In Chapter III, w e study the sm all and large-signal digital m odulation of tw o and three-terminal InG aA s/G aA s lasers, and describe the fabrication and packaging of single-channel transmitter modules. In Chapter IV, w e develop a large-signal SPICE equivalent 10 circuit models for those two- and three-terminal InGaAs/GaAs lasers. In Chapter V, w e briefly discuss modal noise in MM fiber systems, and evaluate the performance of an optical MM fiber ribbon. In Chapter VI, w e describe the fabrication and packaging procedures utilized for the implementation of three different multi-channel transmitter modules, present the measured results, and compare their performances. Finally, we draw conclusions from this work in Chapter VII. References [1] J. W. Goodman, F. J. Leonberger, S. Kung, and R. Athale, "Optical Interconnections for VLSI Systems," Proc. IEEE, 72, pp 850-866 (1984). [2] L. D. Hutcheson, P. Haugen, and A. Husain, "Optical Interconnects Replace Hardwire," IEEE Spectrum, March, pp 30-35 (1987). [3] R. A. Nordin, A. F. J. Levi, R. N. Nottenburg, J. O'Gorman, T. Tanbun-Ek, and R. A. Logan, "A Systems Perspective on Digital Interconnection Technology,” J. Lightwave Technol., 10, pp 811-827 (1992). [4] C. S. Harder, B. J. Van Zeghbroeck, M. P. Kesler, H. P. Meier, P. Vettiger, D. J. Webb, and P. Wolf, "High-Speed G aA s/A lG aA s Optoelectronic Devices for Computer Applications," IB M J. Res. Develop., 34, pp 568-584 (1990). [5] C. J. Brunet, "Hybridizing the Local Loop," IEEE Spectrum, June, pp 28-32 (1994). 11 [6 ] A. Cook and J. Stern, "Optical Fiber Access - Perspectives Toward the 21st Century," IEEE Comm. Magazine, 32, no. 2, pp 78-86 (1994). [7] J. R. Lineback, "Optical Fiber Extends Parallel Interface Link," Electronics, September 9, pp 104-105 (1985). [8 ] N. R. Aulet, D. W. Boerstler, G. DeMario, F. D. Ferraiolo, C. E. Hayward, C. D. Heath, A. L. Huffman, W. R. Kelly, G. W. Peterson, and D. J. Stigliani, "IBM Enterprise Systems Multimode Fiber Optic Technology," IBM J. Res. Develop., 36, pp 553-576 (1992). [9] F. E. Ross and J. R. Hamstra, "Forging FDDI," IEEE J. Selected Areas Comm., 11, pp 181-190 (1993). [10] P. F. DeBuck and S. R. Johnson, "The FT-2000 OC-48 Lightwave System," A T & T Technical Journal, January/February, pp 14-22 (1992). [11] W. A ndrew s, "Peripheral Interfaces Offer Fast N etw orking Solutions," Computer Design, June, pp 59-68 (1992). [12] ANCOT Corporation: "What is Fibre Channel?," Menlo Park, CA (1994) [13] B. H. Hornbach, W. J. Bielawski, R. J. Canniff, and P. A. Stiling, "5ESS-2000 Switch: The Next Generation Switching Systems," A T & T Technical Journal, September/October, pp 4-12 (1993). [14] N. Yamanaka, M. Sasaki, S. Kikuchi, T. Takada, and M. Idda, "A Gigabit-Rate Five-Highway GaAs OE-LSI Chipset for High-Speed Optical Interconnections Between M odules or VLSI’s," IEEE J. Selected Areas in Comm., 9, pp 689-697 (1991). [15] J. W. Parker, P. J. Ayliffe, T. V. Clapp, M. C. Geear, P. M. Harrison and R. G. Peall, "Multifibre Bus for Rack-to-Rack Interconnects Based on Opto-Hybrid Transmitter/Receiver Array Pair,” Electron. Lett., 28, pp 801-803 (1992). [16] T. Sakano, T. Matsumoto, K. Noguchi, and T. Sawabe: "Design and Performance of a Multiprocessor System Employing Board-to-board 12 Free-Space Optical Interconnections: COSINE-1," Applied Optics, 30, pp 2334-2343 (1991). [17] K. W. Jelley, G. T. Valliath, and J. W. Stafford, "High-speed Chip-to- Chip Optical Interconnect," IEEE Photon. Technol. Lett., 4, pp 1157- 1159 (1992). [18] G. L. Ginsberg and D. P. Schnorr, "M utichip M odules and Related Technologies," McGraw-Hill, Inc., New York (1994). [19] T. Nagahori, M. Itoh, I. Watanabe, J. Hayashi, H. Honmou, and T. Uji, "150 M b it/s/ch 12-Channel Optical Parallel Interface Using an LED and a PD Array," Optical and Q uantum Electronics, 24, pp S479- S490 (1992). [20] K. Zurl and N. Streibl, "Optoelectronic Array Interconnections," Optical and Q uantum Electronics, 24, pp S405-S414 (1992). [21] M itsubishi Electric Corp.: ML78512 Laser D iodes for Optical Communication Systems, Itami City, Japan. [22] T. C. Banwell, A. C. Von Lehmen, and R. R. Cordell, "VCSE Laser Transmitters for Parallel Data Links," IEEE J. Q uantum Electron., 29, pp 635-644 (1993). [23] H. Karstensen, Ch. Hanke, M. Honsberg, J. Kropp, J. Wieland, and M. Blaser, "PAROLI - High Performance Optical Bus with Integrated Components," LEOS Sum m er Topical M eeting Digest on Integrated Optoelectronics, paper TH1.4 (1994). [24] A. L. Lentine, H. S. Hinton, D. A. B. Miller, J . E. Henry, and J. E. Cunningham, "Symmetric Self-Elecro-Optic Effect Device: Optical Set-Reset Latch," Appl. Phys. Lett., 52, pp 1419-1421 (1988). [25] D. T. N icholas, J. Lopata, N. K. Dutta, and W. S. H obson, "Fabrication and Performance Characteristics of Optoelectronic Receiver and Transmitter Circuits," LEOS Sum m er Topical M eeting Digest on Integrated Optoelectronics, paper W3.2, pp 21-22 (1994). [26] K. Kaede, T. Uji, T. Nagahori, T. Suzaki, T. Torikai, J. Hayashi, I. Watanabe, M. Itoh, H. Honm ou, and M. Shikada, "12-Channel 13 Parallel Optical-Fiber Transmission Using a Low-Drive Current 1.3- pm LED Array and a p-i-n PD Array," J. Lightwave T echnoi, 8 , pp 883-887 (1990). [27] D. K. Lewis, P. J. Anthony, M. Bendett, and J. D. Crow, "The Optoelectronics Technology Consortium (OETC)," IEEE LEOS Newsletter, October, pp 12-13 (1992). [28] F. Shimizu, H. Furuyama, H. Hamasaki, F. Kuroda, M. Nakamura, and T. Tamura, "Optical Parallel Interconnection Characteristics of 4-Channel 2-G bit/s Bit Synchronous Data Transmission Module," in Proc. o f 42nd ECTC, pp 77-82 (1992). [29] Y. Ota and R. G. Swartz, "Multi-Channel 4-G bit/s (500-Mbit/s-per- channel) Parallel Optical Data Link," in Proc. o f ECOC, pp 167-168 (1992). [30] A. Takai, T. Kato, S. Yamashita, S. Hanatani, Y. Motegi, K. Ito, H. Abe, and H. Kodera, "200-M b/s/ch 100-m Optical Subsystem Interconnections Using 8 -Channel 1.3-pm Laser Diode Arrays and Single-M ode Fiber Arrays," ). Lightwave Technoi., 12, pp 260-269 (1994). [31] C. A. Armiento, A. J. Negri, M. Tabasky, R. A. Boudreau, M. A. Rothman, T. W. Fitzgerald, and P. O. H augsjaa, "Gigabit Transmitter Array M odules on Silicon Waferboard," IEEE Trans. Com ponents, H ybrids, and M anufacturing Technoi., 15, pp 1072- 1080 (1992). [32] T. Horimatsu, N. Fujimoto, K. Wakao, and M. Yano, "Optical Parallel Interconnection Based on Group Multiplexing and Coding Technique," IEICE Trans. Electron., E77-C, pp 35-41 (1994). [33] D. Bursky, "Parallel Optical Links Move Data at 3 Gbits/s," Electronic Design, November, pp 79-81 (1994). 14 CHAPTER II Strained InGaAs/GaAs SQW Lasers 2.1 Introduction In recent years, low-threshold high efficiency semiconductor laser arrays have attracted much attention as candidates for the optical transmitter function in high-speed optical com m unication systems. Very low -threshold lasers offer the advantages of low power consumption, small thermal crosstalk, direct modulation from an ECL output buffer, suitability for optoelectronic integration (OEIC technology), and higher temperature stability [l]-[3]. Temperature stability, an important requirement for system designers, can be further enhanced by using wider bandgap material system such as GaAs. GaAs lasers have better temperature stability than InP lasers because they have higher temperature characteristic constant, Tu. The threshold current of conventional semiconductor lasers had been lowered by an order of magnitude in quantum well (QW) lasers. 15 This improvement is due mainly to the physical scaling o f the active region, and not to the quantum confinement of electrons an d holes in QW regions [2]. The current needed for lasing has two components (i) current needed to maintain electron density at the optical transparency level, and (ii) current needed to overcome the total loss in the laser cavity. So, achieving very low-threshold lasers requires a very low loss laser cavity and a material with very low transparency current density [2]. Loss in a single QW (SQW) active region can be reduced by utilizing graded-index separate confinement heterostructure (GRIN SCH), while the transparency current density can be lowered by using strained material for the SQW [2], [4]-[6]. Strained HR-coated GRIN SCH InGaAs/GaAs SQW lasers with threshold currents as low as 145 pA have been recently reported [7]. D espite the fact that very low threshold lasers w ith high efficiency and large characteristic temperature ( Tn) have been achieved in recent years, the use of these devices in parallel optical data links has not been exploited in commercial products. Since the laser diode must be driven from a current source, an intermediate analog circuit (transconductance amplifier) that transforms voltage sign als into current pulses is required. Also, for large-signal digital modulation, the dc current bias of the laser has to be maintained at a predetermined value, normally close to its threshold value. One method that may be used to circumvent these problems is 16 to utilize a laser-gate [1], [8]-[9], The dc characteristics and performance of strained InGaAs/GaAs SQW laser-gates w ill be detailed in this chapter. 2.2 Growth and Fabrication Our low threshold strained InGaAs/GaAs SQW Fabry-Perot (FP) laser arrays are grown on structured substrate using the temperature- engineered growth (TEG) technique [5]. W ith the TEG technique, a buried heterostructure (BH) is obtained in a single growth step, reducing the usually elaborate post-growth processing to a sim ple shallow Zn diffusion for the p-contacts. The sim plicity of the processing permits the fabrication of more uniform laser arrays than those obtained with conventional fabrication techniques. The epitaxial growth of the structure is carried out in a vertical-geometry metal- organic chemical vapor deposition (MOCVD) system. The growth is done on a n + GaAs substrate patterned with double V -groove structures in order to define a narrow mesa oriented along the (Oil). The following is the growth sequence with the growth temperature of each layer given between parentheses: 2000 A thick n-GaAs buffer layer (830 °C), a 2 pm thick n-Alo.6 Gao.4 As bottom n-cladding layer (830 °C), a 17 1500 A AlA Gai.A -As graded layer with x varying from 0 . 6 to 0 . 1 (750 °C), a 30 A thick GaAs spacer (750 °C), a 1 2 0 A thick GaAs spacer (640 °C), a 90 to 1 0 0 A thick Ino.2 5 Gao.7 5 As quantum well (640 °C), a 150 A thick GaAs spacer (640 °C), a top 1500 A graded AlGaAs layer (750 °C), a 0.7 pm p- Alo.6 Gao.4 As top p-cladding layer (780 °C), a 1 pm n-Alo.6 Gao.4 As top n- cladding layer (700 °C), and a 2 0 0 0 A thick undoped GaAs cap layer (700 °C) [10]. The high growth temperature used during the growth of all the layers below the top n-Alo.5 Gao.5 As layer results in a reduction of the active region width and causes the graded region and the p portion of the top cladding layer to pinch off at the corner of the mesa [1 0 ]. Consequently, a good optical confinement and low current leakage are obtained. The lower temperature used during the growth of the top n- Alo.6 Gao.4 As layer and GaAs cap yields an enlargement in the top of the mesa, thereby, facilitating the laser fabrication [10]. After the epitaxial growth, a 1 0 0 0 A thick plasma assisted CVD Si3 N 4 layer is deposited onto the sample. For laser-gate arrays, w indow s are defined over the m esas for CF4 dry etching of the Si3 N 4 in the regions where Zn diffusion is needed for the anode and gate contacts. Then, T i/P t/A u and A u G e/N i/A u are deposited as the p and n contacts, respectively. 18 2.3 DC Operation of InGaAs/GaAs Laser-Gate Laser-gate is a three-terminal laser with a monolithically integrated intracavity modulator, called gate. A schematic drawing of a laser-gate is shown in Fig. 2.1. Very high efficiency intensity modulation (IM) [1], [9], [11]-[13] and frequency modulation (FM) [14]-[15], as well as broad wavelength tuning [16]-[17] have been obtained with QW laser-gates. We have described this enhanced performance as a photonic transconductance [9], while others have termed the behavior as a gain lever effect [8 ]. It is achieved by taking advantage of the highly sublinear behavior of the QW gain with respect to carrier density. The gain of a strained InGaAs/GaAs SQW laser-gate is plotted in Fig. 2.2 as a function of carrier density, n, using the em pirical formula #(/j) = 15X103 In(n/3.25X101 8 ) [18]. When the device is biased above the lasing threshold, the sum of the optical gain in the gate and anode regions clamps at a constant value (equal to total losses). If the anode gain is increased, then the gate region must automatically lower its gain by ejaculating the extra electrons. When the gate region is also pumped, then its optical gain can not be lowered electrically; instead, it is reduced through the emission of photons [ 8 ], [19]. When the anode section is biased by a constant current source, then a small modulation of the gate section results in a large modulation of the optical output power. 19 G ain (X1 0 cm Gate Anode n+ substrate A ctive Region Cathode Fig. 2.1: Schem atic drawing o f a laser-gate. 10 -1 Anode 5 - Gate 0 - -20 2 10 12 4 6 8 1 8 Carrier density (X I0 cm ) Fig. 2.2: G ain versus carrier density for laser-gate. 20 W e now present the experimental dc measurement results of an uncoated SQW strained InGaAs/GaAs laser-gate w ith a QW thickness of 95 A, an active region width W = 1.7pm, gate section length Lx = 40pm , separation distance between anode and gate L ; , = 10pm, and laser cavity length L( . = 250pm. Before metallizing the top contacts of these laser-gates, a 1 . 2 pm thick polyim ide was spun to reduce the capacitance of the device. The electrical isolation between the gate and the anode was measured to be /?; , = 15 kQ. Using the following relation between the laser threshold, /„,, and temperature, T , /„ a cxp(T/T0) (2.1) 14 -F V = 2 .0 V 1 2-- 8 1 0- - 8 - - 1.5 V 1.0 V 0.5 V 6 - - 0.0 V 2 -- 0 2 4 6 8 10 12 14 16 18 20 22 Anode Current, Ia (mA) Fig. 2.3: C W light output power versus anode current at various gate voltages. The dashed curve is for the case where the gate and anode are shorted together. 21 the characteristic temperature, T„, was found to be 150 °K up to 50 °C. Light is em itted in a single longitudinal m ode at an em ission w avelength of X = 970 nm. In Fig. 2.3 w e have plotted the measured light output power per facet versus the anode current, Ia, at various gate voltages, I ' . With the gate and anode sections shorted together, the CW lasing threshold was 3.5 mA, and the external quantum efficiency was 8 8 % (1.12 m W /m A ) as shown by the dashed curve in Fig. 2.3. Abrupt switching behavior is observed for Iu> 9.8 m A for relatively low V K. Interesting to note that the laser-gate recovers its external efficiency after the abrupt turn-on. A better understanding of 0 . 6-1 0 . 5 - 0 . 4 - -§ • 0 . 2 - &0 C 0 ) u u 3 u c u 4 — * ^ - 0 . 3 - 1 4 mA 0 .0 - - 0 . 2 - - 0 . 4 - 14 mA - 0 . 5 - 0.0 0 .2 0.4 0 .6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2 .2 Gate Voltage, VB (V ) Fig. 2.4: Gale current ( Ig) versus gate voltage ( V K) for anode currents Ia - 4 , 6 , 8 , 10. 12, and 14 mA. 22 the laser-gate behavior can be obtained by considering the measured gate current versus voltage (/„-V£ ) for various /„, as shown in Fig. 2.4, and by referring to the equivalent electrical circuit of the laser-gate and its external connections sh ow n in Fig. 2.5. In the measurem ent presented here, the series resistance of the voltage source, Rs is zero, so V g =Vap. For anode currents less than 10 m A, the gate I-V curves show leaky but otherwise normal diode characteristics. At low V g som e current leaks back from the anode to the external gate circuit through the shunt resistance Rp. At higher applied gate voltages, most o f the current from the external voltage source flow s through the gate because of the lower resistance path. For ^ > 1 0 m A and at higher gate voltages, the increased number o f carriers transferred from the anode to the gate can no longer be ejaculated electrically. This results in a negative resistance and the onset of lasing [8]-[9], Va Rp Vg 0 J anode V / V / g a te I external source1 la se r-g a te diode © 'ap external source Fig. 2.5: Equivalent circuit o f the laser-gate (betw een the dashed lines) and the external supply circuit of the d c m easurem ent. 23 For large-signal digital modulation of the laser-gate, the most important parameter is the transfer characteristics, expressed as the dependence of the light output power on the gate voltage (L-V curves) for various anode currents. This is show n in Fig. 2.6. For anode currents above 8 mA, the laser-gate turns-on abruptly as V g is increased, indicating a highly efficient optical switching. For la = 11 mA (19 mW dc power to the anode), the total photonic transconductance is 11 m W /V for 0.85 V < V g < 2 V. At /, = 11 mA and for V K around 1.0 V, about 2.3 mW of output light power is switched using only a 10 pW < u U 0 > £ o 1 4 m A 8 7 ■11mA ■10mA 6 5 8 m A 4 6 m A 3 2 4 m A 1 0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 Gate Voltage, V (V) Fig. 2.6: Light output pow er per facet versus gate voltage at various anode currents. 24 change in electrical power, yielding a power switching ratio of 230. We should note here that, for the current and voltage ranges used in the dc measurements presented above, the laser-gate operated in a stable regime, and neither self-pulsation nor hysterisis effect were observed. All these attractive properties of strained InGaAs/GaAs SQW BH lasers and laser-gates w ould of little practical consequence if these devices are unreliable. This is the topic of the next section. 2.4 Reliability of Strained InGaAs/GaAs Lasers Laser lifetime may be limited by sudden failure due to dark-line defects (DLDs). DLDs are dislocation networks driven by nonradiative recombination and are initiated from material defects or damaged areas on the laser [20]-[21]. Another mechanism that limits the laser lifetime is the intensity-dependent slow degradation at power densities less than those sufficient to cause instantaneous catastrophic failure. The gradual degradation rate of a laser is defined as the percent change in continuous-wave (CW) operating current for a certain light output power per kilohour (kh), w hile laser lifetime is defined as the time to operating current doubling. InP-based long wavelength (1.3 - 1.5 pm) semiconductor lasers have traditionally showed better reliability than 25 GaAs-based shorter wavelength (0.6 - 1.1 gm) lasers [20]-[21]. However, with the recent improvements in GaAs laser technology, the reliability gap is gradually diminishing. Typical uncoated AlG aA s/G aA s SQW lasers (k = 860 nm) exhibit a degradation rate of 8 - 1 0 % /kh and a lifetime of 8000 h [20]. Surprisingly, strained SQW InGaAs/GaAs lasers having a QW thickness below some compositionally dependent critical thickness, exhibit better reliability than AlGaAs/GaAs lasers [20], [22]- [23]. Beernink et al. estim ated this critical QW thickness for an hio 2 5 Gao.7 5 As strained-layer active region to be around 1 2 0 A [24]. Table 2 . 