JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 13, NO. 6, JUNE 1995 995 Technology Development of a High-Density 32-Channel 16-Gb/s Optical Data Link For Optical Interconnection Applications for the Optoelectronic Technology Consortium (OETC) Yiu-Man Wong, Member, IEEE, Dirk J. Muehlner, C. C. Faudskar, D. Bruce Buchholz, Mikhail Fishteyn, James L. Brandner, Member, IEEE, W. J. Parzygnat, Robert A. Morgan, Member, IEEE, Theodore Mullally, R. E. Leibenguth, Gregory D. Guth, Marlin W. Focht, Member, IEEE, Kenneth G. Glogovsky, John L. Zilko, John V. Gates, Philip J. Anthony, Senior Member, IEEE, Booker H. Tyrone, Jr., Member, IEEE, Timothy J. Ireland, Donald H. Lewis, Jr., Member, IEEE, David F. Smith, Sal F. Nati, David K. Lewis, Member, IEEE, Dennis L. Rogers, Member, IEEE, Herschel A. Aispain, Sudhir M. Gowda, Member, IEEE, Steven G. Walker, Young H. Kwark, Member, IEEE, Richard J. S. Bates, Senior Member, IEEE, Daniel M. Kuchta, Member, IEEE, and John D. Crow, Fellow, IEEE Abstract-A parallel, 32-channe1, high density (140 pm pitch), 500 Mb/s NRZ, point-to-point, optical data link has been fabricated using existing GaAs IC, silicon optical bench (SIOB), and multichip module (MCM-D) technologies. The main components of the transmitter and the receiver modules are a GaAsbased vertical cavity surface emitting laser (VCSEL) array at 850 nm with its IC driver array chip and an integrated metalsemiconductor-metal(MSM) receiver (photodetector and signal processing circuits) array at 850 nm. The package module uses a modified 164 U 0 JEDEC premolded plastic quad flat pack (PQFP) in combination with a polymer film integrated circuit (POLYFIC) chip carrier. The electrical input and output are 500 Mb/s NRZ binary signals. The optical YO in both modules consists of a directly-connectorized(nonpigtail)fiber array block that plugs into the 32 x 1 optical fiber ribbon directly on one side and accepts 32 optical signals from the SEL array or delivers them to the MSM receiver array via a gold-coated 45” polished fiber array mirror. The MACII-32 ribbon cable is an enhanced version of the standard MACIITMconnector ribbon cable. This paper characterizes key components of the optical data link, describes its package design, and discusses preliminary component and optical data link test results. I. INTRODUCTION T HE OPTOELECTRONICS Technology Consortium (OETC), supported by the Advanced Research Project Agency (ARPA), is a precompetitive industrial alliance between Martin Marietta, AT&T, Honeywell, and IBM. This Manuscript received July 5, 1994; revised November 16, 1994. This work was supported by Advanced Research Projects Agency, Microelectronics Technology Office under Contract #MDA972-92-C-0072. Y.-M. Wong, D. J. Muehlner, M. Fishteyn, T. Mullally, J. V. Gates, and P. J. Anthony are with the AT&T Bell Laboratories, Murray Hill, NJ 07974 USA. C. C. Faudskar is with the AT&T Bell Laboratories, Murray Hill, NJ 07974; previously with AT&T Bell Laboratories, North Andover, MA USA. D. B. Buchholz and J. L. Brandner are with the AT&T Bell Laboratories, Murray Hill, NJ 07974; previously with AT&T Bell Labs, Indian Hill Park, IL USA. W. J. Parzygnat is with the AT&T Bell Laboratories, Murray Hill, NJ 07974; previously with AT&T Bell Labs in Whippany, NJ USA. R. E. Leibenguth, G. D. Guth, M. W. Focht, K. G. Glogovsky, and J. L. Zilko are with the AT&T Bell Laboratories, Murray Hill, NJ 07974 USA; previously with AT&T Bell Labs, Breinigsville, PA USA. consortium is advancing the technology, benefits, and use of optoelectronic links for short distance, high bandwidth interconnection. The objective of OETC is to develop and test an optical data link technology that will meet the processing needs of the future, expediting the insertion of optoelectronic links into diverse commercial and military applications. By using a high bandwidth optical fiber medium, the OETC will overcome the existing “metal interconnection” barriers such as reflection, crosstalk, switching (A - I) noise, loaded capacitance, and skin effect that occur at high frequency. Fig. 1 shows the technology deployment in terms of bandwidth-link density design parameter space. The next generation of processor networks will place heavy demands on both processing and communication capability. Typical requirements are high aggregate processing rate, multiple concurrent VO operation capability, and individual and diverse high data-rate U 0 operation capability. Low latency is also an important requirement. The OETC’s goals for the end of 1994 for the 32-channel bus are “logic to logic” 32-bit word data transfers at 500 Mb/s per channel (aggregate data rate of 2 Gbytes per second) with excess latency of 2.0 ns above the “time of flight,” a power dissipation of less than 400 mW per channel, and a bit error rate of better than The OETC is also addressing the difficult problem of transferring technology from research laboratories to prototype fabrication. In the precompetitive stage, the OETC R. A. Morgan was with the AT&T Bell Laboratories, Murray Hill, NJ 07974; now with Honeywell Technology Center, Bloomington, MN. B. H. Tyrone, Jr., T. J. Ireland, D. H. Lewis, Jr., D. F. Smith, and D. K. Lewis are with the Martin Marietta Laboratories-Syracuse, Electronics Parkway, Syracuse NY 13221 USA. S. F. Nati was with the Martin Marietta Laboratories-Syracuse, Electronics Parkway, Syracuse NY 13221 USA; now with IMRA America Inc., Ann Arbor, MI USA. D. L. Rogers, H. A. Aispain, S. M. Gowda, S. G. Walker, Y. H. Kwark, R. J. S. Bates, D. M. Kuchta, and J. D. Crow are with IBM, T. J. Watson Research Center, Yorktown Heights, NY 10598 USA. IEEE Log Number 9411551. 0733-8724/95$04.00 0 1995 IEEE JOURNALOF LIGHTWAVE TECHNOLOGY, VOL. 13, NO. 6, JUNE 1995 9% Transmitters Channel lbo lo00 100 10 Density (Linedcm) 1000 Receivers tine Pitch (pm) Fig. 1. Interconnect technology comparisons. VI1 provide link architecture and characterization, respectively. A summary and conclusions are provided in the last section. 11. BRIEFCOMPONENT DESCRIPTION The OETC is developing components for a 32-channel, parallel bus that will be integrated into a testbed to demonstrate the practicality of an optical data bus for computer interconnection applications. The present bus contains a directly modulated optical transmitter (Tx), a 32-channel fiber ribbon, and an optical receiver (Rx). Fig. 2 is a photograph of the completely fabricated Tx (directly modulated) and Rx modules. In the directly modulated Tx, a 32 x 1 GaAs laser array driver (with a Manchester-encoder and current bias and modulation control circuits) takes 500 Mb/s, Non-Return-to-Zero (NRZ), Emitter Coupled Logic (ECL)-compatible, differential data input and converts it to Manchester-encoded modulation current that drives each laser in a GaAs surface emitting laser array. The optical output of each laser is sent through a MACII-32 fiber array ribbon cable (multimode, graded index) to a GaAs monolithic integrated metal-semiconductor-metal Fig. 2. A photograph of the completely fabricated Rx module (left) and Tx module (right). (MSM) receiver that outputs 32 differential, ECL-compatible signals to drive the next chip set. A. Vertical Cavity S u ~ a c eEmitting Laser (VCSEL) Array The VCSEL structure consists essentially of a region that contains the active quantum wells at its center, encompassed between top and bottom quarter-wave Distributed Bragg Reflector (DBR) mirrors for feedback. This 6-8 pm thick Fabry-Perot cavity emits radiation at a wavelength roughly corresponding to its transmission resonance at approximately 850 nm. The present VCSEL arrays are grown [ l ] by Molecular Beam Epitaxy (MBE) which allows high quality material and layer thickness control. The planar top-emitting VCSEL is gain-apertured using hydrogen-ion implantation. This design was chosen for its performance, producibility, and wavelength flexibility. Fully batch-processed working VCSEL wafers have been demonstrated [l] for various sizes and geometries in various applications. WONG et al.: TECHNOLOGY DEVELOPMENT OF A HIGH-DENSITY 32-CHANNEL 16 Gbls OFTICAL DATA LINK 0 0 0 Preamp/ 0 Postamp Drivers 0 15 - Opt 15 15 997 crosstalk among channels. The clock channel is identical to the data channels up to the comparator circuit. The clock signal then goes to a multiplexer that chooses between the optical clock or an external electrical clock used for wafer testing. The multiplexed clock signal is then used to latch the decoders on the data channel in the first half of the Manchester baud interval. The low power consumption of 2 W is achieved by using -3.5, -3.3, and -2.5 V power supplies, and minimal signal levels to drive off chip. Optical line rates greater than 1 Gbaud with a typical sensitivity of -18 dBm have been demonstrated. Fig. 4 is the schematic of a single channel. C. Laser Driver IC Array 0 0 0 I Transparent Switch Fig. 3. Overall schematic of Rx array. B. MSM Receiver Array The detectodreceiver array is a 32-channel optoelectronic integrated circuit (OEIC) fabricated using standard 1 pm GaAs MESFET technology [2]. An overall schematic of the receiver array is shown in Fig. 3. The array consists of 31 data channels and one clock channel. Each data channel incorporates a metal-semiconductor-metal (MSM) detector, a preamplifier, a postamplifier, a level restoring circuit, a decision/decoder circuit, and off-chip ECL drivers. The MSM detectors are low capacitance, high speed devices that can be reliably integrated into a GaAs MESFET process. They drive transimpedance amplifiers that are ac-coupled to the detectors to minimize the amplifiers’ sensitivity to dc offsets. Level restoring circuits sample the amplifier outputs and provide feedback to enhance the receivers’ dynamic range to greater than 15 dB. The postamplifiers and buffers boost the preamplifier output sufficiently to drive the decoder. Any electrical offset in the postamplifiers and buffers is compensated for by a second level restore circuit. The Manchester decoder is fabricated using source-coupled FET logic (SCFL) to minimize power dissipation and the noise generated on-chip. It