Commercial data link modules from three manufacturers were measured for transmitter launch power and receiver sensitivity when used with the five common graded-index multimode fiber types: 50 pm core/NAm=0.20, 50 μm! NA.=0.23, 62.5/0.275, 85/0.26, and 100/0.29. Depending on the model, transmitters launched from 2.8 to 4.4 dB more power into the 100 μm fiber than the 62.5 μm fiber, whereas receiver sensitivities decreased by up to 3.5 dB. These variations in launch power and receiver sensitivity complicate the selection of an ideal fiber and emphasize the need for specifying data link performance with the particular fiber of interest. In general, the system gain of typical transmitter/receiver pairs is less favorable to large diameter fibers than previously believed.

The development of an optical circuit that multiplexes gigabit-per-second electronic data streams into a very broad bandwidth (greater than 10 Gbit/s) optical signal would greatly enhance interprocessor communications in a parallel computer. One possibility that we are investigating is coherence multiplexing. In this method, time delays exceeding the coherence time of the light source are used to achieve multiplexing. We have analyzed the limitations of coherence multiplexing for high-speed data links. For the geometry that we are investigating, the analysis shows that the signal decreases inversely as the square of the number of channels. Noise sources include photon shot noise, excess noise (classical optical field fluctuations) and detector noise. A suitable optical source must have a smooth broad spectrum. Superluminescent diodes or edge emitting diodes are the best candidates. We show how all these factors affect the total information transfer rate. An integrated optical circuit for coherence multiplexing is being designed and built. This device is an array of four Mach-Zehnder interferometers integrated on a lithium niobate substrate. The differential delays will be achieved using proton exchange. Each interferometer will have two sets of electrodes, one for signal modulation and a second for fine adjustments of the delay.

Optical interconnects have been designed and demonstrated in the Connection Machine. The objective was to reduce the size and power dissipation of optical interconnect transmitters and receivers such that optical fibers could be used inside the processor to replace cabling. The results demonstrated a 512:1 increase in interconnect density. This paper describes the architecture, design and physical characteristics of the Connection Machine optical interconnects and the results of the demonstration.

A new initiative, which has just received National Science Foundation support, has been undertaken in collaboration with the Center for Telecommunications Research at Columbia University. This project is to design and build a multi-Gb/s optoelectronic data transport network using self-routing packets in a multi-hop network. The single electronic word packet payloads are highly compressed using optical techniques, and remain optical from source to target while traversing the switching nodes. Optical packet switching is performed with custom LiNbO3 directional coupler switches. A lean, self-routing hot potato protocol avoids the need for data storage at the switching nodes and provides a fixed node latency equivalent to a few meters of fiber. Sustainable throughput both in to and out of the electronic host at each node should exceed 1 Gb/s, with higher speed bursts. Some technical details of the optical data packet compression and decompression schemes, the hot potato switching protocol, and the wrap-around shuffle-exchange interconnection network will be given. The project timetable anticipates a lower speed, proof-of-principle four node network in three years, and a higher speed, larger, engineering demonstration in five years.

A high speed serial fiber optic switched network is described which is intended for such applications as sensor/signal processor or computer interconnects. Communication is completely asynchronous and any channel may be operated at any data rate up to 1 Gb/s. The network is based on a gallium arsenide electronic crossbar switch, a common network interface for each node, and commercial optical data links. To implement the network interface function, an ASIC chip which will contain both an encoder and decoder section has been designed and is being fabricated in gallium arsenide. The encoder will accept 8- or 16-bit parallel words at any data rate up to 1 Gb/s and encode them into a 4B5B serial bit stream. The decoder reverses this process so that the network is completely transparent to the user. Handshake protocol is via a simple DATA READY/ACKNOWLEDGE scheme and the network interface may also transmit a small number of command words.

Current design and development activities within the computer industiy are beginning to use semiconductor laser diodes as the light source for optical fiber data links. This trend is occurring because laser diodes have many advantages over LEDs as the preferred light source for both multimode and single mode fiber. As a result, laser safety has become an important issue in the design activity of fiber data link products, both because of the need to produce safe products for the marketplace and because laser safety regulations are enforced in most countries. This paper presents an overview of some of the laser safety issues and regu-lations as they apply to optical data links.

This paper report describes the results of the development of a multi-channel Optical Fiber Crossbar Switch (OFCS). The switch operates in the optical domain using GaAs semiconductor lasers to transmit wideband multiple channel optical data over fiber optic cables. A 16 X 16 crossbar OFCS switching system was developed suing GaAs optoelectronic integrated circuit receivers (OEICs) and GaAs integrated laser driver circuits output to GaAs lasers operating at 830 nm. Essentially error free performance has been obtained at a data bandwidth of 410 MBPS. The OFCS can be completely reconfigured in less than 50 nanoseconds under computer control. In addition, the system was designed with a 32 X 32 crossbar switching capability, so that by simply installing additional optical transmitters and receivers the system has the capability for 32 X 32 optical crossbar switching. The fully populated OFCS has the capability to handle 12.8 gigabits per second (GBPS) of data while switching this data over 32 channels without the loss of a single bit during switching. The integrated GaAs transmitter drivers and OEIC GaAs optical receivers with data bandwidths exceeding 2 GBPS.