1 contains a summary of the reported results on the reliability of strained InGaAs/GaAs QW lasers. We note that most of the lasers in Table 2.1 are broad-area devices w hich are not suitable for fiber communication applications. Some of the best reliability results show a A u th o r [Ref.] #QW s /k (nm) be (gm ) C o a tin g A ging p o w er (m W ) Aging Temp. (°C) D e g ra d a tio n ( / k h ) A g in g T im e C W (h r) M e d ia l L ife tim e ( h r ) Stutius et al. [30] 1/950 250 90 50 . 120 >2.5x11)3 Fukuda et al. [311 - 60(1 30 50 - lx lO 4 11)5 Mito et al. 1321 2 /9 8 0 - 30 50 - > SxlO3 - Fu et al. [331 - - 10 70 < 10% lx lO 3 - Okayasu et al. [271 1/980 6(H) 10 30 2% 6xl()3 _ Yellen et al. [221 1/1010 6(H) 70 30 1.3% 2X104 5x10* Yellen e ta l. [341 -/1100 6(H) 70 30 < 1 .8 % l.SxlO 4 4x10* Bour et al. [251 1/930 4(H), co a te d 1(H) 30 < 1% lx lO 4 - Fischer et al. [261 1/1010 6(H) 70 - 1.8% 5x103 104 T able 2.1: R ep o rted results on th e reliability o f strained In G a A s/G aA s Q W lasers. 26 degradation rate of ~ 1%/kh (at 70 mW, 30 °C) and an extrapolated lifetime of 50,000 h [22]. Moreover, these lasers showed extremely low sudden failures in unscreened laser populations [22]-[23], [25]-[26]. This increased reliability of InGaAs lasers has been attributed to both dislocation pinning due to the introduction of indium and to elastic strain accommodation in this strained layer material system [22]. Even though facet oxidation in InGaAs/GaAs strained lasers is similar to that of GaAs QW lasers, the former lasers are more reliable because of their higher resistance to DLD propagation [27]-[29]. We have not studied the reliability of our strained InGaAs/GaAs SQW BH lasers and laser-gates in our laboratory. References [1] R. A. Nordin, A. F. J. Levi, R. N. Nottenburg, J. O'Gorman, T. Tanbun-Ek, and R. A. Logan, "A System s Perspective on Digital Interconnection Technology," /. Lightwave Technoi., 10, pp 811-827 (1992). [2] K. Y. Lau, "Ultralow Threshold Quantum Well Lasers," Chapter 4, in Q uantum Well Lasers, P. S. Zory, Ed., San Diego: Academic Press (1993). [3] K. Y. Lau, N. Bar-Chaim, P. L. Derry, and A. Yariv, " High-speed digital modulation of ultralow threshold ( < 1 mA) GaAs single 27 quantum w ell lasers without bias," Appl. Phys. Lett., 51, pp 69-71 (1987). [4] P. J. A. Thijs, J. J. M. Binsma, L. F. Tiemeijer, and T. Van Dongen, "Submilliamp Threshold Current (0.62 mA at 0 °C) and High Output Power (220 mA) 1.5 fim Tensile Strained InGaAs Single Quantum Well Lasers," Electron. Lett., 28, pp 829-830 (1992). [5] N. C. Frateschi, J. S. Osinski, C. A. Beyler, and P. D. Dapkus, "Low- threshold single quantum well InG aA s/G aA s lasers grow n by metal-organic chemical vapor deposition on structured substrates," IEEE Photon. Technoi. Lett., 4, pp 209-212 (1992). [6 ] T. R. Chen, B. Zhao, Y. H. Zhuang, A. Yariv, J. E. Ungar, and S. Oh, "Ultralow Threshold M ultiquantum Well InGaAs Lasers," Appl. Phys. Lett., 60, pp 1782-1784 (1992). [7] H. Zhao, PhD Thesis, University of Southern California (1994). [8 ] K. Y. Lau, "Dynamics of Quantum Well Lasers," Chapter 5, in Q uantum Well Lasers, P. S. Zory, Ed., San Diego: Academic Press (1993). [9] N. C. Frateschi, H. Zhao, J. Elliott, S. Siala, M. Govindarajan, R. N. Nottenburg, and P. D. Dapkus, "Three-Terminal Bistable Low- Threshold Strained InG aAs/G aA s Laser G rown on Structured Substrates for Digital Modulation", IEEE Photon. Technoi. Lett., 5, pp 275-278 (1993). [10] H. Zhao, M. H. MacDougal, N. C. Frateschi, S. Siala, P. D. Dapkus, and R. N. Nottenburg, "High Efficiency InG aA s/G aA s Single- Quantum-W ell Lasers U sing Single-Step Metalorganic Chemical Vapor Deposition", IEEE Photon. Technoi. Lett., 6 , pp 468-470 (1994). [11] K. J. Vahala, M. A. Newkirk, and T. R. Chen, "The Optical Gain Lever: A N ovel Gain Mechanism in the Direct M odulation of Quantum W ell Semiconductor Lasers," Appl. Phys. Lett., 54, pp 2506-2508 (1989). 28 [12] H. -F. Liu, Y. Hashimoto, and T. Kamiya, "Switching Charateristics and Maximum R epetitive Frequency of InGaAsP/InP Bistable Injection Lasers," IEEE }. Q uantum Electron., QE-24, pp 43-50 (1988). [13] A. F. J. Levi, R. N. Nottenburg, R. A. Nordin, T. Tanbun-Ek, and R. A. Logan, "M ultielectrode Quantum Well Laser for D igital Switching," Appl. Phys. Lett., 56, pp 1095-1097 (1990). [14] N. Moore and K. Y. Lau, "Ultrahigh Efficiency Microwave Signal Transm ission U sin g Tandem -Contact Single Quantum W ell GaAlAs Lasers," Appl. Phys. Lett., 55, pp 936-938 (1989). [15] D. Gajic and K. Y. Lau, "Intensity N oise in the Ultrahigh Efficiency Tandem-Contact Quantum Well Lasers," Appl. Phys. Lett., 57, pp 1837-1839 (1990). [16] K. Berthold, A. F. J. Levi, S. J. Pearton, R. J. Malik, W. Y. Jan, and J . E. C unningham , "Bias-controlled intersubband w avelength switching in a GaAs/AlGaAs quantum well laser," Appl. Phys. Lett., 55, pp 1382-1384 (1989). [17] K. Berthold, A. F. J. Levi, T. Tanbun-Ek, and R. A. Logan, "Wavelength Switching in InG aA s/InP Quantum Well Lasers," Appl. Phys. Lett., 56, pp 122-124 (1990). [18] N. C. Frateschi, PhD Thesis, U niversity of Southern California (1993). [19] K. Lau, "Short-Pulse and High-Frequency Signal Generation in Semiconductor Lasers," /. Lightwave Technoi., 7, pp 400-419 (1989). [20] S. L. Yellen, A. H. Shepard, R. J. Dalby, J . A. Baumann, H. B. Serreze, T. S. Guido, R. Soltz, K. J. Bystrom, C. M. Harding and R. G. Waters, "Reliability of GaAs-Based Semiconductor D iode Lasers: 0.6- 1.1pm," IEEE J. Q uantum Electron., 29, pp 2058-2066 (1993). [21] M. Fukuda, "Laser and LED Reliability Update," J. Lightw ave Technoi, 6 , pp 1488-1495 (1988). 29 [22] S. L. Yellen, R. G. Waters, Y. C. Chen, B. A. Soltz, S. E. Fischer, D. Fekete, and J. M. Ballantyne, "20 000 h InGaAs Quantum Well Lasers," Electron. Lett., 26, pp 2083-2084 (1990). [23] R. G. Waters, P. K. York, K. J. Beernink, and J. J. Coleman, "Viable Strained-Layer Laser at X = 1100 nm", J. Appl. Phys., 67, pp 1132-1134 (1990). [24] K. J. Beernink, P. K. York, J. J. Coleman, R. G. Waters, J. Kim, and C. M. W ayman, "Characterization of InG aA s/G aA s Strained-Layer Lasers w ith Quantum Wells Near the Critical Thickness," Appl. Phys. Lett., 55, pp 2167-2169 (1989). [25] D. P. Bour, D. B. Gilbert, K. B. Fabian, J. P. Bednarz, and M. Ettenberg, "Low Degradation Rate in Strained InG aA s/AlG aA s Single Quantum W ell Lasers," IEEE Photon. Technoi. Lett., 2, pp 173-174 (1990). [26] S. E. Fischer, R. G. Waters, D. Fekete, J. M. Ballantyne, Y. C. Chen, and B. A. Soltz, "Long-Lived InGaAs Quantum Well Lasers," Appl. Phys. Lett., 54, pp 1862-1862 (1989). [27] M. Okayasu, M. Fukuda, T. Takeshita, S. Uehara, and K. Kurumada, "Facet Oxidation of In G aAs/GaAs Strained Quantum-Well Lasers," J. Appl. Phys., 69, pp 8346-8351 (1991). [28] M. Fukuda, M. Okayasu, J. Temmyo, and Jun-ichi Nakano, "Degradation Behavior of 0.98-pm Strained Q uantum W ell InG aA s/A lG aA s Lasers Under High-Power Operation," IEEE J. Q uantum Electron., 30, pp 471-476 (1994). [29] R. G. Waters, D. P. Bour, S. L. Yellen, and N. F. Ruggieri, "Inhibited Dark-Line Defect Formation in Strained InGaAs/AlGaAs Quantum Well Lasers," IEEE Photon. Technoi. Lett., 2, pp 531-533 (1990). [30] W. Stutius, P. Gavrilovic, J. E. Williams, K. Meehan, and J. H. Zarrabi, "Continuous Operation of High-Power (200 mW) Strained- Layer G al-xInxA s/G aA s Quantum-W ell Lasers w ith Emission W avelengths0.87 < A < 0.95pm," Electron. L ett., 24, pp 1493-1494 (1988). 30 [31] M. Fukuda, M. Okayasu/T. Takesita, and M. Wada, "Reliability and Degradation of 980 nm InG aA s/G aA s Strained Quantum Well Laser," in Tech. Dig., 2nd European Symp. Rel. o f Electron. Devices, Failure Phys.,Anal., pp 403-410 (1991). [32] I. Mito and K. Endo, "1.48 pm and 0.98 pm High-Power Laser D iodes for Erbium-Doped Fiber Amplifiers," in Tech. D ig., Opt. Am pl. Their Appl., paper WC1 (1991). [33] R. J. Fu, C. S. Hong, E. Y. Chan, D. J. Booher, and L. Figueroa, "High Power, High Temperature InGaAs Strained Quantum Well Lasers," in Tech. Dig., Int. Conf. Lasers Elcctro-Opt., paper CTuA3 (1991). [34] S. L. Yellen, R. G. Waters, P. K. York, K. J. Beernink, and J. J . Coleman, "Reliable InGaAs Quantum Well Lasers at 1.1 pm," Electron. Lett., 27, pp 552-553 (1991). 31 CHAPTER III Small and Large-Signal Modulation 3.1 Introduction Besides their improved dc performance, quantum w ell lasers have many attractive dynamic characteristics such as high modulation speed and low frequency chirp. The higher speed of QW lasers is the consequence of larger differential gain, defined as the derivative of the optical gain with respect to carrier density. The larger differential gain is caused by a modification of the density of states, and therefore, is a consequence of the quantum confinement of the carriers. A factor of 3 - 4 increase in the modulation bandwidth was predicted for QW lasers in comparison with bulk lasers, and an additional increase by a factor of 2 - 3 was predicted for strained QW lasers [1]. The fastest modulation speed demonstrated up to now with strained QW lasers was much lower than what was predicted because the differential gain was not as large as it was first foreseen [l]-[7]. This is due to gain compression, which is the decrease in the optical gain beyond that anticipated from 32 carrier depletion at a high optical output power. Some of the origins of gain com pression are (i) inhom ogeneous broadening in the gain saturation characteristics due to finite intraband scattering time, (ii) finite capture time of carriers from the separate confinement region into the quantum well, and (iii) diffusion of carriers in the separate confinement region [3]-[6]. Practically, however, laser chip parisitics are still the limiting factor in the modulation bandwidth of m ost QW lasers [2], [ 8 ]. Compared with unstrained QW lasers, strained layer QW lasers can reach higher differential gain at a nominal range of optical gain encountered by devices of usual design [1 ]. In the rest of this chapter, we first present results of small-signal measurements of strained InGaAs/GaAs SQW lasers and laser-gates, then, w e discuss the large-signal digital performance of both devices. 3.2 Background on Small-Signal M odulation of QW Lasers The photon and carrier densities inside the laser medium are governed by the rate equations. The single mode rate equations are given by [1]: dt ed t vg(N)s (3.1a) ! y = rvg{N)S- — + PRH I lit T P (3.1b) 33 where N is the carrier density, S is the photon density, r is the optical confinement factor, I is the pump current, v is the group velocity, i? is the volum e of the active region, rs is the recombination lifetim e (radiative and nonradiative) of the carriers, rp is the photon lifetime, P is the fraction of the spontaneous em ission entering the lasing m ode, R is the spontaneous emission rate, e is the electronic charge, and g(N) is the optical gain (unit is cm*1). We introduce here another symbol for optical gain having s " 1 as unit: G = vg(N). Because the electron density does not fluctuate too far from its steady-state value, the gain can be expressed as a linear function of the small-signal deviation of the electron density, n [1 ]: G(N) = — - (3.2) v ' 1 + eS where G „ is the average optical gain, G' is the differential gain, and e is the gain com pression constant. We note here that the dynam ic behavior of QW lasers is dictated by the small difference between the gain and the cavity loss (a few percent of the gain), and not so much by the absolute gain. Thus, even a very sm all am ount of gain compression, eS, can have dramatic effects on the intrinsic dynamic characteristics of QW lasers [1], [9]. For small-signal analysis, we assume sinusoidal m odulation / = /„ +/Vj2’ 9', S = S „ + sei2*", N = N„+nei2*' (3.3) 34 By ignoring the small product terms ns and s2, we obtain the transfer function relating intensity modulation and current [l]-[ 2 ] where f a is approximately fo ~2n G'S y2 T ,,(l + £$,) and f< t is approximated by f , = _eA_ 2 ;rr (3.4) (3.5) (3.6) Equation (3.4) shows that obtaining higher modulation bandwidth can be achieved by (i) increasing the differential gain coefficient, (ii) reducing the photon lifetime by making shorter cavity lasers, or (iii) increasing the photon density by making narrower index-guided waveguides. The effect of the gain compression is illustrated in Fig. 3.1. In Fig. 3.1(a) the modulation response given by Eqn. (3.4) is plotted as a function of frequency for various values of e. The resonance (or relaxation oscillation) frequency, f r is plotted as a function of the square-root of the photon density, -y/jT in Fig. 3.1(b) using the relation [ 2] fr=^fl2- ~ (3-7) As e is increased, the intrinsic frequency response becomes over- 35 ,e = o C Q 2 0 1 1 a, in Pi a .2 4 - * < G 3 T3 O -10 s -15 10« 1 ()‘ > 1011 Frequency (GHz) Fig. 3.1(a): E ffect o f gain com pression on m odulation response o f QW lasers. 3.5 e = 0 1.5 2 2.5 3 xKF 0.5 1/2 1 /2 (Photon Density) (cm-3) Fig. 3.1(b): Effect o f g ain com pression on the resonance frequency o f QW lasers. 36 damped, therefore, reducing the resonance frequency and the -3 dB bandwidth [l]-[4]. Since this reduction is more severe at high optical output power because of the larger eSa, the advantage of a higher modulation bandwidth at high output power is lost. Eqns. (3.5)-(3.7) express the relaxation oscillation frequency as a function of the internal photon density. In practice, however, it is more useful to express it in terms of output power per facet, P a. where, h is Planck's constant, v is the laser light frequency, a m is the distributed mirror loss, and a, is the internal loss. By simplifying Eqns. (3.5)-(3.7) and utilizing Eqn. (3.8), a very useful approximation of the relaxation oscillation frequency is obtained [2], [10]-[11] In Eqn. (3.10) w, d , and Lt, are the w idth, thickness, and length, respectively, of the active region, and Re is the facet reflectivity. It should be noted here that (i) shorter cavity lasers have larger slopes D but smaller maximum output powers, P nm. and (ii) it's the product D ^P nm that should be maximized in the design of a laser. Other / , = (3.9) where, (3.10) approximate (to first order of eSa) relations relating the intrinsic laser bandw idth (w ithout laser electrical parasitics) and its resonance frequencies are [10]-[11] w here f ^ dB and f ( y _ d B are 3 dB and 6 dB electrical frequencies, respectively. So far, w e have only considered the intensity modulation o f the intrinsic QW lasers. However, The ultimate bandwidth is most often limited by electrical parasitics determined by the laser chip geometry and packaging [l]-[2], [10]-[11]. These parasitics considerably affect the current actually reaching the laser active region . The total response of a laser can then be approximated by the product o f the intrinsic intensity m odulation response described above and the transfer function £ from external source to active-region junction current assuming that the lasing region can be represented by a short circuit [10]-[11]. For a laser driven from a 50 £2 voltage source, w e define (3.11) U,w = V3/r (3.12) ^ _ c u rre n t flo w in g in to in trin sic diode voltage o f th e sig n al so u rce 38 A very simple equivalent circuit of laser parasitics is shown in Fig. 3.2, and consists of (i) an inductance L due to the wire bond to the laser chip, (ii) a parallel capacitance C due to the capacitance of the bonding pad and the current confining structure, and (iii) a parallel resistance R due to the p -contact layer, substrate, and contact resistances [2 ], [1 0 ]-[1 1 ]. We assume here that the laser is wire-bonded to a 50 fi microstrip line. R s = 50 n L ■ m n r r \ W\MA— ^ R C r , Fig. 3.2: Simple equivalent circuit o f laser parasitics. J The ratio £ is then (3.13) where, The 3-dB bandwidth of 4 is then f™ - t j 1 2Q2 + { 2 £ > 2 + 4Q4 R = 5 0 1 0.5 GHz I 0.1 20 50 75 0.01 1 0 0 Capacitance, C(pF) Fig. 3.3(a): 3-dB bandw idth contours in th e capacitance-inductance p lan e for R = 5Q and Rs = 5 0 £2. Using Eqns. (3.13)-(3.16) and assuming Rs = 50 Q, we show in Fig. 3.3(a) and Fig. 3.3(b) the calculated contours in the C -L plane of the constant 40 parasitic-limited bandwidth for lasers w ith resistance R of 5 £2 and 15 £2, respectively. Given that the inductance of a 0.5 m m long typical gold wire-bond is about 0.38 nH and utilizing the bandwidth contours of Fig. 3.3(a,b), we observe that designing a 20 GHz bandwidth laser requires a bonding pad capacitance of about 1 pF and 0.5 pF for R = 5 £ 2 and 14 £2, respectively. The bonding pad capacitance can be decreased by reducing the bonding pad area, utilizing a thick layer of low-dielectric constant polyim ide under the pad, and limiting the area o f current confining p-n junctions [ 2 ], [1 0 ]-[1 1 ]. 0.5 GHz 1 0.1 20 50 75 0.01 0.1 1 0 0 Capacitance, C (pF) Fig. 3.3(b): 3-dB bandwidth contours in the capacitance-inductance plane for R = 14 £ 2 and Rx = 50 £ 2 . 