decodes the differential Manchester coded data into differential NRZ data at 500 Mb/s. The off-chip drivers provide ECL logic level through externally terminated source followers. This high density (32 channels on 140 pm pitch) receiver array is fully differential internally to suppress common-mode noise and The laser drivers were fabricated at a GaAs foundry using an enhancement/depletion-mode (ED) GaAs MESFET process (gate lengths 51 pm) combined with SCFL. The integrated circuit contains a digital and an analog section having 31 data channels, 1 clock channel, and 2 control inputs. Each data channel consists of a differential ECL-compatible input buffer, a 500 Mb/s Manchester encoder, and an analog output section that converts the digital logic signals into modulation currents to drive each VCSEL diode in the array. In the digital section of the laser driver, a programmable clock skew adjustment is included to adjust the clock in 80 ps increments. The two additional control inputs are configured on the driver IC to maintain uniform laser power levels in the presence of variations in temperature, supply voltages, and device aging. These inputs are part of feedback circuits, each consisting of a monitor laser and detector (on the VCSEL chip), a currentto-voltage converter and subtracter (separately integrated in the Tx package), and a laser driver. One circuit sets the low level current limit of the laser drivers and the other sets the upper current limit. As temperature variations, supply voltage changes, and aging cause the optical power of the monitor lasers to change, the monitor detectors will sense these changes and will raise or lower the bias and modulation currents of the laser driver accordingly. This enables the laser driver to uniformly modulate currents of 1-9 mA peak to peak to each laser. Fig. 5 is a block diagram of the laser driver and feedback control. D. Fiber Array Block (FAB) Fig. 6 is a schematic of the FAB, a connector structure that incorporates elements of the Multifiber Array Connector (MACIITM)technology [3]. The FAB serves three purposes: 1) It couples light from the VCSEL array to the fiber ribbon cables and from the fiber cables to the MSM-receiver; 2) it forms an integral connector that mates to the MACII-32 cable directly without a pigtail; and 3) with the heat spreader, it forms a robust platform to start the packaging sequence. E. MACII-32 Fiber Cable This is an enhancement of the standard cable used with the MACIITMtechnology. Standard multimode fiber with 62.5 pm diameter graded index (GRIN) core and 125 pm diameter JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 13, NO. 6, JUNE 1995 998 +3.3v Clock -L MSM s Data out Detector Fig. 4. Decoder U Level Restore Driver w Schematic of single Rx channel. Bias voltage Bias Voltage l - - = l 0.0luF Fig. 5. Block diagram of the laser driver and the feedback control. Electrical In Heat Spreader Fig. 6. Cross section of ATLT Tx Package design. cunm witwing 1.OuF WONG et al.: TECHNOLOGY DEVELOPMENT OF A HIGH-DENSITY 32-CHANNEL 16 Gb/s OFTICAL DATA LINK t 4.45 cm Optical SubAssembly (OSA): 45" (Au Coated) Turning Fiber Array Mirror 10.53 cm p0mo-f Fig. 8. AT&T low profile OETC T(R)x module. MSM Receiver SiOB Fiber Array Blodc I 0. Heat Spreader (Bottom Cover) w \ m r r r . . . . . . . . . . . . . . . . . . . . . . mr c NO.2-56 FHM SCR'S Fig. 7. Rx (Tx) package design concept OETC-AT&T-ARPA. cladding forms the 32 x 1 fiber ribbon. The polyimide protective coating thickness, however, has been reduced to conform to the 140 ,urn (standard optical cable pitch is 250 ,urn) pitch requirement between two adjacent channels. F. Package Platform Common to Both Tx and Rx Modules The optoelectronic package concept shown in Fig. 7 consists of a heat spreader, VCSEL or MSM receiver array, an optical subassembly (OSA), a MCM-D (POLYFIC), lead frame assembly, and a machined aluminum cover. A finned 1/2-in aluminum heat sink (not shown) is added for heat dissipation. Heat Spreader: The heat spreader is machined from SilvarTM [4] to provide alignment features for the silicon base plate and the MCM-D. SilvarTMis chosen for its high thermal conductivity (153 W/m-K) and for a coefficient of thermal expansion of 6.5 x 10-60C that matches that of the POLYFIC substrate (alumina, 6.7 x 10-6/oC). The SilvaTM heat spreader, which is nickel-gold plated after machining, incorporates features for the attachment of an aluminum cover. An aluminum heat sink is attached to the heat spreader. Optical Subassembly (OSA): This consists of a silicon base plate onto which are attached a fiber array block (FAB) and the VCSEL array. The FAB consists of a 32 multimode fiber array sandwiched between two precision etched V-groove Si pieces. The end that is towards and above the VCSEL or MSM receiver array is polished at 45'. It is then gold-coated to form a turning mirror to capture the output light from the VCSEL array or to direct light to the MSM receiver array. The other end is polished at 90" to mate to the MACII-32 ribbon cables. A photograph of the OSA is included in Section I11 where its physical design is described. Wire Conducting (b) Fig. 9 (a) Schematic of the VCSEL layout. (b) Mounting of monitor PIN'S on SEL chip. MCM-D: The multichip module consists of a Film Integrated Circuit (FIC), a lead frame, a heat spreader, a cover, and active devices. The present OETC POLYFIC is a modification of one of AT&T's standard MCM-D technologies [5] that allows 32 optical VO's on one side of the RxRx module. The aluminum cover is designed to be attached to the pre-molded lead frame and the heat spreader. 111. PACKAGEDESIGN Fig. 6 shows the schematic cross section of the Tx module package. In order to keep package commonality, the Rx module is almost identical to the Tx module in design concept except that 1) the VCSEL array, the two associated monitor photodetectors, and the laser array driver IC and its associated optical amplifiers are replaced by an integrated MSM receiver array IC, and 2) there are no integrated termination resistors on the POLYFIC of the Rx module. We will focus our discussion on the Tx module package design. Fig. 8 shows the relevant physical dimensions of the completed assembly without a heat sink. VCSEL Array Fig. 9(a) shows the top side features of the VCSEL array. The area occupied by the laser sites is very small, and much of the chip area is dictated by the inclusion of the two monitor JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 13, NO. 6, JUNE 1995 / Substrate Via / Thin-film Resistors Referem Plane M Conductw (option,) Fig. 10. AT&T multichip module-D (POLYHIC) cross section. PIN diodes (to be discussed later) shown in Fig. 9(b). The metal interconnection linewidth is governed by the data rate requirement of 500 Mb/s Manchester encoder (see electrical design section). Visual alignment marks are laid out on both the front surfaces of the VCSEL and MSM receiver arrays. The alignment tolerance of the 62.5 pm core GRIN multimode fiber FAB to optical array devices will be discussed later. Multichip Module-D Fig. 10 shows a typical cross-section of the polymer hybrid integrated circuit (POLYHIC) [6]. The 1-in square substrate is 99.6% alumina. The low-loss TiPdCuNiAu metalization has a sheet resistance of 0.005 Wsquare. A ground or conductor plane is provided on the backside of the substrate. The ceramic substrate has via holes and a via slot. The slot provides frontto-back side ground plane continuity along a portion of one edge where good termination of the signal from the wire bonds is needed. The triazine-based photodefinable dielectric [7] has a low dielectric constant of 2.8, a dissipation factor < 0.0025, and a typical thickness of 35 pm. The lead frame (a modified 164 VO pre-molded PQFP) has a 25 mil lead pitch along three sides (resulting in 123 VO's) of the ceramic substrate. Physical Design I ) Optical: The mechanical and optical designs of the optical subassemblies for the OETC transmitter and receiver are deliberately very similar, allowing the complete modules to be quite similar also. Fig. 11 is a 3-D view of the transmitter OSA. The VCSEL chip is mounted horizontally on a silicon base plate, with the lasers emitting in the vertical direction. The light from the 32 data lasers is coupled to 32 short sections of 62.5 pm core multimode GRIN fiber in the fiber array block (FAB) via 45" mirrors polished onto the ends of the fibers. The gaps between the VCSEL's and the cylindrical surfaces of the overlying fibers can be filled with an index-matching material to improve the light coupling, to reduce feedback to the lasers due to reflections from the air-fiber interface and to keep unwanted material from getting between the lasers and fibers. Two monitor lasers on either end of the VCSEL array chip are capped by silicon PIN'S, the photocurrents of which are used in feedback circuits to regulate the laser array bias and modulation currents. To ensure that the monitor PIN'S will not change the light-current characteristics of the monitor lasers Fig. 1 1 . 3-D view of the optical subassembly (OSA). by strong feedback from reflections off their nearby surfaces, the pins are tilted at angles of approximately 7". The receiver OSA is almost identical. Total internal reflection (TIR) has been used [8] elsewhere to reflect light into or out of fibers. Unfortunately, TIR is not suitable in our application. We use a 45" angle for the mirror for best matching of the laser beams to the fibers. The minimum angle for TIR in the fiber is about 43", so that any light more than 2" from the optic axis on one side would not be totally reflected. The angular spread of the laser beams incident on the mirrors is likely to be considerably larger than this. At incident angles less than the critical angle, the internal reflectance of silica is low and depends strongly on polarization, so that it is necessary to have an efficient reflective coating with low polarization sensitivity (to keep polarization fluctuations from being converted to amplitude noise) on the 45" fiber ends. Gold is the best metal for the mirror coating. At a wavelength of 850 nm and an internal incidence angle of 45 f lo", gold on silica has a calculated reflectance of -98%, with the two polarizations differing by less than 2%. A disadvantage of gold is that it does not adhere well to bare silica. Metals, like chromium and titanium, that do adhere well are comparatively poor reflectors. To make the gold adhere better to the fiber ends, we used a chemical [9] treatment that puts a nearly optically transparent adhesion layer on the silica fiber ends. Measurements made on FAB's with this coating indicate that the average reflectance is -95% or better and that the difference between polarizations is <5%. 