41 3.3 Small-Signal M odulation of Strained InGaAs/GaAs SQW Two-Terminal Lasers The small-signal frequency response measurement setup is shown in Fig. 3.4. An HP 8510C Network Analyzer [12] was controlled from a com puter using LabVIEW software [13]. A 26.5 GHz M odular Microwave Package (MMP) [14] with a 3.5 mm RF connector (connected to Port 1 of Network Analyzer) is used as a package for the laser. The laser is epoxied directly onto a gold-plated aluminum block using an electrically conductive silver epoxy [15], providing adequate heat sinking. The laser is biased through the bias-T at Port 1 of the Network HP 8510 C Network Analyzer 3.5 mm Gore cable 3.5 mm Gore cable r v M icroscope HP83040 Modular Microwave Package 3-axis micropositioner Computer with LabView controller H P11982A Lightwave converter „Single-mode fiber pigtail Fig 3.4: Experim ental setup for laser sm all-signal m easurem ent. 42 Analyzer and measured at room temperature. Light from the laser is collected using a lensed single-mode (SM) fiber pigtail. At one end, the fiber is lensed with a hemispheric microlens that allows a maximum coupling efficiency of about 48%. This fiber end is mounted on a XYZ m icropositioner having 2 pm resolution. The other fiber end is connectorized with an Angled Physical Contact FC connector (FC-APC) w ith a return loss > 60 dB [16]. The dc light output power at the connector end is measured using a light meter [17]. The high-speed receiver is an HP Lightwave Converter having a transimpedance gain of about 330 C 2 and a 3-dB bandwidth of 15 GHz with a fairly flat response up to 12 GHz [18]. In the measurement of 3-dB frequency of lasers, corrections were m ade for the small roll-off in the receiver response. The RF output of the Lightwave Converter is connected to Port 2 of the Network Analyzer. In this section, w e present the modulation reponse of a two- terminal strained InGaAs/GaAs SQW laser with a cavity length Lc ~ 225 pm. One facet of this laser is covered with a high-reflection (HR) coating resulting in a reflectivity of 95 %, while the other facet is left uncoated. The laser threshold current /„, is 1.5 mA. The laser is wire- bonded to the 50-fl microstrip line of the MMP with a 0.6 mm long wire-bond. The normalized frequency response of this laser at various current bias points (or light output powers) is shown in Fig 3.5. We notice from Fig. 3.5 a significant damping of the intensity modulation 43 response at high output power levels. A figure of merit called the m odulation-current efficiency factor (MCEF) is som etim es used to characterize the speed of low threshold lasers [19]. MCEF is defined as MCEF = (± 'm - (3.17) v I ~ 4i and can be approximated using Eqn. (3.11) for a parasitic-free laser [7], [19] MCEF = — (3.1 8 ) 2 k \ ed 5 -I 0 - 5 - m T J o - n 5 - 3.8mW - 2 0 - l).58m' 2.4mW 1.48mW 0 2 6 1 2 4 8 1 0 Frequency (GHz) Fig 3.5: N orm alized sm all-signal m odulation response o f a HR coated InG aA s laser for various light output powers. 44 where 7 7, is the internal quantum efficiency. The measured f ^ d B as a function of square root of the incremental current, -\jl - Ilh, above threshold is shown in Fig. 3.6. At low bias increases linearly with ■ yjl - flh , then sublinearly w ith increased bias current. This indicates that intrinsic damping a n d /or parasitics are no longer negligible. A linear fit at low incremental currents (see Fig. 3.6) show s a MCEF value 1 0 - 8 - £ (3 I 6- o c Q ) 3 O ' £ C D ■ O I C O 2 - 0 H 0 1 3 4 1 / 2 , Fig. 3.6: M easured 3-dB m odulation bandwidth v ersu s square root o f increm ental current — Ilh . M CEF is obtained by a linear curve fit at low power levels. of 2.8 G H z/ VmA. Apparently, the highest MCEF reported for strained InGaAs/GaAs SQW lasers is 5 G H z / V m A and was obtained with a HR- coated very short cavity ( L, = 120 nm) laser [19], The 3-dB frequencies 45 are also plotted as a function of the square root of the output power, as shown in Fig. 3.7. A linear curve fit of at low power levels results in a slope (modulation power efficiency) of 4.28 G H z /V m W . This slope is typically 3-5 G H z/ VmW in GalnAsP lasers [2], [20]. Some of the recently reported slopes for InGaAs/GaAs QW lasers are (i) 5.8 G H z/ VmW for 3-QW lasers with Lc = 140 pm [8], (ii) 3.8 G H z/ VmW for SQW lasers with Lc = 400 pm [4], and (iii) 10 G H z/ VmW for SQW lasers with Lt. = 120 pm and HR mirror coating [19]. 1 0 - 8 - N I u ^ 6 - o c a > 3 C T ® j >- 4 — m T 3 I cn 2 - 0 - 2 .5 0.0 0 .5 1.0 1 1 2 (light power) 1 .5 2.0 1 2 , Fig 3.7: M easured 3-dB m odulation bandw idtli versus square root o f output power -y/7^- , w ith a linear curve tit o f at low power levels. 46 The small-signal intensity modulation has revealed a significant roll-off at output powers (from the uncoated facet) larger than 2.5 mW. In order to investigate the origin of this bandwidth limitation, w e need to measure the intrinsic resonance frequency of the laser w ithout the effects of its electrical parasitics. This can be done by using the Relative Intensity Noise (RIN) technique [5], [10], [21]. The laser is packaged the same way as in the intensity modulation experiment, with the laser being biased (CW) from a low noise current source. The output signal of the HP Lightwave Converter is amplified using two B&H amplifiers each having a bandwidth of 10 GHz and a gain of 20 dB [22]. Using a bias-T [23], the amplified signal is then ac-coupled to a RF spectrum analyzer [24]. First, with the laser unbiased (OFF), the noise trace from the different active components in the setup is measured and stored in the spectrum analyzer. This trace is subtracted from the n oise trace obtained w hen the laser is dc biased above threshold. The noise spectrum of the laser peaks at its resonance frequency. This is illustrated in Fig. 3.8(a) where the noise spectrum has a maximum at a frequency / = f r = 5.05 GHz for a laser bias of 6 mA. Strong damping of the relaxation oscillation resonance due to gain compression ( eS0) is obtained with larger bias current, as show n in Fig. 3.8(b). The resonance frequencies m easured u sin g the RIN technique are plotted as a function of square root of the output power in Fig. 3.9. W e notice that f r keeps increasing linearly with increasing , as predicted by Eqn. (3.9), with a slope D = 3.45 GHz/ VmW. 47 > • H T3 N . C O T3 O Q > > ( U ► J < u in O 2 -4 2 1 5.55 10 Frequency (GHz) Fig. 3.8(a): Intensity noise spectrum o f laser biased at 6 m A show ing a peak at the resonance frequency f r = 5.05 GHz. > 73 \ C O 73 O H 0 > m • ^ 4 £ < v £ tIKR 7. 4 0 5 GHz 17 dBm 1 5.55 10 Frequency (GHz) Fig. 3.8(b): Intensity noise spectrum o f laser biased at 12 m A show ing a peak at the resonance frequency f t = 7.5 GHz. 48 As described by Eqn. (3.12), the ratio of the 6 dB electrical frequency to the resonance frequency is equal to -y/3 for all output powers even in the case of gain compression an d /or thermal limitations, however, this ratio would approach zero for large output powers in the case of parasitic limitation [10]. This is indeed the case for our laser as shown clearly in Fig. 3.9 and Fig. 3.10. We observe that at = 0.44VmW, 6 ~ '/B = V3 , but this ratio decreases with increasing 10H LL. m 4 - ^ 6 -d B / OH 0.0 0.5 1.0 2.0 1.5 (Output P ow er)1/2 (m W ,/2) Fig 3.9: L aser resonance frequency as m easured using the RIN technique (circles), and 6 dB electrical frequency as m easured with N etw ork A nalyzer (filled circle), as a function o f . Note that the ratio decreases with increasing 49 ■ y J P ^ , indicating the dom inance of the parasitics in lim iting the intensity modulation bandwidth. In the power range where parasitics are limiting the bandwidth, the ratio can be approximated by [ 10] f ""r (3.19) / f - ~ where fp a r is a constant. This is plotted in Fig. 3.10 for f lm r = 7 GHz. This can be explained by considering that our measured laser had an 3.0 H 2 .5 - 2.0 - 1.0 - 0 .5 - 0.0-1 0.0 2.5 0.5 1.0 1.5 2.0 (Output Power)1/2 (mW 1/2) Fig. 3.10: dependence on (squares) indicating the dom inance o f is plotted for D = 3.45 G H z and f.„r = 7.0 GHz. f parasitics. Also, 50 approximate wire-bond inductance of L - 0.5 nH, a resistance R = 14 £2, and capacitance C ~ 2.5 pF. Using these values and referring to the 3-dB bandwidth contours of Fig. 3.3(b), the parasitics have a bandwidth of about 7 GHz. The bandwidth of our lasers can be improved by (i) reducing the laser series resistance, w hich is currently about 14 Q and m ainly due to the resistance of the p-doped region, (ii) reducing the capacitance by decreasing the size of the bonding pad and using a thick layer of polyim ide under the pad, and (iii) shortening the w ire bond length. 3.4 Sm all-Signal M odulation of InGaAs/GaAs Laser-Gates The rate equations used to describe the photon and carrier densities inside a two-terminal laser (Eqn. 3.1a,b) can be m odified to describe a three-terminal laser [1], [25], [26] (3.20a) dt eel r. (3.20b) ^ = rGM (l-h ) + rGah - y S + f}Rsp L it (3.20c) 51 where Nag, JIIX, xmf,, and G M are the carrier density, current density, recombination lifetime, and optical gain of the anode and gate sections, respectively; while h is the fractional length of the anode region and d is the thickness of the active region. A small-signal analysis similar to that in Section 3.2 gives the follow ing expression for the photon density modulation as a function of angular frequency co [1], [25] rG ;„sAi-> < Vo> + r„)l«‘ {321a) ./, (.!«>)' +(r, + r „ )(j< » )! + a , where a, = [g,„c;„(i -h )+ G llo G;„h]rs„ + yx Y it (3.21b) *2 = K G ^ j A ^ - h ) + G,wG:iliyKh]rS,, (3.21c) y,„ = 7 - + c;,«,4 (3.21d) Gk u i i o and G'„ are the gain and differential gain of the gate and anode sections as illustrated in Fig. 3 .11. An additional relation exists between the gains of the anode and gate sections W hen the anode and the gate are shorted together, th en = G J G . =1. It has been reported that inhomogeneous pum ping of the laser-gate results in an increase in the modulation efficiency as compared w ith hom ogeneous pum ping [1], [25], however, the relaxation oscillation frequency remains alm ost unchanged. This can be explained by considering Eqns. (3.21) in the case of a » yx a [25] s 1 J, (,/'< w )2+<yr 2 w ith (°r2 = [Gk o G 'J I ~ h ) + Ga o G 'a „h] TS,, As illustrated in Fig. 3.11, because the gain function is nearly parabolic w e have GgoG 'g o = GmG'lo, except for extreme values of Gg u . By plotting (3.22a) (3.22b) anode 'ao gate Fig. 3.11: G ain o f laser-gate as a function o f carrier density. 53 Eqns. (3.22), the relaxation oscillation frequency remains the same for h > 0.5 and Ggo/G„ > 0.2 [1], [25]. The modulation response of an uncoated strained InGaAs/GaAs SQW laser-gate was measured using the same experimental setup used for the 2-terminal laser (see Section 3.3). The laser-gate has a total cavity length Lv = 195 pm, gate section length Lg = 40 pm ( h = 0.7), and a separation distance between gate and anode of 10 pm. The laser-gate is packaged in an HP MMP [14], with the gate wire-bonded to the 50-fl microstrip line. The length of this wire-bond is about 500 pm, yielding an inductance L ~ 0.38 nH. An 82 pF and 1 nF bypass capacitors [27] are connected to the dc line of the anode. With the anode biased at a constant current of /„ = 1 1 mA, the gate section is voltage-biased and modulated from Port 1 of the Network Analyzer [12]. The normalized small-signal modulation response of this laser-gate at various optical output powers is shown in Fig. 3. 12. The small kink w e observe in the modulation response around 5.5 GHz is due to the laser package. As w ith 2-terminal lasers, the damping of the response increases with increasing output powers. The 3-dB electrical frequencies / 3 _ ,/B versus the square root of the output power per facet are plotted in Fig. 3.13. The 3-dB bandwidth remains linear with respect to up 7.9 GHz, indicating a higher frequency response as compared with that of the 2- terminal laser discussed in the previous section. This is partially attributed to the lower wire-bond inductance in series with the gate. 54 5 0 - co ■ o -10 * M co 0.25m' 2.25mW 1.0m' 1,5m' -20 -2 5 0 2 8 4 6 10 12 Frequency (GHz) Fig. 3.12: N orm alized sm all-signal m odulation response o f laser-gate for various optical output pow ers/facet. The anode current is 11 mA. 1 0 - 9 - 8 - 7 - > . 6 - IT 2 - o H 0.0 1.0 1 2 , 0 .5 1 .5 (p ow er/facet) Fig. 3.13: M easured 3-dB m odulation bandw idth versus w ith a linear curve tit o f at low pow er levels. 55 The fact that the modulated section is now smaller does not necessary im prove the parasitics bandwidth because, while the capacitance C decreased, the resistance R increased, keeping the RC product almost the same. 3.5 Large-Signal Digital M odulation of Two-Terminal InGaAs/GaAs Lasers Predicting the large-signal modulation performance of semiconductor lasers based on their small-signal modulation characteristics is difficult. For instance, since the small-signal modulation depends very strongly on the bias current, the large-signal impulse response is not simply the Fourier transform of the small-signal response [2]. Particularly, if the drive signal has significant components at or above the "one" or "zero" level resonance frequency, or if the "zero" level is at or below the laser threshold, then the large-signal performance is hard to predict [9]-[ll]. A lso, it has been observed that w hen the m odulation depth, m , increases, the output pow er does not vary sinusoidally under sinusoidal modulation, but rather becomes pulse-like [11], [28], and thus the relaxation oscillation resonance occurs at a lower frequency than that predicted by sm all-signal analysis. In a w orst-case calculations, it has been reported that under large-signal modulation, 56 the relaxation oscillation frequency is reduced by a factor of 0.7 for a m odulation depth m = 70% and drops to 0.6 times the small-signal resonance frequency when in ~ 100% [28]-[29]. Two of the m ost important characteristics of large-signal modulation of lasers are the turn-on delay and the pattern effect. Turn on delay (rtl) is the finite time delay betw een the onset of the current pulse and the onset of lasing. This time delay is equal to the time required for the electron population to reach its threshold level. A ssum ing a constant carrier lifetime rv , the turn-on delay of a laser biased close to its threshold is approximately given by [9] f" = T ' T f r (3-23) ‘ ‘ih w here / is the injection current, 7 ,; , is the threshold current, and I„ is the bias current. So, the turn-on delay can be reduced by increasing the laser prebias. An equally serious problem in the large-signal modulation of lasers is intersymbol interference (ISI) or pattern effect. In digital modulation, the starting condition for each pulse is affected by the charge left over from the previous pulse. Specifically, following an optical pulse, the carrier density stays at slightly below the threshold level and decays with a time constant equal to the carrier lifetime. Therefore, the second of tw o consecutive optical pulses will have a larger amplitude [30]. 57 The m ost important formats of pulse code m odulation in current optical communication systems are non-return-to-zero (NRZ) and return-to-zero (RZ) modulation formats. With NRZ modulation, the laser output power stays at the same level (high or low) for the total clock period. However, in RZ format a "one" transmission consists of a transition from OFF to ON to OFF again. So, the RZ format has the disadvantage of requiring twice the bandwidth of that needed in NRZ m odulation. Throughout the rest of the thesis, w e consider NRZ format only. Also, in the rest of the thesis, and unless specified otherwise, the receiver used in the digital modulation experiments consists of a front- illu m in a ted Ino.5 3 Gao. 4 7 As p-i-n photodetector [31] and a heterojunction bipolar transistor (HBT) transimpedance amplifier (TIA) fabricated in the TRW Inc. foundry [32]. The TIA has a transimpedance of 65 dBQ and a 3-dB bandwidth of 2.3 GHz [33]. The receiver has a sensitivity of -26.5 dBm for a BER of 10" 9 at a wavelength of 980 nm [33]. This receiver was designed specifically for optical data link applications [34]. The circuit schematic of an optical transmitter with a strained InG aA s/G aA s SQW 2-terminal laser is as shown in Fig. 3.14. The common-base laser driver consists of a single PNP silicon bipolar transistor (fj~ 4GHz)[35], with the laser cathode (bottom side) connected 58 F ig. 3. 14: O ptical transm itter consists o f a InG aA s 2-terniinal laser anti a com mon- base driver. to Vcc- The use of an PNP instead o f an NPN transistor is dictated by the fact that our laser arrays have a common cathode (p-region up). This is a disadvantage because presently PNP transistors are considerably slower than NPN transistors. The single supply line is bypassed by an 82 pF and a 1 nF bypass capacitors [27], while the input of the transmitter is impedance-matched using a 47 Q. chip resistor [36]. In Fig. 3.15, we show qualitatively the effect of the laser prebias on the its turn-on delay. Using an stream of input pulses at 1 G b/s w ith a high level voltage of 920 mV (corresponding to /, ~ 12 mA), the input low- level voltage is swept from zero to 530 mV, corresponding to a current sw eep through the laser, /,. from zero to 3 mA (the laser threshold). It should be noted here that the ~ 225 ps change in turn-on delay shown 59 in Fig. 3.15 is the combined turn-on delay of both the PNP transistor and the laser. > T 3 > 6 C O 350 p s/d iv Fig. 3.15: Turn-on delay o f the transm itter as the low level o f the input voltage is swept from 0 to 520 m V ( / from 0 to 3 mA) w ith a high level at 920 mV. The eye-diagram s show n in Fig. 3.16 were obtained using a two- terminal laser having a threshold current = 4.5 mA. The output of a 3 G b /s data generator [37] is ac-coupled using a bias-T [16] to the input of the optical transmitter. The bias-T, which has a low frequency cutoff of 10 KHz, is utilized to pre-bias the input. The output light from the laser is coupled into a lensed 8-m long graded-index (GRIN) multi-mode (MM) fiber having a core and cladding diameters of 62.5 pm and 125 pm, respectively. The output of the receiver is displayed on a 20 GHz digitizing oscilloscope [38]. The input to the transmitter consists of a NRZ 215 -1 pseudo-random bit sequence (PRBS) with a low and high- voltage levels of 0.93 V and 1.46 V, respectively. As shown in Fig. 3.16, 60 500 ps/div (b) 300 jpsZdiv 200 ps/div (d) Fig. 3.16: Eye diagram s 2-term inal la-ter transm itter obtained with an input voltage sw ing o f 530 mV and a 2 15-1 NRZ P R B S at (a) 5(X) M b/s, (b) 1 Gb/s, (c) 1.5 G b/s, and (d) 2 Gb/s. 61 a sinusoidal but otherwise open eye is obtained at a data rate of 2 G b/s. Having the 2-terminal laser biased below threshold w ould increase the extinction ratio, but also results in a power penalty w hich arises from increased pattern effect, turn-on delay, and timing jitter. Timing jitter of lasers involves predominantly the spontaneous recombination in the active region [28]. The jitter can be measured using a 20 GHz digitizing oscilloscope [38] that performs statistical analysis on the distribution of jitter. A ssum ing normal distribution, the standard deviation, a is the m ost accurate way to characterize jitter [39]. In Fig. 3.17(a,b) we show the eye-diagrams and the cross-over timing jitter distribution of the same transmitter at 1 G b /s (217-1 PRBS) for two different bias currents. Slower transitions and integration of the relaxation oscillation transients are evident in the eye-diagrams of this figure because a larger area (80 pm diameter) p-i-n photodiode was used, thus reducing the receiver bandw idth. A lso, during this experiment, no bias-T was used at the input of the transmitter; instead, the output of a 1 G b/s data generator [40] w as directly dc-coupled to the transmitter's input. When the laser is m odulated from %(/,/,) to 4.2(llh), the jitter standard deviation is a = 46 ps. W hile keeping the same high level, increasing the current lo w level to causes < x to drop to 31 ps, as shown in Fig. 3.17(b). 62 500 ps/div I t > e o o r o 70 ps/div (a) ■ __ 1 ..... J t t r " 7 _____ : -1 — ► • * + — 1 - .-+¥+-t» — -------» ■ - I ,,.,- - ., 500 ps/div 65 ps/div (b) Fig. 3.17: Transm itter Eye-diagram and jitte r distribution (at 1 Cib/s 2 ^ - 1 PRBS) for a m odulating current high level o f 4 .2 /,h an d low level o f (a) % ( /,/,) an d (t>) hr 63 We also investigated the performance of the transmitter without any laser prebias (current through laser is zero in the OFF state). Shown in Fig. 3.18 are the eye-diagrams obtained w ith a 500 M b /s 217-1 PRBS input data for a current high level of 4.2Ilh (top trace) and 7.5Ilh. Despite the obvious problem with the duty-cycle, the eye opening is w ide enough to make this bias-free modulation useful at data rates up to 500 M b/s, but probably unacceptable at higher bit rates. > ‘ d s . > E o o 1 ns/div (a) > > E o L O 1 ns/div (b) F ig. 3.18: B ias-free m odulation of 2-term inal laser at 500 Mb/s (2 17-1 N RZ PRBS) for a current high level o f (a) 4 . 2 and (b ) 7 .5 /(/l. 64 3.6 Large-Signal D igital M odulation of InG aAs/G aAs Laser- Gates A s discu ssed in the previous section, fo r large-signal digital m odulation of low-threshold two-terminal laser at G b /s , the laser current dc bias has to be maintained at a valu e close to its threshold current. A lso, the laser has to be driven from a current source, so an intermediate analog circuit that transforms voltage signals into current pulses is required, adding com plexity an d increasing pow er consumption. One w a y that may be utilized to avoid so m e of these problems is to use a sem iconductor laser with a m onolithically integrated intracavity absorber, or laser-gate [1], [41]-[43], The gate can be m odulated from a low-power voltage source, enabling ON-OFF switching of milliwatts of optical power w ith an ECL in p u t voltage swing. In this section, w e present the digital modulation performance of a low-threshold uncoated strained InGaAs/GaAs SQW laser-gate w ith a cavity length Lr of 250 pm, a gate section length LM o f 40 pm, a separation distance from gate to anode Ls o f 10 pm. T h e isolation resistance between the gate and anode Rjt is measured to be 15 kO, w hile the characteristic temperature T 0 is 150 °K. The dc performance of this laser-gate was presented in Section 2.3. The drive circuit used for the digital modulation of our laser-gate consists of an emitter-follower using a single silicon bipolar NPN transistor ( f T ~ 10 GHz) [44]. The 65 Output Light Power (mW /facet) CC N P N V in A n o d e G a te 50 ft 200 ft Fig. 3.19: Circuit diagram o f optical transm itter. 8 -I 7 - 6 - 5 - 4 - 3 - 2 - - 1.8 - 1.6 -1 .4 1 .2 1.0 Input Voltage (V) Fig. 3.20: T ransm itter's optical output pow er versus input voltage for anode current of 13.5 mA. 66 circuit diagram of the optical transmitter is show n in Fig. 3.19. Light from the laser-gate is collected by coupling to a 16-m long lensed 62.5/125 pm GRIN MM fiber. The dc transfer characteristics of the transmitter, shown in Fig. 3.20, indicate that 6.3 mW of optical power is switched per facet for an input voltage swing at the base of the NPN transistor of Av,n = vin lo w - vn lli),h of 800 mV (ECL swing) and a constant anode current /„ of 13.5 mA. For the same anode current, the transmitter was modulated with a NRZ PRBS input data of length 217- 1. Fig. 3.21 shows the eye-diagrams obtained at (from top to bottom) 500 M b/s, 1 G b/s, 1.5 G b /s, and 2 G b/s for an input voltage swing of 800 mV. At these settings, the laser-gate has a modulation efficiency of about 27%. Wide open eyes are obtained at data rates up to 2 G b/s. The noise-free bottom rails of the eye-diagrams indicates that lasing practically ceases completely in the OFF state. The eye-diagrams are a very valuable tool to qualitatively evaluate the performance of digital transmission. However, eye-diagrams are acquired using oscilloscopes which can not reveal very low probability (say 1 in 1011) events [45]. A method to quantitatively analyze eye-diagrams is to measure the bit- error-rate (BER), which is the ratio of bits received in error to total received bits. The BER measurement setup is depicted in Fig. 3.22. The output of a pattern generator [37] is connected to the input of the optical transmitter. The light output from the laser-gate is coupled into a 67 lOOmV/div lOOmV/div 85mV/div 85mV/div 1 ns/div 500 ps/div 300 p s/div 200 p s/div Fig. 3.21: Eye-diagram s for a 2 l7- l N R Z PRBS data pattern after transm ission through 16 m o f MM optical fiber. The input signal sw ing is 0.8 V at a d ata rate of (top to bottom ) 5(H) M b/s, 1 G b/s, 1.5 Gb/s, and 2 Gb/s. 68 62.5/125 pm MM fiber and split equally into two other fiber cables. The light output from one fiber is coupled into a receiver. The receiver Error Detector D isp la y Pattern Generator Clock Source LK DATA TRIG 18 GHz SMA cable / | J 'V /.fv P H K Transmitter 3-dB splitter Receiver MM fiber Power Meter Fig. 3.22: Experim ental setup for the B E R m easurem ent. output is, in turn, fed into an error detector [46]. The other fiber allows us to monitor the average light power using an optical power meter [18]. The BER was measured at 1 G b/s w ith a 27-l NRZ PRBS as a function of the input voltage sw ing at the base of the N PN transistor and is show n in Fig. 3.23. The anode current was 13.5 mA. By extrapolating the linear fit of the data, a BER less than 10'14 is achievable with an input voltage swing of just 125 mV, which is substantially lower than present ECL noise margins. However, there 69 ' " " V 1-10 -1 2 -13 < - 1 4 100 110 120 130 Input S ignal Swing (m V ) Fig. 3.23: BER versus input signal swing at th e transm itter m easured w ith a 2 7-l NRZ PRBS data pattern at 1 Gb/s. was a noise floor to the measurement at 3X10’13. This was caused by burst-errors after several hours of error-free operation. These error bursts occurred even during a purely electrical measurement of the receiver's TIA alone. The cause of these errors is not known at present. Another important system parameter in digital transmission is the phase margin (PM), which quantifies both the crossover timing jitter and the duty cycle of the received data. If a bit duration is T, and the tim e interval AT w ithin a single eye opening (one bit) during which the BER is less than a certain value, then the phase margin at that BER value is given by 70 PM = — 360 T (3.24) We measured the phase margin of the transmitter versus the input voltage swing at 1 G b/s NRZ PRBS of length 27-l for a BER of 10'11 as show n in Fig. 3.24. In this measurement, the data was transmitted through 16 m of GRIN MM optical fiber connectorized with a FC/PC connector, and the anode current was kept at 13.5 mA. A phase margin of 255° was obtained at an input voltage swing of 800 mV. 3 0 0 - I , = 13.5mA 200 - 150 — ioo H o.o 0.2 0 .4 0.8 1.2 0.6 1.0 In p u t V oltage S w in g (V) Fig. 3.24: Phase margin after transmission through 16 m of MM fiber versus input voltage swing at a BER of 1C )-1 1 for a 1 Gb/s, 27-l NRZ PRBS data pattern. 71 Also, for the same input voltage sw ing of 800 mV, the same anode current of 13.5 mA, and after transmission through 16 m of MM fiber, w e measured the phase margin versus the data rate for a BER of 10-*1, as shown in Fig. 3.25. Phase margins of 325° and 150° were obtained at data rates of 500 M b/s and 2 G b/s, respectively. A curve fit of the data points indicates a linear decrease of phase margin with data rate. L = 13.5mA 3 5 0 - 3 0 0 - e 2 5 0 - 3) w C Q s : « 2 0 0 - t n a t J 3 a. 1 5 0 - loo-j 0 .5 2.5 1.0 1.5 2.0 D ata R ate (G b/s) Fig. 3-25: Phase m argin after transm ission through 16 m o f MM fiber versus data rate at a B ER o f 1()-*1 for an input voltage sw ing o f 0.8 V (27- l NRZ PRBS data pattern). The success of the large-signal digital m odulation of these InGaAs/GaAs laser-gates eliminates the need for a separate laser driver 72 stage. In actual data link applications, a register flip-flop precedes the transmitter to perform signal retiming, as depicted in Fig. 3.26. The output stage of the register flip-flop can then directly drive the laser- gate. Moreover, another important characteristic of laser-gate is its higher temperature stability as compared with 2-terminal devices. This w ill elaborated on in Chapter VII. N P N G a te C lock Register Flip-Flop 200Q -5.2V -2V Fig. 3.26: Laser-gate use in an actual data link transmitter. 73 3.7 D esign and Fabrication of Single-Channel Transmitter M odules O ptical fiber com m unication system s require properly protected sem iconductor laser diodes, efficient coupling of light from these sources into the optical fiber, and electrical connections in the package w h ich allow high-speed m odulation. W e have d esign ed and im plem ented modules for single-channel transmitters incorporating tw o and three-terminal lasers. These transmitter m odules operate at data rates up to 1.5 G b/s, and are currently being used in our laboratory for various experiments. The laser bar and the electronic components for its driver are attached onto a 5 mm X 5 mm patterned alumina ceramic substrate having a thickness of 250 pm [47]. The metallization on these substrates consists of 400 A TiW, 1000 A Ni, and 1.27 pm Au. The various dies were attached using a tw o-com pound electrically conductive silver epoxy [15]. This epoxy is cured at 120 °C for 15 minutes. Patterning of the substrates was achieved by first transferring the design pattern from a photom ask to the substrate u sin g photolithography process, then chemically etching aw ay the excess metallization. Photomask designs were performed using ACAD [48]. The dies were placed on the ceramic substrate using a "pick&place" m achine [49]. After curing, wire-bonding w as achieved using an ultrasonic wire-bonder [50]. 74 A photograph of the com pleted ceramic substrate of the transmitter m odule is shown in Fig. 3.27. This transmitter uses a 2- terminal strained InG aA s/G aA s SQW laser. The laser driver is a common-base amplifier as show n in Fig. 3.14. UV-cured epoxy Lensed fiber-. 8 2 pF capacitor 'MP tra n sisto r i resisto r 1 nF capacitor Ground pads Fig. 3.27: Photograph o f the com pleted ceramic board o f an optical transm itter w ith a 2-term inal laser. The ceramic substrate is fitted inside a square cutout made in a 2.54 cm X 1.27 cm motherboard, as show n in Fig. 3.28. This microwave board consists of a 250 pm thick Duroid with 0.635 mm thick brass backing [51]. The motherboard, in turn, is fixed inside a gold-plated brass housing. The pattern on the motherboard consists of one 50 £2 75 microstrip transmission line for the input signal, several dc supply lines, and a few ground pads. A SMA launcher [52] is used for the high speed electrical connection to the input of the transmitter, and a dc pin [53] is used for VC c supply, as shown in Fig. 3.28. Fig. 3.28: Photograph show ing the different parts o f a transm itter m odule incorporating a 2-term inal InGaAs laser. Optical fiber pigtailing is the last step in the transmitter module fabrication. W hile the transmitter is being modulated at 1 G b/s, the lensed end of a 62.5/125 pm optical fiber is actively aligned to the laser. The hemispherical microlens at the fiber tip has a radius of curvature of 30-40 pm, w ith a typical coupling efficiency of about 75%. The fiber is then fixed in place using a fast-acting (~ 25 s curing time) UV-cured epoxy (see Fig. 3.27 and Fig. 3.28) [54]. A brass tube is utilized to provide strain relief for the optical fiber. D uroid m other boao >MA connector Ceram ic board G old-plated Brass housing 76 Similar procedure and design are also used to implement single channel transmitter modules w ith the InGaAs/GaAs laser-gates. The ceramic substrate of such a transmitter is shown in Fig. 3.29. Here, the laser-gate is driven from a sim ple emitter-follower stage as show n in Fig. 3.19. UV-cured epoxy Lensed fiber \ i \ Ground pac NPN tran sisto r Anode suppl; 50 O 1 nF Fig. 3.29: Photograph o f the com pleted ceram ic board of an optical transm itter w ith InG aA s laser-gate. Exactly the same motherboard is used in the laser-gate transmitter module. A s depicted in Fig. 3.30, an extra dc feed-through is used for the anode current. 77 Fiber strain relief G old-plated Brass housing Duroid ( m other boaord' Ceram ic board A node currenl supply SMA connector Fig. 3.30: P hotograph show ing th e different parts o f a transm itter m odule incorporating an InG aA s laser-gate. A photograph of a com pleted optical transmitter m odule is show n in Fig. 3.31. The transmitter's optical fiber is connectorized with a FC/PC connector. As will be discussed in Chapter V, these relatively inexpensive fiber connectors provide an excellent return-loss performance. Both types of single-channel transmitters performed very well at data rates up to 1.5 G b/s. However, w e found that the reliability and stability of the UV-cured epoxy in some of the modules were not satisfactory. So, w e have sought a more robust solution to fixing the lensed fiber-end in place. First a small section of the fiber end is m etallized (T i/A u m etallization) in a electron-beam evaporation chamber. Then, the metallized section is soldered to a metallized pad 78 on the ceramic substrate. This method will be implemented in future optical transmitter modules. Fiber strain relief SM A connector Fig. 3.31: Photograph o f the com pleted transm itter module. References [1] K. Y. Lau, "Dynamics of Quantum Well Lasers," Chapter 5 in Q uantum Well Lasers, P. S. Zory, Ed., San Diego: Academic Press (1993). [2] J. E. Bowers and M. A. Pollack, "Semiconductor Lasers for T eleco m m u n ica tio n s," C hap ter 13 in O p tic a l F iber Telecom m unications II, S. Miller and I. Kaminow, Eds., Academic Press (1988). MM fiber F C /P C connector 79 [3] Y. Arakawa and T. Takahashi, "Effect of Nonlinear Gain on Modulation Dynamics in Quantum-Well Lasers," Electron. Lett., 25, pp 169-170 (1989). [4] R. Nagarajan, T. Fukushima, S. W. Corzine, and J. E. Bowers, "Effects of Carrier Transport on High-Speed Quantum Well Lasers," Appl. Phys. Lett., 59, pp 835-837 (1991). [5] W. F. Sharfin, J. Schlafer, W. Rideout, B. Elman, R. B. Lauer, J. LaCourse, and F. D. Crawford, "Anomalously H igh Dam ping in Strained InGaAs-GaAs Single Quantum Well Lasers," IEEE Photonics Technol. Lett., 3, pp 193-195 (1991). [6] R. Nagarajan, M. Ishikawa, and J. E. Bowers, "Effects of Carrier Transport on Relative Intensity Noise and Critique of K factor Predictions of Modulation Response," Electron. Lett., 28, pp 846-848 (1992). [7] B. Zhao, T. R. Chen, Y. H. Zhuang, A. Yariv, J. E. Ungar, and S. Oh, "High S p eed O peration of Very Low Threshold Strained InGaAs/GaAs Double Quantum Well Lasers," Appl. Phys. Lett., 60, pp 1295-1297 (1992). [8] R. Nagarajan, T. Fukushima, J. E. Bowers, R. S. Geels, and L. A. Coldren, "High-Speed InGaAs/GaAs Strained Multiple Quantum Well Lasers with Low Damping," Appl. Phys. Lett., 58, pp 2326-2328 (1991). [9] G. P. Agrawal and N . K. Dutta, Long-W avelength Sem iconductor Lasers: N ew York: Van Nostrand Reinhold (1986). [10] J. E. Bowers, B. R. Hemenway, A. H. Gnauck, and D. P. Wilt, "High- Speed InGaAsP C onstricted-M esa Lasers," IEEE J. Q uantum Electronics, QE-22, pp 833-843 (1986). [11] K. Y. Lau and A. Yariv, "High Frequency Current Modulation of Sem iconductor Injection Lasers" in S e m ic o n d u c to r s a n d Semimetals: Lightw ave C om m unication Technology, W. T. Tsang, vol. Ed., Vol. 22B, pp 69-152, London: Academic Press (1985). [12] Hewlett-Packard Inc., HP 8510C Network Analyzer, Santa Rosa, CA. 80 [13] LabView, National Instruments, 6504 Bridge Point Parkway, Austin, Texas, 78730. [14] Hewlett-Packard Inc., HP 83040 Series Modular Microcircuit Package, Santa Rosa, CA. [15] Epoxy Technology, Inc.: H20E Electrically Conductive Epoxy, Billerica, MA. [16] Picosecond Pulse Labs., Inc.: 5575A Bias-T, Boulder, CO. [17] RIFOCS Corporation, FC-APC pigtail assembly, Camarillo, CA. [18] Hewlett-Packard Inc., HP 8153A Lightwave Multimeter, Santa Rosa, CA. [19] Hewlett-Packard Inc., HP 11982A Lightwave Converter, Santa Rosa, CA. [20] T. R. Chen, B. Zhao, L. Eng, Y. H. Zhuang, J. O'Brian, and A. Yariv, "Very H igh Modulation Efficiency of Ultralow Threshold Current Single Quantum Well InGaAs Lasers," Electron. Lett., 29, pp 1525- 1526 (1993). [21] J. E. Bowers, B. R. Hemenway, A. H. Gnauck, T. J. Bridges, E. G. Burkhardt, D. P. W ilt, and S. M aynard, "High-frequency Constricted-Mesa Lasers," Appl. Phys. Lett., 47, pp 78-80 (1985). [22] S. Weisser, J. D. Raltson, E. C. Larkins, I. Esquivias, P. J. Tasker, J. Fleissner, and J. R osenzw eig, "Efficient H igh-Speed Direct Modulation in p-Doped InGaAs/GaAs Multiquantum Well Lasers," Electron. Lett., 28, pp 2141-2143 (1992). [23] B&H Electronics Corp.: DC10000H and AC9020 H igh-speed amplifiers, Monroe, NY. [24] Hewlett-Packard Inc., HP 8563E 9kHz-26.5GHz Spectrum Analyzer, Santa Rosa, CA. 81 [25] N. Moore and K. Y. Lau, "Ultrahigh Efficiency Microwave Signal Transm ission U sing Tandem -Contact Single Q uantum W ell GaAlAs Lasers," Appl. Phys. Lett., 55, pp 936-938 (1989). [26] D. Gajic and K. Y. Lau, "Intensity Noise in the Ultrahigh Efficiency Tandem-Contact Quantum W ell Lasers," Appl. Phys. Lett., 57, pp 1837-1839 (1990). [27] Dielectric Labs., Inc.: dicap© M icrowave Ceramic Capacitors Cazenovia, NY. [28] B. W. Hakki, F. Bosch, S. Lumish, and N. R. Dietrich, "1.3-pm BH Laser Perform ance at M icrowave Frequencies," /. Lightw ave Technol., LT-3, pp 1193-1201 (1985). [29] T. Ikegami and S. Suematsu, Electron. Comm. Jpn., B53, pp 69 (1970). [30] K. Y. Lau, N. Bar-Chaim, P. L. Derry, and A. Yariv, " High-speed digital m odulation of ultralow threshold ( < 1 mA) GaAs single quantum w ell lasers without bias," Appl. Phys. Lett., 51, pp 69-71 (1987). [31] Fermionics Opto-Technology Inc.: FD60S4 and FD80S2 InGaAs photodetectors, Simi-Valley, CA. [32] M. E. Kim, A. K. Oki, G. M. Gorman, D. K. Umemoto, and J. B. Camou, "GaAs HBT D evice and IC Technology for H igh- Performance Analog and M icrowave Applications," JEEE Trans. Microwave Theory and Tech., 37, pp 1286-1303 (1986). [33] M. Govindarajan, S. Siala, and R. N. Nottenburg, "DC to 2.5 G b/s x 4 p-i-n/H BT Optical Receiver Array with Low Crosstalk," IEEE Photon. Technol. Lett., 5, pp 1397-1400 (1993). [34] M. Govindarajan, Ph.D. Thesis, University of Southern California (1994). [35] NEC California Eastern Laboratories: NE 88900 PNP Silicon High Frequency Transistors, Santa Clara, CA. 82 [36] IMS, inc.: RC-1-0302SG Chip Resistors, Portsmouth, RI. [37] Hewlett-Packard Inc., HP 70841B Pattern Generator (0.1-3 G b/s), Santa Rosa, CA. [38] Hewlett-Packard Inc., HP 54120B 20 GHz Sampling Oscilloscope, Santa Rosa, CA. [39] Hewlett-Packard, "Jitter Analysis Using the HP 54120 Family of Digitizing Oscilloscopes," Product Note 54120-2 (1989). [40] Hewlett-Packard Inc., HP 80000 Data Generator System, Santa Rosa, CA. [41] A. F. J. Levi, R. N. Nottenburg, R. A. Nordin, T. Tanbun-Ek, and R. A. Logan, "M ultielectrode Quantum W ell Laser for D igital Switching," Appl. Phys. Lett., 56, pp 1095-1097 (1990). [42] R. A. N ordin, A. F. J. Levi, R. N. Nottenburg, J . O'Gorman, T. Tanbun-Ek, and R. A. Logan, "A Systems Perspective on Digital Interconnection Technology," J. Lightwave Technol., 10, pp 811-827 (1992). [43] N. C. Frateschi, H. Zhao, J . Elliott, S. Siala, M. Govindarajan, R. N. Nottenburg, and P. D. Dapkus, "Three-Terminal Bistable Low- Threshold Strained InG aA s/G aA s Laser G rown on Structured Substrates for Digital Modulation", IEEE Photon. Technol. Lett., 5, pp 275-278 (1993). [44] NEC California Eastern Laboratories: NE 68000 NPN Silicon High Frequency Transistors, Santa Clara, CA. [45] H. Nishimoto, T. Okiyama, N. Kuwata, Y. Arai, A. Miyauchi, and T. Touge, "New Method of Analyzing Eye Patterns and its Application to H igh-Speed Optical Transm ission Systems," J. Lightw ave Technol., 6, pp 678-685 (1988). [46] Hewlett-Packard Inc.: HP 70842B Error Detector (0.1-3 G b/s), Santa Rosa, CA. 83 [47] Mini-Systems, Inc., Thin-film Division: 99.6% Alumina Ceramic Substrates, North Attleboro, MA. [48] Autodesk, Inc.: AutoCAD©, Version 11, Sausalito, CA. [49] West Bond Inc.: Model 7200A Epoxy Die Bonder, Anaheim, CA. [50] West Bond Inc.: Model 7400A Ultrasonic Wire Bonder, Anaheim, CA. [51] ROGERS Corp., Microwave Materials Division: R T/Duroid 6002, Chandler, AZ. [52] M /A COM OMNI-Spectra: Flange M ount Jack Receptacle, Waltham, MA. [53] EMI Filter Company: B3 C273A Mini-Feed-Thru, Largo, FL. [54] Electro-LITE Corp.: ELC 4482 UV Light Curing Adhesive, Danbury, CT. 84 CHAPTER IU SPICE Circuit Simulation of Optical Transmitters 4.1 Introduction With the ever increasing interest in high-speed digital and analog fiber optic links, there is more urgency to develop efficient computer-aided design (CAD) tools to sim ulate and design optical links. What a photonic system designer needs today is the ability to deal w ith the system parameters of a fiber optic link in the same way as any electrical system. A modular approach seems to be the logical path to simulating optical links. In this approach, each component of the system (the transmitter, the fiber, and the receiver) is presented as a black box. Then, the w hole link sim ulation is achieved by cascading the individual boxes. There are numerous methods for modeling photonic devices. Some of the methods used for modeling semiconductor lasers are listed in the flow chart of Fig. 4.1 [l]-[5]. 