2) Electrical: Controlled impedance lines are provided on the MCM-D. One ground and one power plane are supplied to the circuit. A power or ground lead is placed between pairs of signal leads wherever possible to reduce electrical crosstalk. On the Tx module, signal lines are terminated very close to the driver chip with precision 50 C2 integrated TaN thin film resistors. The signal traces from the lead frame to the laser driver chip or the receiver array chip are laid out with almost identical length to minimize possible electrical signal skew. Decoupling (or power/ground) capacitors are added to the package by solder reflow attachment of discrete capacitors - WONG et al.: TECHNOLOGY DEVELOPMENT OF A HIGH-DENSITY 32-CHANNEL 16 Gbls OPTICAL DATA LINK 1001 Vertical Spring Modiied MACH Socket Clip Silicon FAB Silicon Base Plate Spacer Silicon Base Plate L Alignment Pin from Connector Plug Fig. 13. MACII interface connection for the transmitter and receiver packages (exploded view). The simulation also shows that improved rise and fall times of the output signal are possible by more closely matching the impedances between the signal trace on the VCSEL chip (24 R) and the VCSEL load resistance (250 R). 0 0.5nsldiv (4 Fig. 12. (a) Circuit element schematic from driver IC to VCSEL. (b) and (c) input current and output waveforms for (a). (d) NN and NNN crosstalk waveform on the center quiet line. onto the MCM-D. Data rates (into and out of the module via the lead frame) exceeding > 1 Gb/s have been previously demonstrated [lo]. Signal Integrity A seven section circuit model (Fig. 12(a)) was used to calculate the electrical characteristics of the line connecting the laser array driver chip to the VCSEL array. The characteristic impedance, inductance per unit length, and capacitance per unit length (self and mutual) of each section were computed using an electromagnetic field solver, GreenfieldTM[ 111. From these results, lumped element models for each section were constructed. The input signal is a 5 mA pulse train rise, fall times are 250 ps, and the period is 2 ns. Fig. 12(b) and (c) show the input and the output signals. The output voltage is 1.20 V, 96% of the expected 1.25 V from 5 mA into 250 R. Electrical Crosstalk Effects of the nearest neighbor (NN) and next nearest neighbor (NNN) lines on a central quiet line (electrical crosstalk) were also simulated as shown in Fig. 12(d). Crosstalk due to the two nearest neighbors on both sides of the quiet line is about 4.5% of the active signal of 1.2 V. 3) Mechanical: A new MACII-32 type interface connector was designed for the transmitter and receiver packages. Fig. 13 shows how the spacer, modified clip, and the vertical spring are designed for drop-in assembly over the silicon base plate. The vertical spring applies a 2.0 lb clamp force to the MAC-like FAB. This force counteracts the force from the alignment pins that tends to separate the FAB (1.6 lb). The vertical spring is a backup to the epoxy that bonds the FAB together. This backup is normally provided by the standard socket clip on a MACII assembly. However, the clip was modified by removing one side for drop-in assembly; therefore, the clamping force had to be provided by different means. The modified clip provides about the same alignment pin force as a standard clip (2.25 lb). The spring constant of a standard or modified clip is about 0.1 lb per 0.001 in of change in the distance between the cantilevers that apply the alignment pin force. JOURNAL OF LIGHTWAVE TECHNOLOGY,VOL. 13. NO. 6, JUNE 1995 1002 FAB C OSA MCM-D Fig. 14. OETC Tx module assembly. The clip is preassembled to the spacer before this subassembly is installed on the submount. The spacer is designed to spread the cantilever springs open enough to clear the submount during installation. When the alignment pins plug into the FAB, they spread the cantilever fingers further thus applying the full force of the cantilever springs to the pins. The spacer also positions the clip so that it clears the FAB base plate by 0.005 in. IV. ASSEMBLY Optics Construction of an OSA for a Tx or Rx module includes assembling the fiber array block (FAB), aligning and bonding the VCSEL or MSM detector array to the silicon base plate, and epoxying the FAB to the base plate. The FAB mirrors are aligned over the lasers or detectors before the epoxy is thermally cured. The assembly steps for the FAB are indicated on the left hand side of the flow chart in Fig. 14. First, an array of 32 cleaned-bare fibers (obtained from a section of the OETC 32-fiber cable) is laid into the 32 grooves of a precision etched silicon V-groove part. A second grooved silicon piece, shorter than the first to allow the FAB to overhang the laser or detector chip, is placed over the fibers and epoxy is wicked into the grooves and cured to cement the sandwich together. One end of the longer silicon piece and the fibers it carries are polished at 45" and then the other end of the assembly is polished perpendicular to the fibers, bringing the FAB to the correct overall length. Finally, following the gold adhesion process, a 1200" A-thick gold film is evaporated onto the beveled end of the FAB. 1003 WONG et al.: TECHNOLOGY DEVELOPMENT OF A HIGH-DENSITY 32-CHANNEL 16 Gbls OPTICAL DATA LINK The VCSEL or MSM detector array chip is bonded to the silicon base plate using conducting epoxy preforms, in a jig that positions the chips so that alignment marks on the chips line up with corresponding marks on the base plate. The FAB, in turn, is aligned to the laser or detector chip when it is also epoxied to the base plate, before the epoxy is thermally cured. Alignment of the fiber mirrors to the lasers or detectors is done visually, looking down onto the array with a video camera through a zoom microscope. Approximately 25 or 30 pm, or at least 20% of the fiber diameter, is available in each direction for optimum coupling. Using reference marks on the chips, it is not difficult to center the fiber mirrors over their respective devices to this accuracy. Fig. 9(b) shows a view of a laser chip on the mirror end of a FAB and a mounting of amonitor PIN on a SEL chip. Naturally, the 32 mirrors of the FAB cannot be independently aligned to 32 lasers or detectors; instead, the alignment is done near the middle and checked at the two ends of the array. After the FAB is aligned to the laser or detector chip, the epoxy between it and the silicon base plate, which acts as a lubricant during the alignment process, is heated further and cured. MACII-32 Ribbon Cable with Connector The end of the fiber ribbon is assembled into the connector without stripping the polyimide coating. This increases the connector reliability because the fiber is not weakened by the stripping process. The 32 fibers in the ribbon are positioned by clamping and then epoxying between two precision-etched V-grooved silicon chips and plastic jackets. The end is then polished (Fig. 15(a)), the alignment steel pins and spring clip are assembled, and the housing is installed (Fig. 15(b)). This allows direct “plug and play” connection with the Tx or Rx modules, which have similar interface connector designs, as shown in the schematic drawing in Fig. 16. (b) Fig. 15 (a) Polished end face of a MACK32 connector. (b) Special interface cable for the Tx and Rx module. the resonance concur at elevated temperatures. Moreover, given the inherent connection between layer thickness control and lasing wavelength (determined by the cavity resonance) for manufacturing, it is likewise important for VCSEL’s to exhibit minimum sensitivity to variations in wavelength (thickness). Control is also imperative for uniform operation of one- and two-dimensional VCSEL arrays (for simple driving and monitoring schemes), where thermal heating may be inMCM-D Assembly creased and wavelength may vary across the array. Wavelength The details of the MCM-D assembly process are given by and temperature performance are also intrinsically linked. As the flow chart on the right hand side of Fig. 14. shown in Fig. 18, we have demonstrated [12] over a 50 nm wavelength range of operation with only a factor of two variation in threshold current and a statistically flat threshold v . LINK COMPONENT PERFORMANCE over an -30 nm wavelength range. Within this top-surface emitting VCSEL platform, we have VCSEL Device Characteristics also developed [13] a novel technique of obtaining single For use in commercial environments (e.g., a computer longitudinal and transverse mode emission by exploiting a backplane), it is necessary for VCSEL’s to exhibit adequate monolithic spatial filter. The emission is well behaved and lasing performance over a wide range of temperatures. Fig. 17 remains in a single TEMoo-like mode with typical cw power displays the distribution of threshold current, voltage, and levels exceeding 1.5bmW. Even for drive currents as high power at a 10 mA drive current for temperatures of 20 as 11 mA (-2.5 X I & the transverse side mode suppression and 100°C for one array. Excellent uniformity is obtained ratio was still measured below -40 dB, comparable to or better with standard deviations below 6% for these three properties than most edge-emitting DFB lasers. These devices have been over this temperature range. Note that device performance modulated with bandwidths in excess of 8 GHz. The far field actually improves with increasing temperature. This results is circularly-symmetric with measured FWHh4 divergence from implementing the following design strategies: 1) decrease angles -1 lo, corresponding to a near-field Gaussian (e-’) self-heating by lowering resistance and improving efficiency; diameter = 3.8 pm. Hence, spatially-filtered VCSEL’s are 2) improve carrier confinement by incorporating high con- well-suited for coupling into single mode or multimode fiber finement barriers; and 3) blue-shift the room-temperature and should exhibit diffraction-limited imaging performance for gain from the Fabry-Perot resonance so the peak gain and implementation in free-space optical interconnection schemes. JOURNAL OF LIGHTWAVE TECHNOLOGY,VOL. 13. NO. 6, JUNE 1995 Modified MACll Ribbonkray Assembty ...