85 M easurem ent Based _ Physics Based M axwell- Bloch 3D, M M , laxwell- Bloch ID, SM , Rate Equations E quivalent ^ C ircuit . M onte-C arlo N on- Linear Multi- Mode Single M ode Linear Simulation of Semiconductor Lasers Fig. 4.1: M ethods for sim ulating the behavior o f sem iconductor lasers. The physics-based methods are not very practical to system designers as computations take significant amount of time even w ith today's state- of-the-art workstations [1], [5]. A more desirable approach is to use the rate equations which describe the interaction between the electron and photon populations inside many optoelectronic devices. The rate equations are a set of coupled first-order differential equations which can be solved numerically using, for instance the Runge-Kutta-Verner fifth-order method. Even though this technique is reasonably accurate and fast, it is inadequate in predicting the dynam ic properties of packaged semiconductor lasers because the rate equations d o not incorporate the electrical parasitics of the laser chip and its package [3], 86 [6]-[10], A development of the rate equations approach is to utilize a SPICE-compatible equivalent circuit of the semiconductor laser. In addition to including the laser chip and package parasitics, this method allows the incorporation of the driver circuitry m odels to simulate the complete transmitter m odule [6]-[13]. In this chapter, we use the last approach to develop SPICE equivalent circuit m odels of low threshold strained InGaAs/GaAs SQW lasers and laser-gates. 4.2 SPICE Circuit M odeling of 2-Terminal InGaAs Lasers W e develop a large-signal SPICE equivalent circuit model of very low threshold strained InGaAs/GaAs SQW lasers. The rate equations are first-order differential equations, so they can be solved using a circuit simulation program by the proper mapping of physical quantities into circuit variables. The derivation is based on the m onom ode rate equations w hich describe the rates o f change of the carrier density, n and the photon density, s inside the active region of a semiconductor laser. Referring to Table 4.1 for definitions of the various variables, the monomode rate equations of a uniformly pumped semiconductor laser are [14] dn I n dt ed T = - L - - 2 - r / i , g. v (4.!a) 87 ^ = rn g S- L +rpBn2 d t x v (4.1b) w here the carrier lifetime rv dependents on the carrier density as follows [14] and the inverse of the photon lifetime is expressed as The output power per facet is [10] In the rate equations given above, we assum ed that the carrier and photon densities are constant across the active region. This assumption is valid for the case of narrow-stripe lasers with strong confinement of injected carriers [10]-[11], which is the case for our InGaAs lasers. Besides, the analysis is restricted to single (or nearly single) frequency lasers. The gain in the active region is given by [15] (4.3) (4.5) 88 Symbol || D efinition (Units) n carrier density (cm'3) s photon density (cm-3) e electron charge (C) W width of active region (pm) d thickness of active region (pm) k laser cavity length (pm) & = WdLc volume of active region (cm3) A, nonradiative recombination coefficient (s'1) B Spontaneous recombination coefficient (cm3 s'1) C Auger recombination coefficient (cm6 s_1) r confinement factor T v Carrier lifetime (s) % Photon lifetime (s) I Current injected into the active layer (mA) P spontaneous em ission factor group velocity in active region (cm /s) 8n differential gain coefficient (cm-1) ",r transparency carrier concentration (cm*1) e optical gain saturation coefficient (cm3) V optical frequency (s'1) h Planck's constant (J.s) «„ internal loss coefficient (cm"1) front and rear mirror reflectivity (/?, = R2 = 32%) n quantum efficiency of the laser Sc photon density normalization factor SN normalized photon density (cm'3) Cs junction capacitance (F) V , heterojunction voltage (V) A laser diode series resistance (Q) K series resistance of diode D (£2) T able 4.1: D efinitions and units o f various sym bols used in the derivation. 89 The term es in the gain function accounts for gain compression which limits the intrinsic bandw idth of QW lasers as discussed in the previous chapter. Gain compression is partially attributed to finite intraband carrier scattering time, finite capture time of carriers from the separate confinement (SC) region into the quantum well, and diffusion o f carriers into the SC region [16]. So, all these nonlinear effects are lumped together in the phenomenological term es. Each term in Eqns. 4.1a,b can be modelled w ith a closed-form function of circuit variables, resulting in [6]-[7] / — /] + + 1 ^ 2 1\ + ' (4.6a) (4.6b) where /, = e\JA„ n (4.7) When the input current is below or close to laser threshold, the non- radiative recombination process dominates, and the l-V characteristics can be m odeled by an ideal diode D with a small resistance R. in series w ith it. The last term in Eqn. (4.6a) represents the space-charge capacitance C s of the laser heterojunction which can be expressed as [7]- [8 ] where V D is the diode built-in potential and C „ is the zero-bias space charge capacitance. The equivalent circuit obtained from Eqns. (4.6)- (4.8) is shown in Fig. 4.2. The series resistor Rs in the circuit accounts for the voltage drop outside the depletion region due to the near ohmic p-P isotype heterojunction and the series lead resistance. To obtain manageable numerical values for the photon density, it is necessary to introduce a normalization factor, st., such that s = s,.sN. The voltage at the output port is proportional to the light output power from the laser. The logarithmic function in the expression of the gain (Eqn. (4.5)) was generated using a temperature-independent SPICE circuit model as A, dt Fig. 4.2: Equivalent circuit model o f a sem iconductor laser. 91 Fig. 4.3: T em perature-independent subcircuit to m odel the g ain function in E qn.(4.5) described in details in Ref. [17]. The subcircuit describing the gain function is shown in Fig. 4.3. The laser circuit simulation was solved using the commercial circuit simulator software HSPICE [18]. The 5.0-1 4 . 5 - 4 . 0 - E,3.5 - 8 3 . 0 - 2.5- 1.0 - m easu red sim ulated 0.5- 0.0 0 2 6 3 4 8 10 5 7 9 C urrent (m A ) Fig. 4.4: Sim ulated and m easured o u tp u t light per facet versus injection current o f an InGaA s/G aA s laser. Inset: Sim ulated pulse response o f the laser. 92 sim ulated dc output light versus current (L-I curve) of an InGaAs/GaAs SQW laser with a cavity length of 175 pm. As shown in Fig. 4.4, a very good agreement w ith the m easured L-I curve is obtained. The simulated pulse response of this laser is show n in the inset of Fig. 4.4. The input current pulse had a 2 ns width, and a low and high levels of 0 and 10 mA, respectively. 4.3 SPICE Circuit M odeling of InGaAs SQW Laser-Gate In the previous chapter w e have demonstrated efficient, ON-OFF large- signal digital m odulation of low threshold strained InG aAs/G aA s SQW laser-gates. These devices are modulated by driving the gate section from a wide bandwidth low impedance voltage source, while keeping the anode at a constant current. In this section, w e derive a large-signal SPICE equivalent circuit model of an InGaAs laser-gate driven from an emitter-follower. The approach to this derivation is very similar to that used for two-terminal lasers and described in Section 4.2. To model the laser-gate, the monomode rate equations of a uniformly pumped laser (Eqns. 4.1) are m odified to account for the intracavity saturable absorber. By referring to Table 4.1 and Table 4.2 for variable definitions, these rate equations are rewritten for the 93 conservation of the total number of carriers and photons as follows [19]-[20] (4.9a) where the carrier lifetime, rs and the gain in the active region, g are as given in Eqns. (4.2) and (4.5), respectively. The distributed mirror loss is given by The experimental dc data of the three-terminal lasers given in Fig. 2.3 shows that the electroabsorption in the gate region has both saturable and nonsaturable com ponents which depend on the applied gate voltage, y,. By adding these tw o components, the phenomenological loss coefficient in the gate region can be expressed as [19]-[20] (4.10) (4.11) 94 The first term in Eqn. (4.11) represents the nonsaturable component of the loss coefficient, expressed as a linear function of . The second term in Eqn. (4.11) approximates the saturable component of the loss in the gate region [19], [21]-[22]. We note here that a ,, a 2, k u and k 2 are treated as fitting parameters. Each term in Eqns. (4.9) can be modeled with a dependent current source, resulting in [6]-[7], [20] /„ = /,+v , 2 + v , 3 +f ir + /"-+c' ^ L An dt dt (4.12a) where, ir + rpi? = hs^C ^ + sN \ — + — K ' dt \ R V RgJ (4.12b) I\=<'AA„n la = r e A W c S bx = B B C r L,iix{aa + am) (4.13) The effect o f the saturable absorber is described by a voltage-dependent resistance Rg given by = < ;(i,/L > ,« ,(v ,) (4.i4) In the laser-gate equivalent circuit illustrated in Fig. 3.5, R and Rg are combined together such that 95 K = RPHRg (4.15) It was necessary to use som e procedures [23], such as scaling, to avoid SPICE convergence problem s w hen solving for the subcircuit representing Eqn. (4.15). The subcircuit representing Re q v is shown in Symbol | Description K gate section length (pm) K anode section length (pm) = WdLa active volume of anode section (cm3) = WdL* active volume of gate section (cm3) h anode current (mA) v* applied gate voltage (V) « i nonsaturable absorption coefficient (cnr1) «2 nonsaturable absorption normalization constant (cm'lV*1) a.,,, maximum saturable absorption coefficient (cm"1) kv k2 voltage normalization constants (V-1) ss n saturation photon density (cm'3) K anode heterojunction voltage (V) T able 4.2: D efinition and units o f variables Fig. 4.5: Equivalent circuit m odel o f InGaAs laser-gate. 96 details in Appendix A. As shown in Fig. 3.19, the laser-gate is driven from an emitter-follower using a single N PN bipolar transistor [24]. The sim ulated dc and large-signal digital responses of the optical transmitter (driver + laser-gate) were obtained using the commercial circuit simulator software HSPICE [18]. The simulated and measured dc output light versus anode current of the optical transmitter for various input voltages to the base of the transistor, vin , are depicted in Fig. 4.6. m easured sim ulated 7 - E = 1.6 V = 2 .8 V = 0 .8 V 0 10 20 5 15 Anode C urrent (mA) Fig. 4.6: Sim ulated (solid) and m easured (dashed) output light per facet versus anode current for input voltages vjn = 0.8, 1.6, and 2.8 V. Using the above circuit model, we have also simulated the eye- diagram of our optical transmitter. For this simulation, w e have used a m odified HSPICE program that allow s the representation of non 97 return to zero (NRZ) pseudo-random bit sequence (PRBS) inputs in the source-file by a SPICE-like statement [25]. For a 1 G b /s, 27-l PRBS NRZ input pattern with a voltage swing of 800 mV and rise and fall times of 100 ps, the resulting simulated eye-diagram of the optical transmitter is as show n in Fig. 4.7. 5. Dns Fig. 4.7: Sim ulated eye-diagram o f the three-term inal laser transm itter for input voltage sw ing o f 800 m V, input signal rise and fall tim es o f 100 ps, and a 27-l NRZ P R B S pattern. 4.4 Future Work: Even though the SPICE equivalent circuit m odels presented above have simulated the responses of the two and three-terminal lasers with reasonable accuracy, more is to be done. Below, w e list som e future 98 work needed to improve the simulation accuracy of the light emitter and to extend the model to simulate the entire optical data link. (i) Extract more accurate values for the physical parameters, such as the gain compression coefficient e , of the strained InGaAs/GaAs SQW lasers from experimental measurements. (ii) Extract laser chip parasitics from measurements of the small- signal scattering parameters (S-parameters). On-wafer probing is preferable for such measurements. (iii) Utilize microwave measurement techniques to model the electrical parasitics of the laser package. (iv) Start w ith the multimode (instead of the monomode) rate equations in the derivation of laser the circuit model [26]. (v) Develop an improved formulation for the rate equations of the three-terminal InGaAs/GaAs lasers in order to obtain a more accurate SPICE circuit model. (vi) Develop a SPICE equivalent circuit model for the optical receiver. (vii) Develop a simple model for the MM graded-index fiber. The model should include the effects of fiber loss, dispersion, and modal noise. (viii) Combine the models of transmitter, fiber, and receiver to simulate the entire optical data link. 99 A "truly complete" semiconductor model is too complicated to be o f practical interest. Rather, it's important to consider w hich laser structures and behaviors the model is supposed to simulate, then make app roxim ations accordingly. The large num ber o f adjustable parameters contained in a laser m odel can easily obscure the results. Therefore, the model should not be more sophisticated than necessary. References [1] A. S. Daryoush, N . Samant, D. Rhodes, and D. Sturzbecher, "Photonic CAD for H igh Speed Fiber-Optic Links," M ic ro w a v e Journal, March, pp 58 (1993). [2] J. Buus, "Principles of Semiconductor Laser Modelling," IE E Proceedings, 132, Pt. J, pp 42-51 (1985). [3] A. J . Lowery, "A two-Port Bilateral M odel for Sem iconductor Lasers," IEEE J. Q uantum Electron., 28, pp 82-92 (1992). [4] A. F. Elrefaie, J. K. Townsend, M. B. Romeiser, and K. S . Shanmugan, "Computer Simulation of D igital Lightwave Links," IEEE J. Selected Areas Comm., 6, pp 94-105 (1988). [5] A. J . Lowery, "New Dynamic Semiconductor Laser M odel Based on the Transmission-Line M odelling Method," IE E Proc., 134, Pt. J, pp 281-289 (1987). [6] R. S. Tucker, "Circuit Model of Double-Heterojunction Laser Below Threshold," IEE Proc., 128, Pt. I, pp 101-106 (1981). 100 [7] R. S. Tucker, "Large-Signal Circuit M odel for Sim ulation of Injection-Laser Modulation Dynamics," /EE Proc., 128, Pt. I, pp ISO- 184 (1981). [8] R. S. Tucker and D. J. Pope, "Microwave C ircuit M odels of Semiconductor Injection Lasers," IEEE Trans. M icrow ave Theory and Tech., MTT-31, pp 289-294 (1983). [9] R. S. Tucker and I. P. Kaminow, "High-Frequency Characteristics of D irectly M odulated InGaAsP Ridge W aveguide and Buried Heterostructure Lasers," /. Lightwave Technol., L T -2, pp 385-393 (1984). [10] R. S. Tucker, "High-Speed Modulation of Semiconductor Lasers," /• Lightwave Technol., LT-3, pp 1180-1192 (1985). [11] D. S. Gao, S. M- Kang, R. P. Bryan, and J• J - Coleman, "Modeling of Q uantum -W ell Lasers fo r C om puter-A ided A nalysis of Optoelectronic Integrated Circuits," IEEE J. Q uantum Electron., 26, pp 1206-1215 (1990). [12] M. Nakamura, N . Suzuki, and T. O zeki, "The Superiority of O ptoelectronic Integration for H igh-Speed Laser D io d e Modulation," IEEE J. Quantum Electron., QE-22, pp 822-826 (1986). [13] H. Elkadi, J . p. Vilcot, S. Maricot, and D. Decoster, "Microwave Circuit Modeling for Semiconductor Lasers Under Large and Small Signal C onditions,"Microwave and Optical Tech. Lett., 3, pp 379-382 (1990). [14] G. P. Agravval and N. K. Dutta, Long-W avelength Sem iconductor Lasers:, New York: Van Nostrand Reinhold (1986). [15] N. C. Frateschi, Ph.D. Thesis, University of Southern California, 1993. [16] K. Y. Lau, "Dynamics of Quantum W ell Lasers," Chapter 5 in Q uantum Well Lasers, P. S. Zory, Editor, San Diego: Academic Press (1993). 101 [17] V. B. Litovski and Z. M. Mrcarica, "Macromodeling w ith SPICE's N onlin ear C ontrolled Sources," IEEE C ircuits & D evices, November, pp 14-15, (1993). [18] Meta-Software Inc.: HSPICE Version H92, Campbell, CA. [19] J. O'Gorman, A. F. J. Levi, R. N. Nottenburg, T. Tanbun-Ek, and R. A. Logan, "Dynamic and Static R esponse of M ultielectrode Lasers,'"Appl. Phys. Lett. , 57, pp 968-970 (1990). [20] H. Elkadi, J. P. Vilcot, and D. Decoster, "An equivalent circuit m odel for multielectrode lasers: Potential Devices for Millimeter- W ave A pplications,"Microwave and Optical Tech. Lett., 6, pp 245- 249 (1993). [21] J. O'Gorman, A. F. J. Levi, T. Tanbun-Ek, and R. A. Logan, "Saturable Absorption in Intracavity Loss M odulated Quantum Well Lasers,1 "Appl. Phys. Lett. , 59, pp 16-18 (1991). [22] A. M. Fox, D. A. B. Miller, G. Livescu, J . E. Cunningham, J. E. Henry, and W. Y. Jan, "Exciton Saturation in Electrically Biased Quantum W ells "Appl. Phys. Lett. , 57, pp 2315-2317 (1990). [23] C. H ym ow itz, "Step-by-Step Procedures H elp You Solve Spice Convergence Problems," E D N , March, pp 121-124 (1994). [24] NEC California Eastern Laboratories: NE 68000 NPN Silicon High Frequency Transistors, Santa Clara, CA. [25] M. Govindarajan, S. Siala, and R. N. Nottenburg, "Optical Receiver Systems for High-Speed Parallel Digital Data Links," Submitted for publication to /. Lightwave Technol. (1994). [26] H. A. Tafti, K. K. Kamath, G. Abraham, F. N. Farookhrooz, and P. R. Vaya, "Circuit M odelling of M ultimode Semiconductor Lasers and Study of Pulse Broadening Effect," Electron. Lett., 29, pp 1443- 1445 (1993). 102 CHAPTER U Optical Ribbon Fibers for Parallel Data Links 5.1 Introduction Communication using cylindrical glass optical fibers has a number of extremely attractive features such as very large bandwidth-distance product, sm all size and w eight, im m unity from electrom agnetic interference, very low crosstalk, and very low transmission loss. Single mode (SM) silica fibers, which are currently w idely used for long-haul optical telecom m unication, have a very high bandw idth-distance product and almost no modal noise. However, due to their small sizes (core diameter 5 - 10 pm), SM fibers suffer from stringent optical alignment tolerances and low coupling efficiency. For short-haul (< 1 km) optical communication at data rates around 1 G b /s/ch a n n el, multi-mode (MM) optical fiber is the preferred transmission medium because of its relaxed alignment tolerance, large coupling efficiency, and relatively lower cost. MM fibers with graded-index (GRIN) profile 103 in the core region have significantly higher bandw idth-distance product, lower loss, and greater rigidity to resist bending than step- index fibers [1]. W e have chosen silica GRIN MM optical fiber with core and cladding diam eters of 62.5 and 125 pm, respectively, as the transm ission m edium in our optical data links. These high-quality Corning fibers have optical attenuation of 3 and 1.5 dB/km at the w avelengths 980 nm and 1.3 pm, respectively, and a numerical aperture N A = 0.275 ± 0.015 [2]. The bandwidth-distance product is specified by the manufacturer to be 200 MHz.km at X , = 850 nm and 600 MHz.km at X = 1.3 pm [2]. Noteworthy is the ever increasing interest in utilizing very low- cost all-plastic MM optical fibers in optical interconnections operating at short wavelengths (650 - 780 nm) [3]-[4]. Plastic optical fibers (POF) have relatively large core diameter (500 - 1000 pm), so they are easier to handle and align. However, they still exhibit large attenuation (~ 125 dB/km at X = 650 nm) and significant intermodal dispersion [3]-[4]. In this chapter, we first discuss noise sources in GRIN MM optical fiber systems and present the effect of this noise on the digital performance of our optical data link. In the last section, we evaluate a 12-channel MM optical ribbon fiber for synchronous parallel transm ission. 104 5.2 N oise in Graded-Index M ulti-M ode Fiber Systems The three basic noise sources in high bit rate MM fiber optic communication [1], [5]-[9]: (i) laser noise which is an intensity fluctuation of the total spectral pattern . This is stimulated by the reflections back into the laser cavity from the laser/fiber and fiber/fiber interfaces. (ii) Partition noise is caused by fluctuations in the modal intensity distribution of the modulated laser source. The main prerequisites of this noise source are multi-longitudinal emission from the laser and wavelength-dependent losses within the link. (iii) Modal noise appears as unwanted amplitude modulation of the received signal and is present when coherent sources are used in MM fiber systems with mode-selective loss. Because of its importance, modal noise is treated in more details in the rest o f this section. When monochromatic light is coupled into a MM fiber, it can propagate only in a finite number of m odes, N , w hich is given approximatelyfor GRIN MM fiber with parabolic index profile by [1], where V is the normalized frequency of the fiber and is given for a fiber with core radius a and numerical aperture NA by V = ~ a ( N A ) (5.2) So, for a parabolic index MM fiber w ith a core diameter of 62.5 pm and a NA of 0.275, the total number of guided modes is N = 758 at A = 980 nm and N = 430 pm at A =1.3 pm. Each of these modes propagates at a slightly different velocity in the fiber resulting in the emergence of a number of w aves from the far end of the fiber having different phase and angle. The interference of these waves causes speckle patterns observed at the end of MM fibers as fluctuations which have characteristic times longer than the resolution time of the detector [1], [7]. The speckle patterns are formed by the interference of modes from a coherent source with a coherence time greater than the intermodal dispersion tim e, 8T w ithin the fiber. So, for a light source w ith uncorrelated source frequency width 8 f, the source coherence time is l / 8 f , and the m odal noise occurs when [1] S f » j p (5.3) Since the average number of speckles is proportional the number of m odes N, fibers with larger core diameters exhibit more speckles. It should be noted here that the speckles are coarser if the fiber is underfilled (i.e.; only a few of the fiber modes are excited), and that the 106 more speckles there are in a fiber cross-section, the smoother the image w ould appear at the fiber end [7]. The presence of a static speckle pattern is, by itself, not a significant problem, however, dynamic changes in this speckle pattern causes amplitude modulation of the transmitted signal as it passes a point of mode selective loss (MSL). Origins of MSL include fiber-to-fiber misalignm ent in connectors and splices, fiber vibrations, laser-to-fiber coupling, and fiber microbending. M odal noise is then generated when the correlation between two or m ore modes w hich give the original (static) interference is differentially delayed by these fiber disturbances [1], [7], [11]. Modal noise may be reduced (or even eliminated) by [1], [7], [11] (i) the use of single-mode fibers (ii) the removal of disturbances along the fiber especially by careful design of fiber connectors. (iii) The use of MM fibers with large numerical aperture, as these fibers support the transmission o f a large number of m odes, resulting in diminishing the contrast of individual speckles. (iv) The use of a broad spectrum source [11]-[13]. A broad spectrum laser source can be obtained by either increasing the w idth of the single longitudinal mode and thus reducing its coherence tim e, or by increasing the number of longitudinal m od es and averaging out the speckle pattern. If modulated w ith short pulses (a few nanoseconds) from just below (or at) threshold, single-longitudinal 107 m ode semiconductor lasers oscillate in multiple longitudinal modes of relatively low coherence. This occurs because short pulses do not provide enough time for the laser to settle into its normal coherent state. We illustrate this phenom enon in Fig. 5.1, w here the optical spectrum o f the sin gle-longitu dinal m ode strained 2-term inal InG aAs/G aAs SQW laser is shown when the laser is m odulated from slightly above threshold at 500 M b /s with 1010... input pattern. In this experiment, light from the laser is transmitted through 6 m of GRIN MM 62.5/125 |im optical fiber and coupled to the FC /PC input connector of an optical spectrum analyzer [14]. SENS I -M B dBm 950.11 955.11 960.11 X (nm) Fig. 5.1: Optical Spectrum o f a 2-terminal InG aA s/G aA s laser m odulated from slightly above threshold at 500 M b/s with ...1010... input pattern, resulting in peak X - 955.1, m ean X - 955.85 nm, FW H M - 2.97 nm , m ode spacing AA - 0 .4 nm (A / - 131.53 G H z), a - 1.26 nm . 108 It is worth to mention here that fiber m ode scramblers can reduce modal noise only in the case the fiber is underfilled at launch. M ode scramblers distribute the power over a greater number of modes, decreasing the mean speckle size and, thus, reducing modal noise [7]. We have investigated qualitatively the effect of fiber disturbance, and hence increased modal noise on the digital performance of our optical data link. The transmitter m odule consists of a strained InG aAs/G aAs SQW laser-gate driven from a low-impedance voltage source, as described in Section 3.5. Light from the three-terminal laser is collected using a lensed 62.5/125 pm GRIN MM fiber. The tapered hem ispheric microlens at the fiber end is fabricated using only chemical etching. The microlens allows typical coupling efficiency of ~ 80% and significantly limits the optical feedback back into the laser-gate as compared with a flat endface fiber. The lensed fiber end is fixed in place using UV-cured epoxy. The other end of the fiber cable is terminated w ith a standard FC/PC connector. These connectors are chosen because of their excellent insertion loss, return loss (~ -35 dB), and repeatability, and their medium cost [15]. The optical spectrum of the modulated three-terminal laser is measured for a constant anode current of la = 13 mA, and an input...101010... bit stream with voltage sw ing at the base of the NPN transistor of 900 mV. The measured optical spectrum at 500 M b/s and 1 G b/s are shown in Fig. 5.2(a) and (b), 109 > Q N PQ 73 u - » S ENS - 3 7 |dE)» - j D / n u i ■ 5 5 . SE d B i 4 HIDE j na 1 flUEF RGE HOD * * J i J f m i L i , f l ( 0 » % f r ' T 935.02 965.02 950.02 X (ru n ) Fig. 5.2(a): O ptical Spectrum o f a 3-term inal InG aA s/G aA s laser m odulated at 500 M b/s w ith ...1010... input pattern for Ia = 13 mA, resulting in peak X = 950.02 nm, m ean X = 952.05 nm, FW HM = 8.96 nm, and o = 3.8 nm. _ > *5 C D 73 in SENS 15 B0 - J / M B * •30.1: d B » t UIOE i n n I RUEf RGE i n n 3 1 .... i l W M n ' l r p * 1* f * ■ r i I f ” *? 935.02 950.02 965.02 X ( n m ) Fig. 5.2(b): O ptical Spectrum o f a 3-terminal InG aA s/G aA s laser m odulated at 1 G b/s w ith ...1010... input pattern for Ia = 13 mA, resulting in peak X = 948.71, m ean X = 951.83 nm , FW HM = 9.77 nm, and a = 4.15 nm. 110 respectively. Asym m etric broadening of the lasing spectrum is observed to increase with data rates, which is similar observed by O'Gorman et al. in intracavity loss m odulated distributed feedback (DFB) lasers [16]. This broadening is attributed to the large carrier density variation and optical overshoot that accompanies the switching process of the gate section [16]-[17]. As it can be seen from Fig. 5.2(a,b), the optical spectrum of the laser-gate m odulated with ...101010... bit stream is relatively broad. A slightly smaller broadening is observed w ith PRBS modulation. To investigate the effect of m odal noise, w e intentionally disturbed the 62.5/125 |im fiber section linking the transmitter to the receiver. Each the of transmitter and receiver m odule includes a 4 m long fiber section w ith 900 |im diam eter plastic jacket and connectorized at the far end with a standard FC/PC connector. Both connectors are then mated together using a low cost commercial FC mating sleeve. The laser-gate is modulated at 1 G b /s w ith a 217 NRZ PRWS pattern for an input voltage sw ing of 1 V, and all the eye- diagrams are accumulated for the same period of time. For no fiber disturbances, the eye-diagram of the data link is as shown in Fig. 5.3(a). Bending the fiber at one location to a radius of 5 mm results in a fairly slight increase of noise in the upper rail of the eye-diagram, as shown in Fig. 5.3(b). Bending the fiber to a radius of approximately 1.25 mm 111 lOOmV/Div 100 mV/Div lO O m V /D iv 500 p s /D iv (a) 500 p s /D iv (b) 500 p s /D iv (c) 500 p s /D iv (d) Fig. 5.3: E ye-diagram s obtained at 1 Gb/s w ith 2 17 NRZ PRW S for in p u t voltage sw ing o f 1 V, w hen the fiber is (a) not disturbance, (b) bent to a radius o f 5 mm, (c ) bent to a radius o f 1 m m , (d) shaken at random. 112 causes a ~ 100 mV decrease in the eye vertical opening, but without a significant increase in noise, as can be seen from Fig. 5.3(c). The eye- diagram in Fig. 5.3(d) is accumulated while the fiber is being shaken at random, resulting in just a small increase in noise. For a quantitative investigation of the modal noise in MM optical data links, it is necessary to measure BER under various fiber disturbances and connector im perfections. Our prelim inary and qualitative m easurements using commercial and inexpensive fiber connectors indicate that degradation in our data link due to modal noise is limited to acceptable levels. The observed attenuation in the received signal when the fiber is bent to a small radius is caused by radiation losses in the higher-order fiber m odes (i.e.; m odes close to cutoff). So, w hen launching light into the fiber, it is preferable to excite predominantly the lower-order m odes guided near the fiber axis. 5.3 Low-Skew Optical Ribbon Fiber For Parallel Data Links Optical ribbon fibers is the logical transmission medium for high-speed parallel optical interconnection due to their low crosstalk, compactness, high throughput. Moreover, high-performance ruggedized miniature ribbon fiber connectors are now commercially available. Bidirectional 113 10-channel optical data links w ith an aggregate data rate of 3 G b /s are presently available at a unit price of $1000 [18]. These relatively modest prices are possible partially because low cost fiber ribbon and connector technologies exist, and also because of relaxed alignment tolerance due to the use of 62.5 or 50 pm diameter core MM fiber in the transmitter and receiver array modules. The ribbon fiber described in this section [19] consists of 12 glass fiber channels spaced 250 pm apart. Each channel consists of a Graded- index MM fiber with core and cladding diameters of 62.5 and 125 pm, respectively. The numerical aperture o f these high-quality Corning fibers is 0.275. The measured coupling efficiency to any of the channels of an as-cleaved ribbon is shown in Fig. 5.4 as a function of the laser-to- fiber separation. In this measurement, a 2-terminal InGaAs/GaAs laser diode (A . = 980 nm), a calibrated broad-area silicon photodetector, and a 2-pm resolution micropositioner are utilized. The position of the fiber in the transverse and lateral directions is fixed for maximum coupled light into the fiber. As can be seen from Fig. 5.4, the maximum coupling efficiency of the flat endface fiber is ~ 21%. The optical crosstalk between neighboring channels of the ribbon fiber is also measured. For this measurement, a 2-channel FC /PC connectorized lightw ave multimeter [20] is utilized. If the laser CW light power coupled into one channel is P m , and the light detected in the adjacent 114 J 11 . ■ 11 I I I I I . I 11 . . ■ I I I ■ I I I ■ . ■ I I ■ I ■ . I ■ I . . I ■ ■ I ■ I. . . ■ 2 4 ~ 2 2 - 2 0 - sp t 1 8 - S 1 2 - .E 1 ° " 1 - 8 - o 1 0 - 4 - 2 - 0 - ^ I I I I | I I I I | I I I I | - | I I I | T I I I |T ■ I ■ | ■ I I I | I I I I | I I'l I | I I I I | 0 100 2 0 0 3 0 0 4 0 0 5 0 0 Laser-Fiber ribbon separation (gm) Fig. 5.4: Absolute coupling efficiency versus longitudinal laser-to-fiber separation distance of an as-cleaved optical ribbon fiber, with th e lateral and transverse positions o f th e fiber at m axim um coupling efficiency. channel is P o ff , then the interchannel crosstalk is defined as The reason for using a factor o f 20 (and not 10) in Eqn. (5.1) is because optical power is converted into electrical current at the receiver side. Fig. 5.5 shows the relative coupling efficiency and the interchannel crosstalk as a function of laser-to-fiber reparation distance, w hen the position of the fiber in the lateral and transverse directions is fixed for c ro ssta lk (d B ) = 201og — (5.4) 115 maximum coupling efficiency. A 150 pm longitudinal offset from the maximum coupling position results in a ~ 4 dB drop in the coupled power. In Fig. 5.6, the relative coupling efficiency and the interchannel crosstalk are plotted as a function of lateral offset w hen the separation distance between the laser and fiber in the longitudinal direction is 128 pm. At these settings, lateral offset of up to ~ 60 pm can be tolerated. - 4 0 - A - crosstalk - o - Relative coupling efficiency — 1 —2 -4 2 - 4 4 —3 *g -46 i -48 - - 4 —5 8 -50 —6 — 7 —8 -56 —9 — 10 -58 - 6 0 h - 11 0 100 3 0 0 200 4 0 0 5 0 0 yo 2. B ) f - f < ■ ( D O O c • O 5 ' (a m 3! S' S' 3 o a. G O Laser-fiber ribbon separation (pm) Fig. 5.5: R elative coupling efficiency and interchannel crosstalk o f the ribbon fiber as a function o f the longitudinal laser-to-fiber separation distance. 116 0 - — A - crosstalk - o - relative coupling eff. -10 — 5 3-20 o —10 8 - 3 0 laser-fiber sep. = 128pm -5 0 -60 I — 25 0 50 100 200 150 Lateral O ffset (pm ) Fig. 5.6: R elative coupling efficiency and interchannel crosstalk o f the ribbon fiber as a function o f the lateral offset o f the fiber w ith respect to the laser active region. The 12-wide optical ribbon fiber is connectorized using a m iniature M echanically Transferable (MT) m ultifiber connector, shown schematically in Fig. 5.7 [21]-[22]. The MT ferrule, fabricated using transfer m olding method, has fiber U-grooves spaced 250-pm apart precisely (1 pm accuracy) arranged between tw o holes for the guiding pins. The total alignment error of this connector is estimated to be ~ 2 pm [21]. Average connector loss of 0.327 dB was obtained for 62.5 pm core fiber w ithout matching material, w ith connector loss variations less than 0.05 dB [21]. 117 Ferrule Fiber Array Guide Pins \ Ribbon Fiber Spring Clip Boot Fig. 5.7: Illustration o f the optical ribbon fiber M T connector (A fter [22]). W hat limits the range of potential applications for parallel optical links based on MM ribbon fiber is the variation in signal propagation delay between the fiber channels. Such interchannel skew is caused by variation in the refractive index of the fibers forming the ribbon. W e have experimentally quantified this interchannel skew and thereby set a practical maximum limit to bandwidth-distance product per channel of low -cost parallel MM fiber links for synchronous transmission. Using a Fabry-Perot laser emitting at X = 1.3 pm, a fast photodetector, and a high-speed oscilloscope, the interchannel skew across the 12-channel ribbon fiber is accurately measured (accuracy of 1 part in 105) [23]. A conventional 103 m long 12-fiber ribbon w as first measured, resulting in a total maximum skew of 1005 ps, or ~ 10 p s/m , and a standard deviation o = 3.2 ps/m . Also, the skew in this ribbon fiber is found to be uncorrelated [23]. 