--.. ..-. .. ........ .......: ' ( . -. I a . ..--.. 0':. .......=r : i: W.&=;. ................ .-.-A .: ............. . ; .................. .. ..' (a) Fig. 16. Interface connection layout. 20°C 70°C 100°C 32 Ith 24 s 0 E 2 16 CT L U, 8 0 0 1 2 3 4 5 d6 Parameters (mW-V-mA) Fig. 17. Histogram of a VCSEL 32 x 1 array threshold current, voltage, and power at 10 mA drive current taken at 20, 70, and 100OC. VCSEL Wafer-Level Characterization A major advantage of the VCSEL is that the most important device measurements (including L-I-V, temperahre performance, wavelength and spectral properties, beam quality, and polarization) can be made and qualification accomplished at the wafer level 4. With proper probe station automation, a wafer containing thousands of VCSEL's or VCSEL arrays, can be comprehensively tested and qualified before the expensive, laborious process of dicing and separating the wafer into chips. This distinct advantage over edge-emitters is an important impetus driving VCSEL development. A full wafer takes roughly a day to test and qualify. 1005 WONG et al.: TECHNOLOGY DEVELOPMENT OF A HIGH-DENSITY 32-CHANNEL 16 Gbls OFTICAL DATA LINK 14 - I I I 1 I I I 12 - " 4.5 mA 2 0 10 12 14 16 18 20 22 24 Row Site Wavelength (nm) Fig. 20. Iop wafer array distribution. Fig. 18. VCSEL threshold current versus wavelenth at 2OOC. lth: Wafer Array Uniformity l o t 1 I -----column ............ 4th 5th column -6th column .t 10 12 14 16 18 Row Site 20 22 24 (a) Wavelength: Wafer Array Uniformlty 0 0 1000 2000 Time (ps) Fig. 21. Eye diagram obtained from modulating the impedance-matched VCSEL at 700 Mb/s, RZ, PRBS 223 - 1. I I I I I I I I 10 12 14 16 18 20 22 24 Row Slte (b) Fig. 19. Radial dependence of the Ith and wavelength. VCSEL Array Uniformity On a 2-in.-diameter VCSEL wafer containing 42 nonedge VCSEL arrays and miscellaneous test chips, nearly all individual lasers were found to lase, resulting in a yield of 1422 out of 1428 devices. Approximately 88% of all arrays are fully operational. About 60% of all arrays exhibit less than f 1 0 % o (Ith) mean (Ith) ratio within each array. This intraarray uniformity is exploited in the OETC driver circuit design to eliminate the need for a feedback monitor photodiode and circuitry for each individual channel in the laser array IC driver chip. The inter-array I t h uniformity, however, shows a radial dependence (except some deviation for arrays near the wafer edge) due to variations of thickness and resulting operating wavelength (Fig. 19). Data connected by lines indicate the array average and standard deviation of the threshold current (or wavelength) of arrays in different columns on the wafer. This dependence results from the wafer rotation in the MBE growth process. These preliminary results, demonstrating uniform array operation (Fig. 20) and reasonable array yield, are very encouraging for highly parallel ODL application of VCSEL arrays. High-speed SEL Device Modulation For data-link system studies, a VCSEL was bonded in a high-speed package and tested with an Anritsu BER (bit error rate) system. Light was coupled into a 1 m long 62.5 pmcore graded index multimode fiber. This device was biased at 4.5 mA) and modulated with a current about 6 mA ( l t h swing of -3 mA (hence the laser was completely turned off -1.5 mA below threshold). The VCSEL is impedancematched by a parallel 62 R resistor. Fig. 21 displays the - JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 13, NO. 6, JUNE 1995 1006 1Gb/s 6mA (SM) -20 -18 -16 -14 -12 Received Power (dBm) Fig. 22. Bit error rate versus received power for a multitransverse mode (MTM) VCSEL transmitted (PRBS 27 -1) through a 6 m long 50 pm 125 pm GRIN MM fiber link for the given biases, i t h 6 mA. The ac modulation was approximately f 6 mA. -24 -20 -22 - wide open eye diagram obtained (albeit with the 850 MHz bandwidth detector) by operating the VCSEL at 700 Mb/s, RZ, 223- 1pseudorandom bit sequence. An error free transmission over a 23 hour period results in a BER or lower. Note that this low BER was obtained by butt coupling directly into the fiber facet, without any Manchester coding, and irrespective of any reflection-induced noise. Although single transverse-mode VCSEL's are necessary for efficient coupling into single-mode fiber-based optical interconnects, they may not be required or even as desirable for multimode fiber interconnections. This is because modal noise may, in fact, be reduced due to spatial averaging and decreased coherence of the larger multitransverse mode VCSEL. This has recently been demonstrated 1151, [16] where BER's < were measured using a multitransverse-mode VCSEL operating at 1 Gb/s. The BER's versus received power for a multimoded VCSEL at different drive currents and modulation rates are compared in Fig. 22. It is seen from the 6 mA (1 mode) and the 9 mA (3 modes) data, that almost no penalty is paid using multimode emission into this 6 m long 50 pm-core GRIN multimode fiber link. Moreover, even at speeds up to 3 Gb/s, BER's < are obtained for multimode operation [17]. Here, the speed is limited by the equipment and not the laser itself; thus, higher modulation rates may yet be attainable. VCSEL to FAB: Coupling EfJiciency, Alignment Tolerance, and Optical Crosstalk We have measured the coupling efficiency of a VCSEL laser to multimode fiber in a FAB as a function of the relative position of laser and fiber mirror to determine how critical the laser-fiber alignment is. The air gap between the laser window and the cylindrical surface of the fiber was about 40 pm. The results are shown in Fig. 23. The peak coupling efficiency was approximately 93%. The fiber could be moved in both horizontal directions over a range of more than 25 pm while keeping the coupling greater than 90% of the maximum. In general there seems to be at least a h12.5 pm} adjustment range for near-optimum coupling of a VCSEL laser to the multimode fiber via the 45' mirror, or at least f10% 20 0 40 (Micrometers) Fig. 23. Alignment tolerance of VCSEL to FAB. et - I (SEL) = 15mA (MM) / e - .------I (SEL) = lOmA (MM) ........-I' 0 20 I I I I I I 40 60 80 100 120 140 160 Approximate SEL-Fiber Gap (vm) Fig. 24. OETC Tx nearest neighbor crosstalk. of the fiber diameter. This large adjustment range makes it unnecessary to do any sort of active (power-on) alignment or solder reflow self-alignment. Taking care to adjust the depth of field properly and to control the direction of the light illuminating the 45" mirrors from the side, it is possible to do the alignment of the fiber array to the laser or detector array by visually observing the process under a nonstereo microscope. Channel-to-channel optical crosstalk has been measured at mocked-up transmitter and receiver ends of the OETC data link. Fig. 24 shows the nearest neighbor crosstalk for the transmitter as a function of the VCSEL to fiber air gap, with the laser operating mostly in a single mode at 10 mA and in a multitransverse mode at 15 mA. Crosstalk is higher in the multimode case, but is below -46 dB for gap sizes below 80 phm. the laser operating in a single mode it would be expected to be lower yet. To check compatibility between the MAC-like fiber array block on the optical subassemblies and the 32 fiber MAC connector that plugs into the FAB's on the optical subassembly, we measured the fraction of the light transmitted across the FAB-MAC interface with one fiber of the FAB illuminated by a VCSEL via the 45" mirror in the usual way. The result indicated that the loss at this interface was 0.5 dB or less, consistent with the coupling loss measured between identical MAC connectors. 1007 WONG et al.: TECHNOLOGY DEVELOPMENT OF A HIGH-DENSITY 32-CHANNEL 16 Gb/s OPTICAL DATA LINK Detector size: Average = 0 . W B Std. Dev. = 0.23dB 150 - Samples = 2048 ,I 0 0.1 ( I r oa . . . . . . . 0.9 1.0 1.1 1.2 1.3 1.4 1.5 MACII-32 Conmctor Lasa (de) Fig. 25. MACII-32 connector loss data (64 connectors, 850 nm wavelength, 623125 MM fiber). 25 . Core: 6 2 . 5 NA4.275 ~ 0 I I I I I I I I I Dlstancefrom f l k (run) Fig. 26. Beam size for MM fiber with 45 degree turning mirror. MACII-32 Connector Insertion Loss Thirty-two ribbon cables have been fabricated. Fig. 25 shows the connector insertion loss distribution at 850 nm wavelength for 2048 measurements. The average insertion loss and its standard deviation are 0.49 and 0.23 dB respectively. Fiber Filled with White U g M FAB to MSM Detector: Coupling EfJiciency, Alignment Tolerance, and Optical Crosstalk At the receiver end of the fiber link, we want all of the light emerging from a multimode fiber and reflecting from the output 45” mirror to fall on the active area of the corresponding MSM detector on the 32-element detector array. Since the actual shape of the spot of light emerging from the fiber will depend on which modes of the fiber are illuminated at the output end, which will vary from fiber to fiber and is not known to us, we have assumed that all of the fiber modes are filled and require that the resulting light spot fall entirely on the detector. The cylindrical face of the fiber that the light emerges from focuses the beam in the transverse direction only, so that the light spot is oval. A measurement of the size of the oval spot of visible white light emerging into air after reflection off a fiber end mirror, shown in Fig. 26, agrees well with the calculated envelope. The MSM detector active area is an octagon 100 pm long in the longitudinal direction and 80 pm wide in the transverse direction, as indicated by the dashed lines in Fig. 26. For an air gap in the expected range of 20-80 pm, the light spot size at the detector is at least 20 pm smaller than the detector in both directions, leaving sufficient room for a visual alignment to get the entire light spot on the detector. Optical crosstalk at the receiver end is expected to be higher than at the transmitter end because the fibers over the lasers accept only rays incident in a fairly small cone while any light falling on the MSM detector active area will be detected. The receiver crosstalk was measured for two diameters, 80 and 100 pm, of MSM detectors on a test chip with an air gap of about 50 pm between fibers and detectors. To ensure that all fiber modes were uniformly filled for the sake of repeatability, white light from a high intensity bulb was used. As illustrated in Fig. 27, the nearest neighbor crosstalk was -21 dB using the 100 pm detector and about -36 dB for Detector Diameter = 1Oqrm =squn Fig. 27. Crosstalk with IJ3M sample MSM detectors. the 80 pm detector. Second nearest neighbor crosstalk in both cases was about 3 dB lower. For the elongated OETC detector, the crosstalk level will be most sensitive to the detector size in the direction toward its neighbors, so that for the actual receiver the crosstalk should be close to that measured with the 80 pm detector. If an index matching fill is used between fibers and detectors, the crosstalk can again be expected to decrease due to diminished multiple reflections. Laser Driver Array On-wafer testing of the laser driver arrays was performed using a custom probe card and the TOPAZ-VTM Design Verification System along with its Parametric Measurement Unit. The TOPAZ-Vm system is an automated tester capable of generating digital patterns for input pins and doing real time comparisons of digital data on output pins at data rates up to 110 MHz. Four wafers of laser driver arrays (120 driver arrays) were tested on a single channel basis for dc parameters and low speed functionality. Eliminating the edge devices on each wafer, the driver arrays yielded 58% (58 of 100). The uniformity of the current outputs at the “high” and “low” levels was within the design goal of f 1 0 % of the average current. Figs. 28 and 29 are the “high” and “low” average results over all devices from the four wafers viewed channel-to-channel. JOURNAL OF LIGHTWAVE TECHNOLOGY,VOL. 13, NO. 6, JUNE 1995 1008 Average of 58 High-Current Tests Avg -11.61 Max -10.73 ............................................................................ Average of 58 Low-Current Tests -4.0000 Min - - --4:56 Avg -4.43 -4.1000 StDev~ 0:12 Mm -4.12 zE - E -4.2000 ........................................................................................ ?! L 5 -f 0 -4.3000 c r 0 .- -4.4000 i; -4.5000 -4.60001 O I I r I N I ~ I I ~ I ~ I I w I I I ~ w r r r I r ~ I r I I O r r r r I I I N m r r N I I I O ~ N N N I I I I w ~ w N N N N I ct-t-l O o r N N N m m m O Driver Channel Number (#) Fig. 29. Laser driver low current uniformity. MSM Receiver Array A comprehensive system was developed for automated testing of the receiver arrays. Fig. 30 shows the wafer test setup. The test system and probe card allow single pass, onwafer testing of each receiver channel. The testing includes measurements of both dc parameters and high-speed testing. DC parameters included the supply currents and the output ECL level compliance. The high-speed testing includes bit error rate tests and measurements of the relative signal delay through the receiver chain. In addition, a test using a second Optical fiber Probe was Performed in which the clock signal was supplied optically to the center clock channel and the data read from one of the adjacent receiver channels. This test was performed for only one pair of channels in each array due ~ w ~ ~ O 1009 WONG er al.: TECHNOLOGY DEVELOPMENT OF A HIGH-DENSITY 32-CHANNEL. 16 Gbls OPTICAL DATA LINK OETC. L o t 3. Wafer 7 I ' I I /I I 2- U c a, 3 U a, L LI -11 U Fig. 30. Functional description of the wafer test station. to the difficulty of using dual optical probes. No measurable degradation was observed relative to the same measurement using the external electrical clock. The parametric wafer testing revealed that the EFET and DFET thresholds were well within specification but the DFET drain conductances were somewhat large. Nevertheless, the single channel wafer yields were quite high. Fig. 31 shows a histogram of receiver sensitivity measured on one of the best wafers at 1 Gbaud (500 Mb/s data rate). The single channel receiver yield was approximately 97% for receivers with sensitivities greater than -6 dBm. Fig. 32 shows an eye diagram at the output of one of these receivers in transparent mode with the decoding latch disabled. It is believed that the small duty cycle distortion is induced by the latch operating in transparent mode. VI. LINKDESIGN:ARCHITECTURE, SIMULATION, AND POWER BUDGET :24 -21 -18 -15 -12 -9 -6 -3 S e n s i t i v i t v (dBm @ l e - 7 e r r o r r a t i o 1 External Clock Latch M d e 1823 of 1984 data Fig. 31. Histogram of Rx sensitivity. Fig. 32. Eye diagram of Rx in transparent mode. a Manchester encoded modulation current that drives each of the transmitters. MACIIm connectors couple the light into a 32-channel fiber ribbon and onto the MSM detectors in the front end of the optical receiver. Each of these components is described in detail in Section II. Fig. 33 depicts the link architectures. Architecture Two link architectures are being developed to demonstrate Link Simulation the practicality of parallel, optical data buses for computer The link's performance was accurately simulated to verify interconnection applications, semiconductor laser-based and that the technical approach would achieve the desired link modulator-based. [18] The major difference between the two performance. From the measurements made on the protoarchitectures is the light source. The laser-based approach type VCSEL's, modal noise was identified as potentially uses the directly modulated VCSEL diode array (detailed the limiting factor in achieving the OETC Bit Error Rate description given in Section 11-A), while the modulator-based (BER) target of Using a quasi-single mode VCSEL, approach uses a remotely-located, single-mode, temperature biased well above threshold to minimize skew and relative stabilized laser and a four-element GaAs/AlGaAs waveguide intensity noise (RIN), the link performance becomes very amplitude modulator array. The externally modulated trans- sensitive to mode-selective loss. Experimentally measured mitter package consists of the laser, a single-fiber input, a 1 BER floors due to mode selective loss mechanisms of 3 dB are x 4 fanout circuit, four waveguide modulators, and four fiber illustrated in Fig. 34. This sensitivity was quantified using a outputs, all mounted on a ceramic header. Push-pull pairs of simulation tool [19] based on modal noise analysis [20]-[22]. these modulators are connected to form a waveguide version The analysis is valid for noise penalties less than 2 dB, where of the Mach-Zehnder interferometer, in which electrically it remains reasonable to model the total receiver noise as a induced polarization rotation produces interference between Gaussian random variable. Actual high-frequency modal noise phase coherent light waves. has a more negative exponential distribution, [23] so that this In the laser-based approach described in this paper, a GaAs analysis will tend to underestimate the severity of the resultant driver circuit takes differential data input and converts it to BER floor. JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 13, NO. 6, JUNE 199.5 1010 Features 32-channels - array based Data bandwidth 500 Mbps per channel Line pitch - 140 pm per channel Data transfer latency e 2ns I I I,--------- aa& Connectors Fig. 33. The optical link architecture. = MSL 70z = 25z 2 x 0.6dB mean, 0.2dB SD L-----T c 1.0 5 C 5 IO-* 0.8 0 C .c 0.6 n .0 .c o N L + ln 5 10-5 U * L 2 G 0.4 W I .c U - 0.2 3 10-6 6 ' 0 0 0 .m 4 10" 10'8 1 2 3 Optical Link Penalty, d B d u e to modal noise ' c Fig. 35. Cumulative distribution of the optical link penalty in the presence of mode-selective losses. - 2 - 1 Relgtive 0 1 2 3 Optical Power, dB Fig. 34. Measure BER versus received optical power. The calculations assumed a laser with extinction ratio of 1 : 4 operating principally single mode, but with occasional mode-hopping, two connectors in the link (as proposed in the basic minimalist OETC application), and a receiver with the threshold detector biased at the mid-signal level. The calculations were made for two sets of values of connector losses, 0.6 f 0.2 dB. Fig. 35 shows the results of the simulation. For mean connector losses each of 0.6 dB, these results show that the modal noise penalty should be less than 1.5 dB, even with a poor fiber mode-fill of 25%; however, for mean connector losses each of 1.2 dB, it is essential to ensure a large modefill. If this is not done, then a penalty larger than 1.5 dB may be expected. The computer simulation indicated that the proposed link specification for laser modulation conditions and connector losses would be acceptable. 