118 Then, a different 102-m long ribbon fiber, where the fiber channels forming this ribbon are sequentially cut from the same pull, was measured. The maximum total variation in optical pulse delay measured across this 12-fiber ribbon is just 127 ps, or 1.25 p s/m , with a standard deviation < r = 0.39 ps. To the best our knowledge, this is the low est interchannel in 12-wide ribbon fiber reported so far [24]-[26]. Besides, a correlation in skew exists and decreases with increasing distance separating the fiber sections being compared. For synchronous parallel optical transmission, SM fiber loses its intrinsically higher bandwidth-distance advantage when interchannel skew of the ribbon becomes the limiting factor. For instance, a skew of 2.1 p s /m in an 8-channel SM ribbon fiber, as reported in Ref. [24], w ould limit the synchronous transmission distance to less than 500 m at 1 G b/s. Our low skew ribbon fiber avoids the use of costly optical or electronic skew-correcting components [25] and is at least as good as SM fiber in satisfying the bandwidth-distance/channel requirements of low cost synchronous parallel links. This can be quantitatively determined by m easuring the m axim um allow ed time delay betw een the transmitted data and clock signals and still maintain the BER below a certain value at a constant data rate. In our experiment, tw o channels (one for clock, one for data) of a laser-gate transmitter array module operating at a wavelength X = 980 nm were used [27], along w ith a 12- 119 channel DC-coup led digital optical receiver module [28]. The inset of Fig. 5.8 shows the eye-diagram obtained w ith the low skew 102 m long MM ribbon fiber at 622 M b/s with a 27-l NRZ PRBS input pattern. As shown in Fig. 5.8, a clock-to-data delay up to 1.3 ns can be tolerated while keeping the BER less than 10'11. Thus, the 12-channel 102 m M M ribbon fiber PRBS 2 7-1, 6 22M b/sec 650 ps/Div 04 w C O -10 -800 -400 4 0 0 800 0 Clock-to-Data Delay (ps) Fig. 5.8: E ye-diagram an d BER versus clock-to-data delay o f parallel synchronous digital transm ission over 102 m o f low skew MM ribbon fiber at 622 M b/s w ith 27-l N RZ PRBS. transm ission span over this very low skew ribbon fiber at 622 M b/s/channel with BER < 10-11 can be increased to a maximum of ~ 1 km. The maximum 8-channel transmission span w ith an aggregate 120 data rate of 622 M byte/s should be greater than 1 km. This increase in transm ission distance is very important in data com m unication applications such as computer interconnections in local-area networks, and also for future high-speed hybrid local loops linking switching terminals to curbside distribution boxes for broadband services to the hom e. References [1] J. M. Senior, "O ptical Fiber C om m unications, P rinciples and Practice," 2nd Edition, Chapter 3, Cambridge: Prentice Hall (1992). [2] ARCOMM: Fiber Systems, Product Information, Lenexa, KS. [3] R. S. Bates and S. D. Walker, "Evaluation of All-Plastic Optical Fibre Computer Data Link Dispersion Limits," Electron. Lett., 28, pp 996- 998 (1992). [4] R. S. Bates, "Equalization and M ode Partition Noise in All-Plastic Optical Fiber Data Links," IEEE Photon. Teclmol. Lett., 4, pp 1154- 1157 (1992). [5] J. R. Jones, "Optical Fiber C om m unications System s," Volume 1, Chapter 10, Raleigh: North Carolina State University (1986). [6] R. E. Epworth, "Phenomenon of Modal N oise in Fiber Systems," T echnical D ig est o f Topical M ee tin g s on O p tica l F iber C om m unication, Washington D. C., paper T hD l, pp 106-108 (1979). 121 [7] R. E. Epworth, "Modal Noise: Causes and Cures," Laser Focus, 17, pp 109-115 (1981). [8] A. M. J. Koonen, "Bit-Error-Rate Degradation in a Multimode Fiber Optic Transmission Link Due to Modal Noise," IEEE J. Selected Areas Comm., SAC-4, pp 1515-1522 (1986). [9] M. J. Lum, D. M. Fuller, A. Hadjifotiou, and R. E. Epworth, "M odulation-Induced M odal N oise in D igital System s - The Prediction and M easurem ent of Bit Error Ratio," Proc. 10th European Conference on Optical Communication (ECOC), Stuttgart, pp 240-241 (1984). [10] K. O. Hill, Y. Tremblay, and B. S. Kawasaki, "Modal N oise in Multimode Fiber Links: Theory and Experiment," Optics Lett., 5, pp 270-272 (1980). [11] R. Dandliker, A. Bertholds, and F. Maystre, "How Modal N oise in M ultim ode Fibers D epends on Source Spectrum and Fiber Dispersion," IEEE f. Ligtlnuave Technol., LT-3, pp 7-12 (1985). [12] K. H. Hahn, M. R. Tan, Y. M. Houng, and S. Y. Wang, "Large Area M ultitransverse-M ode VCSELs for Modal N o ise Reduction in Multimode Fibre Systems," Electron. Lett., 29, pp 1482-1483 (1993). [13] L. Goldberg, and D. Mehuys, "High Power Superluminescent Diode Source," Electron. Lett., 30, pp 1682-1684 (1994). [14] Hewlett-Packard Inc.: HP 70951A Optical Spectrum Analyzer 600- 1700 nm, Santa Rosa, CA. [15] D. Horwitz, "Selection Guide to Fiber-Optic Connectors," IEEE Circuits and Devices Magazine, 9, pp 45-46 (1993). [16] J. O'Gorman, A. F. J. Levi, T. Tanbun-Ek, and R. A. Logan, "Asymmetric Line Broadening in Intracavity Loss M odulated Quantum Well Distributed Feedback Lasers," Appl. Phys. Lett., 58, pp 669-671 (1990). 122 [17] K. Berthold, A . F. J. Levi, T. Tanbun-Ek, and R. A. Logan, "Wavelength Sw itching in InGaAs/InP Quantum W ell Lasers," Appl. Phys. Lett., 56, pp 122-124 (1990). [18] D. Bursky, "Parallel Optical Links M ove Data at 3 Gbits/s," Electronics Design, November, pp 79-82 (1994). [19] US CONEC Ltd.: 62.5/125 pm 12-wide Optical Ribbon Fiber, Hickory, NC. [20] Hewlett-Packard Inc.: HP 8153A Lightwave Multimeter, Santa Rosa, CA. [21] T. Satake and W. P. Blubaugh "New Applications for Miniature M ultifiber Connector," Tech. Digest of LEO S 6th A n n u a l M eeting, San Jose, CA, paper FPW5.4, pp 179-181 (1993). [22] T. Satake, T. Arikawa, W. P. Blubaugh, C. Parson, and T. K. Uchida, "MT Multifiber Connectors and New Applications," Proc. 4 4 th Electronic C om ponents & Technology C onference, W ashington D.C., pp 994-999 (1994). [23] S. Siala, A. P. Kanjamala, R. N. Nottenburg, and A. F. J. Levi, "Low Skew M u ltim ode Ribbon Fibres for P arallel O ptical Communications," Electron. Lett., 30, pp 1784-1786 (1994). [24] A. Takai, T. Kato, S. Yamashita, S. Hanatani, Y. Motegi, K. Ito, H. Abe, and H. Kodera, "200-M b/s/ch 100-m Optical Subsystem Interconnections U sin g 8-channel 1.3pm Laser Diode Arrays and Single-M ode Fiber Arrays,” IEEE f. Ligtlnvave Technol., 12, pp 260- 269 (1994). [25] T. Horimatsu, N . Fujimoto, K. Wakao, and M. Yano, "Optical Parallel Interconnection Based on Group M ultiplexing and Coding Technique," IE1CE Trans. Electron., E77-C, pp 35-41 (1994). [26] K. Kaede, T. Uji, T. Nagahori, T. Suzaki, T. Torikai, J. Hayashi, I. Watanabe, M. Itoh, H. Honm ou, and M. Shikada, "12-Channel Parallel Optical-Fiber Transmission Using a Low-Drive Current 1.3- pm LED Array and a p-i-n PD Array," }. Lightwave Technol., 8, pp 883-887 (1990). 123 [27] S. Siala, H. Zhao, M. Govindarajan, R. N. Nottenburg, and P. D. Dapkus, "Synchronous DC-Coupled Parallel Optical Data Path Using Three-Terminal InG aAs/G aAs Lasers," Electron. Lett., 30, pp 1165- 1166 (1994). [28] M. Govindarajan, S. Siala, and R. N. Nottenburg, "12 x 622 M b /s DC-Coupled Synchronous Optical Receiver Array for Parallel Digital Datalinks," Electron. Lett., 30, pp 1177-1178 (1994). 124 CHAPTER Ul M ulti-Channel Transmitter Array M odules 6.1 Introduction Having successfully designed, im plem ented, and evaluated single channel optical transmitters using tw o and three-terminal InGaAs lasers, w e focus our attention now on multi-channel transmitter array modules for synchronous DC-coupled parallel digital optical data links. Parallel transmission requires a very low interchannel crosstalk (<-30 dB) at the transmitter and receiver arrays. To measure the interchannel crosstalk in the transmitter array, it is necessary to use individual receivers (single-channels). Several D C -coupled single-channel receivers were packaged in our laboratory using the same brass housing and microwave mother board as those used for the single-channel transm itters (see Section 3.6). Besides interchannel crosstalk, interchannel skew must also be m inim ized in parallel synchronous transmission. Most often, interchannel skew lim its the bandwidth- distance product of synchronous transm ission. M oreover, the 125 uniformity of the lasers in the array is of paramount importance for parallel transmitters. For instance, a large variation in the lasing threshold results in a large variation in the turn-on delay of the various lasers in the array. This, in turn, translates into significant interchannel skew. In this chapter, w e describe the fabrication and presents the results of m ulti-channel optical transmitter m odules using three different types of light emitters: 2-terminal InGaAs laser array, InGaAs laser-gate array, and commercial 2-terminal InP laser array. 6.2 Four-Channel Transmitter M odule Using Two-Terminal InGaAs/GaAs Laser Arrays We have discussed in details the large-signal digital m odulation of 2- terminal strained InGaAs/GaAs SQW lasers in Section 3.4. These lasers have a common cathode and have been driven from a common-base amplifier as depicted in Fig. 3.14. The same drivers are used in a four- channel transmitter module. The laser bar and the drivers circuitry components were epoxied [1] onto a 0.5 cm X 1 cm patterned 254-|im thick ceramic substrate [2]. Fig. 6.1 show a photograph of the completed ceramic board. Discrete commercial PNP transistors w ith f T ~ 4 GHz 126 were utilized [3]. The ceramic substrate was then attached to a Duroid motherboard [4] patterned with four 50 Q microstrip transmission lines and one supply line (Vcc)- To minimize the interchannel skew, the high-speed transmission lines on the motherboard were designed in ACAD [5] to have the same electrical length. Laser Bar PN P Transistors Fig. 6 .1: Photograph o f the com pleted ceram ic board for the four-channel transm itter with tw o-term inal InG aA s laser array. The lasers in the array are spaced 250 pm apart and had lasing thresholds of 2.4 ± 0.2 mA. In this transmitter array, the spacing betw een the channels is 500 pm, so every other laser is wire-bonded to the collector of a transistor. Using a lensed MM fiber m ounted on a XYZ m icropositioner, w e obtained the eye-diagram s of the four channels (one at a time) as depicted in Fig. 6.2. The input signal consisted of a 217 NRZ PRWS pattern at 1 G b /s with low and high 127 300 ps/ div Fig. 6.2: Eye-diagram s o f th e four channel transm itter w ith tw o-term inal InGaAs laser array. The eye-diagram s were m easured one at a tim e with an input signal at lG b/s 2 1 7 NRZ PRW S and low and high voltage levels o f 0.9 V and 1.4 V, respectively. 128 voltage levels of 0.9 V and 1.4 V, respectively, and was DC-coupled to the transmitter input [6]. The single supply line was kept at Vcc = -10 V. It can be observed from the bottom rails of the eye-diagrams in Fig. 6.2 that the lasers in the array were lasing in the "ZERO" state. The rise and fall times were measured to be around 320 ps. We note that this transmitter module was not fiber-pigtailed because ribbon fiber was not available in our laboratory at that time; and because our package design w as not adequate for pigtailing one channel at a time using single fibers. Thus, no bit-error-rate measurements were attempted on this transmitter m odule because a small m ovem ent in the fiber-holding micropositioner could result in a large number of errors to occur. 6.3 Four-Channel Transmitter M odule U sing InGaAs/GaAs Laser-Gate Arrays W e have im plem ented a synchronous DC-coupled parallel optical transmitter m odule using an array of strained InG aAs/G aAs SQW laser-gates. The gate section of each of the three-terminal laser in the array is connected to the emitter lead of an N PN common-collector amplifier. Commercial silicon bipolar transistors with f T = 10 GHz were utilized [7], This four-channel transmitter has a single anode current /„ and collector bias supply V /, . The laser-gate array and the 129 electronic components are epoxied [1] onto a 254-pm thick patterned ceramic substrate [2]. A photograph of the transmitter's completed ceramic substrate is shown in Fig. 6.3. The ceramic substrate is then attached to a microwave motherboard [4] patterned w ith four 50-fli m icrostrip transm ission lines and tw o dc su p p ly lines. The motherboard is, in turn, mounted onto a brass submount to which SMA connectors [8] and dc feed-through pins [9] for the high-speed and dc signals, respectively, are fixed. Light from the laser-gate array is lensed MM fibers laser-gate array imon anode 200 £ 2 N PN transistors 50 £2 m icrostrip lines F ig. 6.3: Photograph o f the com pleted ceram ic substrate o f the four-channel laser- gate transm itter m odule. 130 collected using four MM graded-index fibers with core diameter of 62.5 gm. Each of the fiber cables has a FC/PC connector at one end, with the other end terminated w ith a hemispheric microlens. Using an array of V-grooves machined 500 gm apart in the brass submount, the lensed ends of the fibers are actively aligned (one at a time) to the laser-gates and then fixed in place using ultraviolet-cured epoxy [10]. Thus, this transmitter module is electrically and optically connectorized. Utilizing the same receivers used in previous sections [11], the eye-diagram of each of the four channels is obtained at 1 G b /s with a NRZ PRWS input data of length 217-1 for an input voltage swing of 0.8 V (ECL swing) and anode current o f 50 mA. The eye-diagrams shown in Fig. 6.4 are obtained after transmission through 200 m of MM fiber. Very good uniformity and wide-open eye-diagrams with large contrast ratio are achieved. The bit-error-rate (BER) is measured as a function of the input voltage swing at 1 G b/s NRZ 215-1 PRBS for synchronous transmission (clock is transmitted as one of the channels) through 16 m of 62.5/125 gm MM fiber. The results are depicted in Fig. 6.5. By linear extrapolation of the fitted data, a BER of 10‘14 is achievable with an input voltage swing of just 370 mV. However, a BER floor of 10*13 lim ited the measurement. Again, this BER floor is caused by bursts of 131 200 mV/div __ ] _________ 5 0 0 ps/div Fig. 6.4: Eye-diagram s o f the four-channel laser-gate transm itter at 1 Gb/s w ith N RZ 2 17-1 PRW S for input voltage sw ing o f 8(X) m V and fiber length o f 2 0 0 m. 132 errors that occur after several hundreds seconds of error-free transm ission. , 0 10 4 0 6 0 8 0 -10 0 -12 0 •14 0 -16 0 260 280 300 320 340 360 380 Input V oltage Swing (mV) Fig. 6.5: BER versus input signal sw ing m easured w ith 1 Gb/s N RZ 2*^-1 PRBS after synchronous transm ission through 16 m o f M M fiber. The phase margin of the synchronous link is an important system parameter because it quantifies both the timing jitter and the fidelity of the transmitted NRZ digital data. In Fig. 6.6, the phase margin is plotted as a function of the data rate with a 27- l NRZ PRBS for a BER of 10'11 and input voltage swing of 800 mV. Phase margins of 311° and 220° are measured at data rates of 1 and 2 G b/s, respectively. 133 These results represent an im provem ent over those reported previously for synchronous optical transmission [12], [13]. 360 340 ~ 3 2 0 O) ■g 300 T 280 5> „ „ Jo 260 2 240 0) to 220 JZ 200 180 160 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 Data R ate (Gb/s) Fig. 6.6: Phase m argin versus data rate for 8(X ) mV input signal sw ing for synchronous transm ission with 27-l NRZ PR B S. Crosstalk between the channels of the optical transmitter array is also investigated. The optical crosstalk is m easured to be very low (below -60 dB). The total interchannel crosstalk is measured as follows: w ith one channel being digitally m odulated, the signal from a neighboring channel (which is kept biased) is displayed on a fast digitizing oscilloscope [14], and its root-mean-square (RMS) value is 134 - 10 H - 2 0 - - 3 0 - u - 5 0 - -60-1 0.0 0.5 2.5 1.0 1.5 2.0 3.0 Data Rate (Gb/s) Fig. 6.7: O verall interchannel crosstalk versus data rate (filled circles) o f the four- channel laser-gate transm itter array. L ow er curve (diam onds) represents the m easured feed-through from gate to anode o f a single laser-gate. measured. The measured overall interchannel crosstalk versus data rate is the upper curve in Fig. 6.7. The overall crosstalk is -28 dB at 1 G b/s and degrades to about -15 dB at 3 G b/s. To investigate the origin of this interchannel crosstalk in the transmitter array, we measured the electrical feed-through between the anode and gate of a single three- terminal laser. The results of this measurement is presented by the lower curve in Fig. 6.7. The anode-to-gate feed-through is lower than -35 dB up to 3 G b /s with no bypass capacitors for the anode supply. Providing signal bypass to the anode supply line with one 82 pF 135 parallel-plate chip capacitor [15] reduces the anode-to-gate feed-through to about -45 dB at 3 G b/s, as shown in Fig. 6.8. Further improvement in the isolation between anode and gate is achieved w ith a 1 nF capacitor in parallel with the 82 pF capacitor, as shown in Fig. 6.8. So, the relatively high interchannel crosstalk is not due to electrical feed- through from the gate to the anode of one laser-gate and, therefore, to neighboring laser-gates. The interchannel crosstalk is electrical in origin and arises from the power supply lines of the digital circuitry. Several plated-through via holes in the ceramic substrate w ould considerably enhance the ground connection, resulting in reduced electrical crosstalk in the transmitter array module. - 20-1 - O - No b y p ass - O - with 8 2 pF - A - 82 pF / / 1 nF - 3 0 - CQ ■ o - 4 0 - u T o < / > v> o L _ -5 0 - ( _ > - 6 0 - -70-1 0.5 0.0 1.0 1.5 2.5 3.0 2.0 Data Rate (Gb/s) Fig. 6.8: Electrical feed-through betw een the gate and anode o f a laser-gate in the case of no by p ass capacitance (diam onds), 82 pF capacitor (circles), an d 82 p F capacitor in parallel w ith 1 nF capacitor (triangles). 