1011 WONG et al.: TECHNOLOGY DEVELOPMENT OF A HIGH-DENSITY 32-CHANNEL 16 Gbls OPTICAL DATA LINK Ma 1 .o 3 tO.OOdB 0.83 - -0.81dBrn 0.8 o.70 E $ -1.02dBrn -1.43dBrn 0.66 0.6 -2.29dBrn a Min 0.4 -7MF - 0.4 -4.2d0rn -4.95dBrn 0.2 0.02 - 0.0 IO.,UBRl SEL CPt MAC Pack $1 Fiber MAC $2 RX Det RX Design Optical Link Fig. 36. Optical link power budget. Power Budget The optical link power budget shown in Fig. 36 illustrates successive loss contributions for each element of the link. Maximum, nominal, and minimum curves represent the best to worst case link behavior anticipated. The maximum curve assumes the greatest peak optical power from the laser and minimum losses due to the link components. This would provide the largest signal at the receiver. The minimum curve presumes the lowest peak power for the laser source and the maximum losses for the link components. Nominal values were used to predict typical performance. The range of optical power at the detector was used to define the range over which the receiver should operate and provide adequate system margin. VII. COMPLETE LINKPERFORMANCE CHARACTERIZATION Eight OETC 32-channel optical data links have been successfully developed and are currently being characterized. A link consists of a Tx/Rx pair with optical fiber of lengths from 1.0-98.0 m between them. Thirty-one of the channels are used for data and one channel is dedicated to the clock. Fig. 37 illustrates the test fixture and modules with 2.0 m of optical fiber. A 500 MHz clock signal, which can be applied simultaneously to all 32 channels, is used to determine the link functionality. Once the link is determined to be functional, data can be applied. The clock/data is viewed at the Rx in its transparent mode, a diagnostic mode in the Rx before the data is latched. Data at 500 Mb/s and 622 Mb/s have been transmitted over each of the 31 data channels individually and over as many as 19 parallel channels with low error rate. The number of parallel channels that could be simultaneously Fig. 37. Test fixture and modules. exercised was limited by the availability of high-speed data sources. Fig. 38(a) and (b) show the eye diagrams of the transmitter module and receiver module outputs at 500 Mb/s. The optical data are Manchester-encoded in the Tx. As a result of the encoding, the eye pattern is essentially unchanged for 27- 1 or 223- 1 pseudorandom bit sequence (PRBS). The overall BER of a 98 m link having 17 parallel channels was measured to be less than 8 x The link performed over temperatures ranging from 0-7OoC. These data were taken using a 27- 1 Pseudorandom bit sequence at a clock frequency of 500 MHz. The channel-to-channel uniformity of the transmitter output was better than f 1 5 2 pW. The maximum skew between channels is approximately 500 ps; JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 13, NO. 6, JUNE 1995 1012 wide, parallel path over 98 m between two points that operates continuously at the data rate of 500 MbMchannel with a bit error rate of less than 8 x ...... ACKNOWLEDGMENT 50 rnv Idiv Support from the other members of the AT&T Bell Laboratories VCSEL fabrication team in Breinigsville, PA, and Murray Hill, NJ, is acknowledged. REFERENCES . . . . . . . . . . . . . . . 45.6211s 1nsldiv 55.6211s (a) . .. .. . I .......... . . . . . . 200 mv Idiv A . . . . ................. . . . . .............. . . . . . . . i -~ . ---. 45.6211s ..... . . 1 nsldiv 55.62ns (b) Fig. 38 (a) Eye diagram of the transmitter module at 500 Mbls (Manchester encoded). (b) Eye diagram of the receiver module before data is latched at 500 Mbls (Manchester encoded). however, the skew adjustment of the clock on the laser driver reduces the relative skew to less than f 250 ps. Interchannel crosstalk due to optical leakage or electrical interference among the channels has not been observed. The transmitter dissipated an average of 6.2 W, and the receiver dissipated an average of 2.0 W. VIII. SUMMARY AND CONCLUSION A 1 x 32, 140 pm pitch, 500 Mb/s N E , point-to-point optical data link was fabricated using existing GaAs IC, SiOB, and MCM technologies. The transmitter module has an 850 nm VCSEL diode array, with an IC driver array chip. The receiver module includes an integrated receiver chip with MSM photodetectors, MESFET amplifiers, decision/decoders, level restorers, and off-chip drivers. In both modules, the active devices are assembled with a POLWIC and packaged in a modified 164 VO JEDEC premolded PQFF’. The optical medium is a 1 x 32, fiber-optic ribbon cable interfaced to the laser and receiver chips with a special interface connector. The data link has demonstrated the successful operation of a 32-bit R. A. Morgan, L. M. F. Chirovsky, M. W. Focht, G. Guth, M. T. Asom, R. E. Leibenguth, K. C. Robinson, Y. H. Lee, and J. L. Jewel], “Progress in planarized vertical cavity surface emitting laser devices and arrays,” SPIE 1562: Devices for Optical Processing, pp. 149 (SPIE, Bellingham, WA, 1991); R. A. Morgan, K. C. Robinson, L. M. F. Chirovsky, M. W. Focht, G. D. Guth, R. E. Leibenguth, K. G. Glogovsky, G. J. Przybylek, and L. E. Smith, “Uniform 64 x 1 arrays of individually-addressed vertical cavity top surface emitting lasers,” Electron. Lett., vol. 27, p. 1400, 1991. J. F. Ewen, K. P. Jackson, R. J. S. Bates, and E. B. Flint, “GaAs FiberOptic Modules for Optical Data Processing Networks,” J. Lighhvave Technol., vol. 9, no. 12, p. 1635, 1991. W. H. Knausenberger, et al., “Optical interconnection technology for AT&T shelf and frame level equipment,” in Proc. 26th Annu. Connector and Interconnection Symp. and Trade Show, Anaheim, CA, Sept. 20-22, 1993, pp. 81-100. MACTM connector technology was developed at AT&T Bell Labs and now is a product of Berg Electronics. This material, a silver-invar composite, is a product of the Metallurgical Material Division of Texas Instruments. A. V. Shah, E. Sweetman, and C. K. Hoppes, “A review of AT&T’s POLYHIC multichip module technology,” in Proc. NEPCON West 91, Feb. 1991; Greg E. Blonder, R. A. Gottscho, and King L. Tai, “Interconnection processes and materials,” AT&T Tech. J., vol. 69, p. 46, 1990. C. C. Shiflett, D. B. Buchholz, C. C. Faudskar, R. D. Small, and J. L. Markham, “High-density multilayer hybrid circuits made with polymer insulating layers (POLYHICs),” in Proc. Int. Symp. Microelectronic (ZSHM), 1986, p.481. Developed by AT&T, see US Patent 4,554,229. K. P. Jackson, A. J. Moll, E. B. Flint, and M. F. Cina, SPIE Proc., vol. 994, p. 40, 1988. R. Fila, et al., AT&T Internal Memorandum. J. L. Brandner, C. C. Faudskar, M. E. Lindenmeyer, S. R. Hofmann, D. B. Buchholz, and J. E. Ballentine, “Electrical characterization of POLYHIC, a high density, high frequency, interconnection and packaging medium for digital circuits,’’ in Proc. 39th Electronic Components Con$, 1989, p. 759; J. L. Brandner and S. R. Hofmann, “Electrical characteristics of 25 MIL pitch JEDEC F‘QFP surface mount lead frames for multi-chip modules,” in Proc. IEEE ECTC, May 1991. Quantic Laboratories, Inc., Suite 200, 281 McDermot Avenue, Winnepeg, Manitoba R3B059, Canada. I. M. Catchmark, R. A. Morgan, K. Kojima, R. E. Leibenguth, M. T. Asom, G. D. Guth, M. W. Focht, L. C. Luther, and G. P. Przybelek, “Extended temperature and wavelength performance of vertical cavity surface emitting lasers,” Appl. Phys. Len., vol. 63, p. 3122, 1993. R. A. Morgan, G. D. Guth, M. W. Focht, M. T. Asom, K. Kojima, L. E. Rogers, and S. E. Callis, “Transverse mode control of vertical cavity top-surface emitting lasers,” ZEEE Photon. Tech. Lett., vol. 5 , p. 374, 1993. -, “Advances in Vertical Cavity Surface Emitting Lasers,” in SPIE OE LASE ‘94,Vertical Cavity Surface Emitting Arrays, Int. Symp. OR Optoelectron. and Microwave Eng., invited paper, vol. 2147, pp. 97-1 19, Bellingham, WA, 1994. K. H. Hahn, M. R. Tan, Y. M. Houng, and S. Y. Wang, “Large area multitransverse-mode VCSELs for modal noise reduction in multimode fiber systems,” Electron. Lett., vol. 29, p. 1482, 1993. D. M. Kuchta and C. J. Mahon, “Mode selective loss penalties in VCSEL optical fiber transmission links,” IEEE Photon. Technol. Lett., vol. 6, no. 2, pp. 288-290, 1994. D. M. Kuchta, R. A. Morgan, K. Kojima, M. T. Asom, G. D. Guth, M. W. Focht, and R. E. Leibenguth, “Multiple transverse mode VCSELs for high speed data communications,” in Con$ Proc. LEOS Annu. Meet., 1993. WONG et al.: TECHNOLOGY DEVELOPMENT OF A HIGH-DENSITY 32-CHANNEL 16 Gb/s OFTICAL DATA LINK J. P. G. Bristlow, S. D. MuWlerjee, M. Nisa Khan, M. D. Hibbs-Brenner, C. T. Sullivan, and E. Kalweit, “High density waveguide modulator arrays for parallel interconnection,” SPIE Proc., vol. 1389, Nov. 1991. R. J. S. Bates, unpublished work, 1992. R. Dandliker, A. Bertholds, and F. Maystre, “How modal noise in multimode fibers depends on source spectrum and fiber dispersion,” J. Lightwave Technology, vol. LT-3, no. 1, pp. 7-12, Feb. 1985. A. M. J. Koonen, “Modal noise in multimode fiber links with distributed mode-selective losses,” J. Opt. Commun., vol. 5, no. 4, pp. 141-143, 1984. _ _ , “Bit-error-rate degradation in a multimode fiber optic transmission link due to modal noise,” IEEE J. Select. Areas Commun., vol. SAC-4, no. 9, pp. 1515-1522, Dec. 1986. M. J. Lum, D. M. Fuller, A. Hadjifotiou, and R. E. Epworth, “Modulation-induced modal noise in digital systems-The prediction and measurement of bit error ratio,” in Proc. 10th European Con6 Opt. Commun., 1984. Yiu-Man Wong (A’86) received the Ph.D degree in physics in 1980, from the University of Rochester, NY, on the theory of Exciton transport His postdoctoral work at Catholic University of America is on time-dependent phase transition Since joining AT&T Bell Laboratones in 1984, he has worked on a dielectric isolated high voltage IC design (Reading, PA), established electrormgration design rule for the 0 9 p m CMOS technology, process integrated a fully planarized tungsten plug first-level metal process for the 0.6 p m CMOS technology (Allentown, PA), developed a resist and flip-chip bump bonding process for the 2-D Quantum Well Infrared Photodetector Array, and optirmzed bump bonding processes and thermal design for a 1-D edge emitting InP uncooled laser array (Breinigsville, PA) He is now a member of the Integrated Photonic Research Group, Murray Hill, NJ, and is technically managing the present optoelectronic MCM packaging efforts in an ARPA-sponsored Optoelectronic Technology Consortium Dirk J. Muehlner received the Ph.D. degree in physics in 1970 from the Massachusetts Institute of Technology, with a thesis on balloon observations of the cosmic background radiation. He taught and did research at MIT as an instructor and assistant professor in the Physics Department before joining AT&T Bell Laboratories in 1979. At Bell Labs, where he is a Distinguished Member of the Technical Staff, he has worked on magnetic bubble devices, photodetectors and lasers, and passive optical components. c. c. Faudskar graduated from ~~h Dakota state College of Science.Electronic Technology in 1969 and joined AT&T Laboratories the Same year, He is a member of theTechnical staff in the Ad. vanced Integrated Modules Business Unit at AT&T, North Andover, MA. H~ is currently a Multichip Module design engineer and has been working in hybrid integrated circuit and multichip module design and package development. D. Bruce Buchholz received the B.S. degree in chemical engineering from the University of Illinois and the M.S. degree in chemistry from Northwestern University in 1977 and 1983, respectively. He joined AT&T Western Electric in 1977, transferring to AT&T Bell Laboratories in 1984. He has worked in the area of thin film hybrids for 17 years with assignments in manufacturing engineering and process optimization, new technology transfer, process development, and product design. He is currently working in a free space optical switching department. 1013 Mikhail Fishteyn was born in the U.S.S.R. in 1947. He received the M.S. degree in electronics from the Leningrad Aircraft Instrumentation Institute, St. Petersburg, Russia, in 1970. From 1972-1989 he worked for the Vilnius Institute of Electronic Measuring Devices. As engineer and team leader, he took a part in design and development of a series of pulse and word generators and OTDR’s. In 1992 he started to work in AT&T Bell Laboratories in the Passive Optical Components Department. James L. Brandner (M’72-M’94) received the B.S.E.E. degree in 1965 from the University of Illinois at Urbana and the M.S. degree in metallurgy and material science in 1969 from Lehigh University, Bethlehem, PA. He has had various assignments at AT&T and since 1981 has been involved in the design and characterization of hybrid ICs, and multichip modules. His interests are currently in high-frequency characteristics of interconnection media. W. J. Parzygnat received the Bachelor of Mechanical Engineering from Villanova University, the Master of Science in engineering mechanics from the University of Denver, and the Ph.D. degree in theoretical and applied mechanics from Cornell University. He is a member of the Technical Staff at AT&T Bell Laboratories, Whippany, NJ. He joined Bell Laboratories in 1981 where he worked on the development of optical products. Prior to joining Bell Laboratories, he worked for the Xerox Corporation and the General Electric Company in the area of mechanical analysis and design. Robert A. Morgan (M’88) received the B.S degree in physics (summa cum laude) from the University of Wisconsin-Eau Claire in 1984 and the M.S. degree (1986) and the Ph.D. degree (1988) in optical sciences from the University of Arizona where he was both an SPIE Scholar and an IBM Graduate Fellow From 1988-1994 he was a member of the technical staff at AT&T Bell Laboratones conducting research in optoelectronic and lightwave devices. Since 1994 he has been a senior pnncipal research scientist at the Honeywell Technology Center, Bloomington, MN, serving as hncipal Investigator and Program Manager for Vertical Cavity SurfaceErmtting Laser (VCSEL) and optical interconnection technology. He has over I 0 0 Dublications and holds 3 Datents in vaned areas of optics including VCSEL’s, optical interconnects, semiconductor epitaxial optics, quantumwell physics and devices, self-electrooptic effect devices, Quantum-Well Infrared Photodetectors (QWIP’s), optical bistability and nonhear Optics of semiconductors, linear and nonlinear interferometry, three-wave mixing, and novel laser systems. Dr.Morgan is a member of the Optical Society of America and IEEE Lasers and Electro-optics Society. Theodore Mullally received the B.E. degree from Stevens Institute of Technology, Hoboken, NJ, in 1988. Upon joining AT&T Bell Laboratories in 1991, he was involved in research and development of Surface Emitting Laser (SEL) technologies. Since 1991, he has been responsible for design and implementation of semi-automated wafer characterization of SEL arrays. R. E. Leihenguth, photograph and biography not available at the time of publication. . . JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 13, NO. 6, JUNE 1995 1014 Gregory D. Guth received the B.A. degree in communications and the B.S. degree in chemistry in 1984 from Southern Illinois University at Carbondale. From 1984-1989 he worked at AT&T Bell Laboratories, Murray Hill, NJ, where he was involved in process development of long wavelength integrated optoelectronic devices. Since 1989 he has been at AT&T Bell Laboratories, Solid State Technology Center, Breinigsville, PA, where he is currently involved in the process development of advanced photonic devices. Marlin W. Focht (S’70-M’73) received the B.S. degree in electrical engineering in 1973 from N.J.I.T., Newark. He first joined AT&T Bell Laboratories in 1967 and worked in material characterization and LED developmpt. From 1970-1980 he worked for Research Devices, Inc. as V.P. of production, infrared microscope development and forensic marketing. Since rejoining AT&T Bell Laboratories in 1980, he has been working on laser ’ and optoelectronic processing and development at Murray Hill and more recently at the Solid State Technology Center, Breinigsville, PA. Kenneth G. Glogovsky received the B.S. degree from Manhattan College in 1977 and the Ph.D. degree in physical chemistry from Brown University in 1982. In that year he joined the Research Division of the Raytheon Company where he worked on numerous thin film deposition technologies for GaAs devices. In 1986 he joined Microwave Semiconductor Corporation where he was involved in plasma process development for GaAs devices. In 1988 he joined AT&T Bell Laboratories, Reading, PA, where he led a team to develop aluminum metalization and etch for application in high density GaAs digital circuits. He is currently doing process development for quantum well optoelectronic devices at AT&T Bell Laboratories, Breinigsville, PA. John L. Zilko was born on July 13, 1950 in Bristol, CT.He received the B.S. degree in physics from Union College, Schenectady, NY,and the Ph.D. degree in material science from the University of Illinois, Champaign-Urbana, where he performed research on the growth of multilayer, alloy, and metastable semiconductor thin films by sputter deposition. He is a Technical Manager in the Lightwave Device Research Department of AT&T Bell Laboratories where he is responsible for the development of material growth and characterization for semiconductor lasers. In 1979, he joined AT&T Bell laboratories, Murray Hill, NJ, where he was involved in the development of liquid phase and metal-organic epitaxial process for GaAsand InP-based lasers, FET’s, and other optoelectronic components. As part of this effort, he worked on metal-organic vapor phase epitaxy (MOVPE) for the growth of Fe-doped semi-insulating InP and applications of this technology for lightwave and electronic devices, especially high-speed laser diodes. In 1989, he transferred to AT&T Bell Laboratories, Breinigsville, PA, where he has continued to perform research and development of MOVPE, other fabrication technologies, and optoelectronic packaging for advanced quantum well optoelectronic discrete and array devices. Qs present research interests ‘ lie in the area of high temperature lasers. John V. Gatas received the Ph.D. degree in solid state physics in 1976 from the University of California, Davis, on low temperature NMR. He conducted postdoctoral research at the University of Illinois at Urbanazhampaign in low temperature x-ray studies of solid hydrogen and helium. Since joining AT&T Bell Laboratories in 1979, he has worked on Magnetic Bubble Memories, Josephson Junctions and lightwave devices. He has been the supervisor of groups performing research on surface and edgeemitting LED’s, photodetectors, PIN and LED arrays, laser packaging, and LiNbOs modulators and switches. He is currently the supervisor of the Integrated Photonic Research group responsible for design and processing of Silicon Optical Bench technology for passive waveguide and micromachining of silicon for optical component packaging. Philip J. Anthony (M%-SM’92) received the physics B.S. (1974) and Ph.D. (1978) from the University of Dayton and the University of Illinois-Urbana, respectively. As a member of the technical staff in AT&T Bell Laboratories, he worked on epitaxial laser material growth, laser reliability, and managing a design team for the 1.7 Gb/s laser for the FT Series G long haul fiber optic transmission product. In 1987, he formed a new department in exploratory Dhotonic devices at a new Bell Laboratories, Solid State Technology Center i n Breinigsville, PA, that developed FET-SEEDS for photonic switching, QWIP’s for infrared sensing, and VCSEL’s for optical interconnection. Since 1992, the mission of his research department in Murray Hill, NJ, is to launch and nurture new optoelectronic products until they grow large enough to attract business unit interest. The areas of interest are passive optical circuits (wavelength routers, power splitters), high-speed LiNbO3 modulators, polarization-independent switches, and silicon optical bench approaches to optoelectronic packages (an example is the present ARPA-sponsored OETC project). Booker H. Tyrone, Jr. (M’834’85-M’86-S’88M’91) received the B.S., M.S., and Ph.D. degrees in electrical engineering from the University of Texas, Austin, in 1980, 1986, and 1991 with his dissertation on common-mode compensation of fiber optic interferometric sensors. Prior to joining Martin Marietta LaboratoriesLockheed (formerly GE Aerospace) in November 1992, he maintained and instructed the maintenance of U.S. Air Force (USAF) ground-based radar systems, analyzed the data and managed USAF programs for inertial guidance and navigation systems, and taught as an instructor and assistant professor in the Electrical Engineering department at the USAF Academy. At Martin Marietta, he has worked as a senior development engineer to integrate the OETC digital fiber optic data bus. Timothy J. Ireland was born in Batavia, NY, in 1963. He received the A.A.S. degree in computer graphics from the State University of New York (SUNY) Technical College, Alfred, NY, in 1984, and the B.S. degree in computer science from the SUNY College of Technology, Utica, NY,in 1990. He joined GE Electronics Laboratory, Syracuse, NY (now Martin-Marietta Laboratories-Syracuse, in 1984 and was responsible for custom physical layout of CMOS and GaAs integrated circuits. Since 1988 he has been involved in developing and implementing CAD software for integrated circuit design, while providing digitaUanalog IC design and layout support. WONG et al.: TECHNOLOGY DEVELOPMENT OF A HIGH-DENSITY 32-CHANNEL 16 Gbls OPTICAL DATA LINK Donald H. Lewis, Jr. (S’82-A’82-A’91) received the AASEE degree from the State University of New York at Alfred, and the B.S. degree in electrical engineering from the SUNY at Utica. He is currently a Design Engineer with Martin Marietta Electronics Laboratories-Syracuse, NY. He is responsible for design and test of digital GaAs circuits for data communication and encryption applications. He maintains a digital GaAs logic design environment, including logic simulator and schematic capture libraries for TriQuint QLSI design tools. After joining the Electronics Laboratory in 1985, he was responsible for physical layout and design verification of 30+ GaAs circuit designs and is the principal author of the TriQuint Semiconductor QED/A Design Rule Verification Ruleset. David F. Smith received the B.S. degree in computer engineering from Rochester Institute of Technology in 1986. Upon graduation, he joined the GE Electronics Laboratory (now Martin Marietta LaboratoriesSyracuse where he developed and maintained a set of computer aided design tools primarily associated with integrated circuit design. Currently, as a design engineer with the Integrated Electronics Group at Martin Marietta Laboratories-Syracuse, he is primarily focused on the design and characterization of multichip modules. Sal F. Nati received the B S. and M S degrees in electrical engineering from Syracuse University He is currently a senior engineer at IMRA America, Ann Arbor, MI, where he is involved in the design of high-speed digital circuits. Prior to joining IMRA in August of 1993, he was with Martin Manetta Laboratones-Syracuse, NY (formerly, GE Electronics Laboratory. He was responsible for the development of digital GaAs technology and its application within GE Aerospace. Since 1983, he has participated in the design and development of GaAs circuits for a variety of high-speed data communication, encryption, signal processing, and microwave control applications. Before joining the Electronics Laboratory in 1980, he was with the GE Corporate Research and Development Center, Schenectady, NY. His work was directed toward the applications of microcomputer hardware and software for GE commercial products. He IS an adjunct member of the faculty at Syracuse Univerwy, where he has taught a graduate level course on digital GaAs circuit design. Mr Nati served as a member of the technical program committee of the IEEE GaAs Integrated Circuit Symposium for three years He has several publications and holds eight U S. patents. David K. Lewis (M’88) received the BSc. (Hons) degree in electronic engineering from Leicester Polytechnic (UK) in 1977, and the M.B.A. degree from Syracuse University in 1988. He was manager, Integrated Optoelectronics Programs at Martin Marietta Laboratories-Syracuse, NY, a role which include management of the OETC program. Prior to joining GE’s Electronics Laboratory in 1983, he developed digital telephony components at Mullard Applications Laboratory (UK) and designed custom silicon IC’s for Smiths Industries Ltd. At GE he designed and was project manager for several silicon VLSI circuits including in 1986, a 40 MHz clock rate 32-b CMOS RISC microprocessor. In 1988 he became Manager of the Electronics Integration Group carrying out the development of complex multitechnology -_ modules including silicon and GaAs analog and digital circuits as well as optoelectronic components. This position evolved into his role in 1992 when he became responsible for optoelectronics programs, and has continued with the merger of GE Aerospace into Martin Marietta in 1993. Mr. Lewis is a member of IEEE-LEOS and was Syracuse Chapter Chairman of IEEE-CAS in 199C1991. 1015 Dennis L. Rogers (M’83) received the M.S. and Ph.D. degrees in theoretical physics from University of California, Davis, in 1970 and 1977. He joined IBM’s Research staff at the Thomas J. Watson Research Center, Yorktown Heights, NY, in 1977. There he has worked on the development of high-speed, integrable, MSM detectors and monolithic, optical receiver circuits. Dr. Rogers has received IBM Outstanding Innovation Awards for work on optical receiver IC’s in 1981 and 1988 and an Outstanding Technical Achievement Award in 1991. Herschel A. Ainspan received the B.S. and M.S. degrees in electrical engineering from Columbia University, New York, NY, in 1989 and 1991, respectively. In 1989 he joined the IBM Thomas J. Watson Research Center, Yorktown Heights, NY, where he has been involved in design and test of IC’s for highspeed optical and infrared links for computer applications, using both Si and GaAs technologies. Sudhir M. Gowda (S‘88-M’91-S’92-M’92) was born in India in 1965. He received the B.Tech. degree in electrical engineering in 1987 from the Indian Institute of Technology, Madras, India. He received the M.S. and Ph.D. degrees in electrical engineering in 1989 and 1992, respectively, from the University of Southern California, Los Angeles. From 1987-1988 he was a Teaching Assistant in the Electrical Engineering Department at the University of Southern California. From 1988-1989 he was a Research Assistant at the MOSIS Service of USCnnformation Sciences Institute, Marina del Rey, CA. From 1989-1992 he worked as a Research Assistant at USC. In 199C1991 he was actively involved in the development and teaching of a senior-level class on VLSI design. He also helped to manage the VLSI Signal Processing Laboratory at USC. He contributed significantly to the development of the BSIKplus submicron MOS model for analog and digital VLSI circuits. In December 1992, he joined the VLSI Design and Communication Technology Department at the IBM Thomas J. Watson Research Center, Yorktown Heights, NY, to work on high-speed computing and computer communications. He has published over twenty technical papers in scientific journals and conferences. His research interests include architectural design and detailed circuit implementation for high-speed computers, signal processors, and communication circuits. Steven G . Walker was born in Minneapolis, MN, in 1961. He received the B.S. degree in applied physics from Cornell University. His past work experience includes atmospheric modeling, GaAs device parametric measurement techniques, and automation. He is currently an engineer at the IBM T. J. Watson Research Center and is working on highspeed on-wafer optoelectronic test and measurement of O/E receivers and photodetectors. Young H. Kwark (S’8GM’82-S’82-M’84) received his degrees in electrical engineering from MIT and Stanford Universities. His past work experience has included high efficiency photovoltaics, microwave characterization of MESFET’s, and design of receivers for fiber optic data links. Currently, he is involved in application of digital signal processing techniques to enhance wireless propagation. Richard J. S. Bates, (M’81SM’85) photograph and biography not available at the time of publication. 1016 Daniel M. Kuchta (SM”LM’92) was born in San Francisco, CA, on April 23, 1964. He received the B.S., M.S., and Ph.D. degrees in electrical engineering and computer science from the University of California, Berkeley, in 1986, 1988, and 1992, respectively. He is currently a research staff member for IBM at the T. J. Watson Research Center. His current research interests include high-speed modulation of VCSEL’s and low cost components for fiber optic data communications. View publication stats JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 13, NO. 6, JUNE 1995 John D. Crow (S’68-M’78-SM’90-F‘93) received the B.S., M.S., and Ph.D. degrees in electrical engineering from the University of California, Berkeley, in 1966, 1968, and 1972, respectively. He joined IBM’s Research staff at the Thomas J. Watson Research Center in 1974, after two years of work on high power propagation in optical fibers and photochromic planar lightguides at the Coming Glass Works Research Center. He developed a flip-chip laser array package for applications to liquid crystal projection displays, and fiber optic links. In 1979, he became manager of the Fiber Optic Technology group, doing the initial research on the IBM 3044 channel extender, as well as a prototype 200 Mb/s link leading to the S/390 ESCON serial mainframe U 0 channel. In 1985, he managed a group developing multichannel OEIC chip and packaging technologies. Since 1990, he has been involved in multiprocessor optical network design and prototyping. In 1992, he became program manager for joint projects between IBM Research’s Communications Technology Department and other companies, the government, and universities. Dr.Crow has received IBM Research Division awards for work on injection laser arrays for liquid crystal displays in 1979, for high-speed fiber optic link development in 1983, and for Opto-electronic IC chip development in 1989. In 1986, he was on the technical program committee of the IEEE Optical Fiber Communication Conference. In 1987, he was appointed to the Photonics Panel of the U.S. Government’s National Research Council. In 1989, 1990, 1991, he has been on the program committee of the IEEE LEOS Annual Meeting; and in 1989, 1990 on the IEEE Electronics Components and Technology Conference program committee. In 1992 he co-chaired the IEEEVLEOS Topical meeting on Integrated Optoelectronics, and is currently editing and contributing to a book on this subject. In 1990, he participated in the formation of the OptoElectronics Industry Development Association, and is currently completing a report on a technology roadmap for the computer industry.