136 6.4 Ten-Channel Transmitter M odule U sing Two-Terminal InP Laser Arrays In addition to our work with InGaAs/GaAs laser array technology, w e have recently implemented a ten-channel parallel optical transmitter using commercial strained InGaAsP/InP laser array [16]. In this section, w e describe the design and fabrication of the multi-channel transmitter module and present the results of our preliminary measurements. The laser diode array consists of ten Facet-Selective growth Buried-Heterostructure (FSBH) Fabry-Perot (FP) lasers fabricated by MOCVD process [17], [18]. These strained m ultiple quantum w ell (MQW) lasers operate in a stable, single transverse m ode at a w avelength of 1310 nm, providing standard continuous (CW) light output of 6 mW [16]. At a 5 mW CW operation, the laser output beam divergence angles are typically 25° and 30° in the planes parallel and perpendicular, respectively, to the laser junction plane. The length of the laser cavity is 200 pm, with a separation of 250 pm between the active regions of adjacent lasers. Both facets of each laser are high- reflection coated resulting in reflectivities of 60% and 90% for the front and rear facet, respectively [17]. These 1.3 pm lasers have a common anode (i.e; the top contact is n-type), which is advantageous over the com m on-cathode InGaAs lasers described earlier. Com m on-anode 137 laser arrays can be driven from simple driver circuitry utilizing high speed N PN transistors; w hile common-cathode laser arrays require PNP transistors. The uniform ity of these 1.3 pm laser arrays is excellent, with threshold current of (2.4 ± 0.1) mA at 25 °C across the 10- element array. However, InGaAsP/InP lasers have lower temperature stability than InG aA s/G aA s lasers. This is explained by the characteristic temperature, T a, which is approxim ately 150 °K for InGaAs/GaAs lasers [19] and around 50 °K for InGaAsP/InP lasers. So, typically, the threshold current of these InP lasers increases from 2.4 mA at 25 °C to 7.5 mA at 80 °C, w hile the typical slope efficiency decreases from ~ 0.31 W /A at 25 °C to 0.21 at 80 °C. A very important advantage that InP lasers currently have over InGaAs lasers is higher reliability and longer lifetime, with an estimated mean-time-to-failure (MTTF) of about 10 years [20], [21]. The InP laser array w as supplied to us with the laser bar soldered onto a SiC submount. The common anode and the cathode of each of the lasers are w ire-bonded to bonding pads on the subm ount. Furthermore, the SiC subm ount is m ounted next to a patterned ceramic substrate on a 3.4 mm X 3 mm X 1.5 mm gold-plated metallic block. We used silver epoxy [1] to attach this laser mount in a small recess machined in the brass housing of our transmitter array module, as shown in Fig. 6.9 and Fig. 6.10. Components of the laser driver array 138 Laser subm ount Brass housing substrate NAN transistors 47 Q ch ip resistors 50 f t m icrostrip lines Fig. 6.9: Photograph o f th e com pleted ceram ic board o f the 10-channel transm itter module. The laser driver array circuitry is im plem ented using discrete com ponents. circuitry are assembled on 1.5 cm X 1 cm patterned ceramic substrate [22], as depicted in Fig. 6.9. Each laser driver consists of a common-base am plifier using a sin gle NPN silicon bipolar transistor [7], The transmitter array uses a single supply line o f Vcc = 5 V. The ten 50-£2 microstrip transmission lines etched on the 9.8 cm X 7.1 cm DUROID motherboard [4] were designed in ACAD [5] to have the same electrical length in order to m inim ize the interchannel skew in the transmitter. H igh-speed electrical connection is achieved with the use of SM A connectors [8], as shown in Fig. 6.10. 139 DUROID m otherboard 50il m icrostip lines SMA connector Com m on G round pin G old-plated ss housing DC pin Laser array subm ount Ceram ic substrate Ribbon fiber h o ld e r' Fig. 6.10: Photograph o f the ten-channel 1.3 pm transm itter array module. The preliminary digital measurements of this transmitter m odule is performed using a graded-index MM lensed optical fiber. As depicted in Fig. 6.10, the transmitter module w as designed to accommodate a 12- channel MM optical ribbon fiber to collect light from the laser array. Pigtailing of this transmitter will be carried out in the near future. As in digital measurements described earlier for the 980 nm transmitters, single-channel receivers designed and fabricated in our laboratory are used to measure the 1.3 pm transmitter array. We note here that the InGaAs p-i-n photodiodes used in our receivers have a responsivity of about 0.9 A /W at 1.3 pm as compared with 0.65 A /W at 980 nm [23]. U sing a 20 GHz digitizing oscilloscope [14], and by dc-coupling the output of a data pattern generator [6] to the input of one transmitter 140 channel, w e investigated the digital performance at different laser biases. The eye-diagram in Fig. 6.11 is obtained for a bias-free modulation at 500 M b/s for a 217 NRZ PRWS input pattern and high- level current through the laser of 19 mA. At these conditions, the standard deviation of the cross-over timing jitter is measured to be 50 ps. Bias-free modulation of this laser is probably acceptable (at least as far as phase margin is concerned) up to a data rate of 500 M b/s. Modulation at higher data rates may require nonzero bias levels. For a 1 ns/ div Fig. 6.11: B ias-free m odulation at 5(X) M b/s for input pattern o f 217 PR W S and high- level current through the laser o f 19 mA. 2 current bias level Ih =Iin w = — where I,h = 2.5 m A and for the same high-level current through the laser Illiy h = 19 mA, w e obtained the eye- diagram shown in Fig. 6.12 at 1 G b /s with a 217 NRZ PRWS input pattern. At these conditions, the standard deviation of the timing jitter dropped to 31 ps. Increasing the current bias to equal the laser threshold 141 ieebiep||iiri I j | i 500 ps/ div 2 F ig. 6.12: Eye-diagram obtained at 1 Gb/s, w ith 2 17 N RZ PRW S for I h = IU m = — I,h 3 and = 1 9 m A (/„, = 2 .5 m A ). 500 ps/ div 70 ps/ div Fig. 6.13: Eye-diagram (top) obtained at 1 G b/s, with 2 17 N RZ PRW S for Iln w = Ilh= 2.5 m A and lh ig h = 1 9 m A , and corresponding tim ing jitter o f a = 26 ps. 142 current ( lh = /,,,), while keeping everything else the same, results in the eye-diagram o f Fig. 6.13. In this case, the tim ing jitter standard deviation is a =26 ps. The effect of the laser bias on the duty-cycle of the 1.3 pm transmitter is described by Fig. 6.14, where the duty-cycle is measured as a function of the ratio of laser bias to laser threshold ) at 500 M b /s and 1 G b /s. The duty-cycle measurement is achieved using a ...1010... pattern displayed on a 20 GHz oscilloscope [14]. The lower duty- cycles at 1 G b /s for lhw > Ilh as compared to those at 500 M b /s is mostly the result of the "rounding" in the pulse shape at higher data rates. 5 2 - 5 0 - 4 8 - 4 6 - o 4 4 - 4 0 - 3 8 - 500 M b/s 1 G b /s 3 6 - 3 4 - I th = 2.5 mA 3 2 H 0.0 0.2 0.8 1 .4 0 .4 0.6 1.0 1.2 I t o w / l | h Fig. 6 .14: T ransm itter duty-cycle versus the ratio o f laser bias to laser threshold ( W , ) at 500 M b/s and 1 Gb/s. / lh 143 As mentioned earlier, the uniformity of lasers in the array is excellent. This uniformity in the dc characteristics translates to a uniformity in the digital modulation performance, as can be seen from the eye-diagrams shown in Fig. 6.15 of some of the channels of the transmitter array. In this measurement, an ECL input voltage signal from -1.7 V to -0.9 V is dc-coupled to the input of the laser driver at a data rate of 1 G b /s for a 217 NRZ PRWS data pattern. A t these conditions, the laser is biased slightly above threshold ( ~ 1.05 Ilh) and the current high-level through the laser is approximately 17 mA. The rise and fall times are approxim ately 195 ps and 175 ps, respectively. 6.5 Conclusions We can now draw some conclusions and make a few comparisons of the InGaAs versus the InP lasers and the two-terminal versus three- terminal devices as edge-em itting light sources for optical digital parallel data links. A major advantage that strained InGaAs/GaAs laser technology has over InP lasers is temperature stability, w hich considerably simplifies the system design of the transmitter. InGaAs/GaAs lasers do 144 350 ps/ div Fig. 6.15: E ye-diagram s o f some o f the channels o f our 1.3 p m laser transm itter array m odule at 1 Gb/s/channel for an ECL input voltage from -1.7V to -0.9 V and a 2 17 N R Z PRW S pattern. 145 not require (at least for temperatures up to 50 °C [19]) the use of Thermal Electric Coolers (TEC), nor do they necessitate the use of elaborate, intelligent laser drivers that account for changes in operating temperature. The use of InP lasers taxes the system design w ith some sort o f temperature-effect correcting mechanism. This can consist of (i) attaching a TEC to the laser array m ount, (ii) incorporating a heat- sensing element w ith feedback circuitry to re-adjust the modulating current levels out of the laser driver, (iii) or try to push the problem to the receiver side (not advisable!) and "pay the price" in the form of complex decision circuitry. Besides temperature stability, InGaAs lasers are m ore efficient that InP lasers due to their higher gain. The m agnitude of this advantage is, however, reduced by the fact that commercial high-speed InGaAs detectors have low er responsivity at 980 nm than at 1.3 pm [23]. On the other hand, InP laser technology is presently more reliable than that of strained InGaAs/GaAs. But, as discussed in Section 2.4, the reliability gap between these two laser technologies is shrinking as th e InGaAs tech n ology keeps im proving. A lso, further im provem ent in the uniform ity of InG aA s/G aA s laser arrays is necessary. The maturity of the InP laser technology was motivated mostly by the fact that minimum loss in silica optical fiber occurs near 1.3 pm. 146 However, for short-haul (< 1 km) parallel optical transmission, loss in optical ribbon fiber is not important, as the m ain issues (for ~ 1 G b /s/ch ann el) are cost and interchannel skew [24]. So, for data link applications using silica optical fiber, the wavelength is not dictated by minimum loss or dispersion in the fiber. The experimental results we have obtained and presented in this thesis prove that three-terminal lasers have many advantages over two-terminal lasers. Laser-gates can be driven from a voltage source, such as from an emitter-follower output stage of a register flip-flop, thus eliminating the need for a separate driver circuitry. Besides, laser- gate is compatible with the presently popular emitter-coupled logic (ECL) family, requiring no extra power supply nor level-shifting circuitry. In addition, laser-gates achieve not only efficient digital modulation, but also modulation depths close to 100% because lasing ceases in the "OFF" state. A particularly important feature of laser-gate is the effect of temperature on its modulation response. It has been experimentally observed that the characteristic temperature, Tn, of InGaAs/InP laser- gates is m ainly related to the anode section, and is practically insensitive to variations in the gate voltage [25]. So, unlike two- terminal lasers, an increase in the operating temperature of the laser- gate results in higher modulation efficiency [25]. 147 Despite these advantages, more work is needed to improve the uniformity of laser-gate arrays and to evaluate their reliability. References [1] Epoxy Technology, Inc.: H20E Electrically C onductive Epoxy, Billerica, MA. [2] M ini-Systems, Inc., Thin-film Division: 99.6% Alumina Ceramic Substrates, North Attleboro, MA. [3] NEC California Eastern Laboratories: NE 88900 PNP Silicon High Frequency Transistors, Santa Clara, CA. [4] ROGERS Corp., Microwave Materials Division: R T/Duroid 6002, Chandler, AZ. [5] Autodesk, Inc.: AutoCAD©, Version 11, Sausalito, CA. [6] Hewlett-Packard Inc., HP 80000 Data Generator System, Santa Rosa, CA. [7] NEC California Eastern Laboratories: NE 68000 NPN Silicon High Frequency Transistors, Santa Clara, CA. [8] M /A COM OMNI-Spectra: Flange Mount Jack Receptacle, Waltham, MA. [9] EMI Filter Company: B3 C273A Mini-Feed-Thru, Largo, FL. [10] Electro-LITE Corp.: ELC 4482 UV Light Curing Adhesive, Danbury, CT. 148 [11] M. Govindarajan, S. Siala, and R. N. Nottenburg, "DC to 2.5 G b /s x 4 p-i-n/H B T Optical Receiver Array w ith Low Crosstalk," IEEE Photon. TechnoI Lett., 5, pp 1397-1400 (1993). [12] H. Nobuhara, K. Nakajima, K. Tanaka, T. Odagawa, T. Fuji, and K. Wakao, "Zero-Bias M odulation of Tensile-Strained In G aA s/ InGaAsP Quantum W ell Lasers w ith W ide Phase Margins," Electron. Lett., 29, pp 138-139 (1993). [13] T. Horimatsu, K. Fujimoto, K. Wakao, and M. Yano, "Optical Parallel Interconnection Based on Group Multiplexing and Coding Technique," IEICE Trans. Electron.,, E77-C, pp 35-41 (1994). [14] Hewlett-Packard Inc., HP 54120B 20 GHz Sampling Oscilloscope, Santa Rosa, CA. [15] Dielectric Labs., Inc.: dicap© M icrowave Ceramic Capacitors Cazenovia, NY. [16] M itsubishi Electric Corp.: ML78512 Laser D iodes for Optical Communication Systems, Itami City, Japan. [17] K. Matsumoto, Y. Miyazaki, E. Ishimura, H. Nishiguchi, T. Shiba, K. Goto, A. Takemoto, E. Omura, M. Aiga, and M. Otsubo, "2.5 G b/s Parallel Transmission Without Excess Bit Error Rate in a 1.3 pm Strained MQW LD Array for O ptical Interconnection," in Conference Digest of I 4 tl1 IEEE Int. Semiconductor C o n f, pp 127-128 (1994). [18] Y. Ohkura, T. Kimura, T. N ishim ura, K. M izuguchi, and T. Murotani "Low Threshold FS-BH Laser on p-InP Substrate Grown by All-MOCVD," Electron. Lett., 28, pp 1844-1845 (1992). [19] H. Zhao, Ph.D. thesis, University of Southern California (1994). [20] M. Fukuda, "Laser and LED Reliability Update," ]. Lightw ave Technol., 6, pp 1488-1495 (1988). [21] C-E. Zah, R. Bhat, B. N. Pathak, F. Favire, W. Lin, M. C. Wang, N. C. Andreadakis, D. M. Hwang, M. A. Koza, T-P Lee, Z. Wang, D. Darby, 149 D. Flanders, and J. J. Hsieh, "High-Performance Uncooled 1.3-pm A l xG a y I n i - x- y A s / I n P Strained-Layer Quantum-W ell Lasers for Subscriber Loop Applications," IEEE J. Q uantum Electron., 30, pp 511-520 (1994). [22] Dielectric Labs., Inc.: 99.6% Metallized Alumina Ceramic Substrates, Cazenovia, NY. [23] Ferm ionics O pto-T echnology: H igh Speed InGaAs PIN Photodiodes, Simi Valley, CA. [24] R. A. Nordin, A. F. J. Levi, R. N. Nottenburg, J. O'Gorman, T. Tanbun-Ek, and R. A. Logan, "A Systems Perspective on Digital Interconnection Technology," /. Lightwave TechnoL, 10, pp 811-827 (1992). [25] J. O'Gorman, A. F. J. Levi, T. Tanbun-Ek, and R. A. Logan, "Saturable Absorption in Intracavity Loss M odulated Quantum W ell Lasers," Appl. Phys. Lett., 59, pp 16-18 (1991). 150 CHAPTER U N Conclusions Parallel optical interconnection using ribbon fiber is a very promising solu tion to data transfer bottleneck in advanced sw itching and computer systems. The very recent launching of some parallel optical data link commercial products [l]-[2] is a testimony that this technology is maturing rapidly. In this thesis, w e have explored some of the issues of transmitter arrays and multi-mode optical fibers for parallel digital data links. We have fabricated and evaluated the performance of 1 G b /s/ch an n el single and multi-channel transmitter m odules using edge-emitting, low-threshold, strained InGaAs/GaAs single quantum well lasers and laser-gates emitting at a w avelength of 980 nm. In addition, we have also implemented and presented preliminary results on multi-channel transmitter module using commercial edge-emitting 1.3 pm strained InGaAsP/InP multiple-quantum well laser array [2]. 151 We have successfully demonstrated highly efficient, ON-OFF large-signal dc-coupled digital m odulation of InG aA s/G aA s three- terminal lasers at 1 G b/s/channel. This has been achieved by driving the modulator section (or gate) from a wide-bandwidth low-impedance voltage source, while keeping the gain section at a fixed current. We have discussed the advantages that laser-gates have over two-terminal lasers for digital link applications. Laser-gates can be driven from a voltage source, such as an em itter-follow er output stage of the transmitter's register flip-flop, thus eliminating the need for a separate driver circuitry (transconductance amplifier). The three-terminal laser is compatible with standard emitter-coupled logic ECL gates and may be easily interfaced to other logic families. In addition to efficient digital modulation, modulation depth close to 100% is obtained with laser- gate. A nother im portant advantage of laser-gate is its higher insensitivity to temperature changes. This is due to the dependence of the characteristic temperature of the device mainly on the anode, and not the modulator, section. However, more work is needed to evaluate the reliability and improve the uniformity of laser-gate arrays before they can be exploited in any commercial product. Even though 1.3 pm InP laser array technology is presently more reliable and uniform, it suffers from temperature instability as compared w ith strained InGaAs/GaAs lasers. Two-terminal 1.3 pm 152 laser arrays w ould require the use of either thermal electric coolers or laser driver arrays w ith temperature-effect correcting circuitry. For short-haul optical interconnections, InG aA s/G aA s lasers m ay be attractive because of their larger gain and significantly higher temperature stability. Moreover, w e have presented in this thesis large-signal SPICE equivalent circuit models for two and three-terminal InGaAs/GaAs lasers based on single-mode rate equations. These circuit models allow the inclusion of the electrical parasitics of the laser chip and its package, and also permit the incorporation of the driver circuitry. The laser and laser-gate circuit simulations were solved using the commercial circuit simulator HSPICE. The simulated dc and digital modulation responses agreed reasonably well with experimental results. Improved m odels can be obtained by extracting more accurate values for the physical parameters and electrical parasitics of these lasers. In addition, w e have measured the interchannel skew in 12- channel 62.5/125 pm graded-index multimode optical ribbon fibers for parallel synchronous transmission [3]. A record low skew of 1.25 p s/m is reported for a fiber-optic ribbon where each channel is formed from fiber sequentially cut from the sam e pull [4]. The 12-channel synchronous transmission span over this very low skew ribbon fiber at 622 M b /s/c h a n n el with a bit-error-rate less than 10-11 has been 153 increased to about 1 km. This increase in transmission distance is very important not only for data communication applications, but also in future high-speed local loops linking switching terminals to curbside distribution boxes for broadband digital services to the home. References [1] D. Bursky, "Parallel Optical Links Move Data at 3 Gbits/s," Electronic Design, November, pp 79-82 (1994). [2] M itsubishi Electric Corp.: ML78512 Laser Diodes for Optical Communication Systems, Itami City, Japan. [3] USCONEC Ltd.: 62.5/125 (im 12-Wide Optical Ribbon Fiber, Hickory, NC. [4] S. Siala, A. P. Kanjamala, R. N- Nottenburg, and A. F. J. Levi, "Low Skew M u ltim o d e Ribbon Fibres for P arallel O p tical Communications," Electron. Lett., 30, pp 1784-1786 (1994). 154 CHAPTER U111 Appendix A R 4020 equivalent . y i - ^ ¥ ) v(4020,0) < 10f j ^ 1 3012 K ,u fh 9 i+ ■ T - y ) v v 0 O U O ) ^ 10 l + _ _ f£ _ L _ •V-d-»c2 v,) (scale) ■ 155

Linked assets