Systems which permute the time-slots within a frame, called time-slot interchangers, are used in digital communications to map an incoming transmission associated with the i-th time-slot to an outgoing message for a receiver associated with the j-th time-slot. In a time-multiplexed optical computer environment, a time-slot interchanger corresponds to a multiprocessor interconnection network in a spatially parallel multiprocessor system. This work attempts to perform permutation among time-slots by deriving a time-domain version of the spatial shuffle-exchange network. In so doing, a new and a powerful structure called serial array architecture has been developed which has demonstrated very high potential for substantial hardware reduction. The shuffle permutation is the more interesting part of the stage and is the topic of this work. Detailed analysis and theory of time-slot interchange with specific reference to performing perfect shuffle has been developed. Different parameters for measurement based on time-slot shift have been presented and used in analysing the interchange problem and studying the power and limitation of a 2x2 photonic directional coupler. Three important classes of architectures using directional couplers have been presented for performing perfect shuffle on odd and even-sized frames of different lengths. Finally the basic relationship between ternary base and the 2x2 switch used in the feedback mode has been brought out and used to develop minimum switch architectures for perfect shuffle.

Free space optical interconnections offer some beneficial solutions to microelectronic problems, e.g., pin-ins, pin-outs, testing, and wafer scale integrations. When applied to an array of multiprocessors for parallel computing, free space interconnects can offer 3-D interconnection topology, 2-D parallel data inputs and outputs and shared memory architectures. Furthermore, if the optical interconnections are programmable, we shall be able to incorporate several computational architectures efficient for implementing several computational algorithms in one machine. In this review, many of these potential benefits of utilizing free space optical interconnections will be described in detail and various research topics currently under investigation at UCSD will be discussed.

We describe a crossbar switch architecture based on acousto-optic beam deflection. The architecture has O(N) hardware complexity throughout, while exhibiting minimal insertion loss, low crosstalk, and fast reconfiguration. Because of the small amount of signal degradation imposed by this switch, it is suitable for nonregenerative applications within fiber-optic networks. By increasing hardware complexity, broadcasting can also be achieved within the framework of this architecture. We report experimental performance of a 1x4 switching element. Insertion loss ranges from 2 - 6 dB, worst-case signal-to-crosstalk ratios are in excess of 30 dB, and reconfiguration times are on the order of one microsecond.

The interconnection requirements of fine-grained computing are examined and compared to the requirements of coarser grained, multiplexed systems. Specifications for the interconnection medium are developed and compared to the performance of available optical source and interconnection components. The use of polyimide waveguides for both applications is considered and the probable architecture of a multiboard fine-grained system is described.

In this paper, a novel backplane interconnect technology based on multiplexed waveguide holograms in low-loss planar waveguides is proposed. Two different types of waveguide holograms are employed in this novel interconnect technology: holographic waveguide couplers serve the coupling functions between optical planar waveguides and laser sources/detectors; and planar waveguide holograms are used to distribute signals in various directions. Both holograms can realize fan-out and/or fan-in operations. In particular, long interaction lengths possible in planar waveguide holograms can result in high angular selectivity.

The device requirements and system design issues for board-level optical links differ from their counterparts in optical links used for longer distance communications. We have fabricated a research prototype of a board-level optical link which is designed to operate in a high-speed digital equipment internal interconnect environment. The experimental transmitter and receiver packages are specifically designed for subminiaturized applications. We have explored the use of photolithographically patterned waveguides made of polymers as an alternative to the use of fiber at the board level. The major aspects of the device design are reviewed, but the emphasis of this paper is the systems-level design issues. We will discuss the design goals and the performance of the link. The results of this work demonstrate the feasibility of optics for signal distribution at the board level.

Continued scaling of silicon CMOS technologies to smaller feature sizes will provide impressive opportunities for integration of very compex processing systems on single ICs and single wafers. This suggests a considerable scaling of "large-scale" systems such as distributed computing environments to much smaller, more highly monolithic real-izations. Such "scaled systems" will require a corresponding scaling of communications networks, perhaps reduced to single wafer sized networks and switching circuits. Presently, electronic interconnections characterize monolithic circuit interconnections while optical communications is making significant inroads for efficient realization of system communications. In the future "scaled" world of ULSI systems, it remains unclear whether optical interconnection techniques increasingly favored for large-scale system communications will remain favored. However, the system organizations and operating protocols are likely to change significantly as present large-scale systems are scaled. It is suggested here that a principle objective will be to develop networks with disposable bandwidth, well in excess of the actual bandwidth required. Under such conditions several of the control issues limiting efficient use of distributed systems are greatly relaxed. Optical interconnects, integrated for example on silicon wafer substrates, provide the high bandwidth sought. However, resynchronization delays and queues will be very difficult in the optical domain. It is suggested here that the overall network function can be divided into relatively slow electronic control of the network paths and high bandwidth (optical or electrical) links and switches, providing a high performance, hybrid network realization.

In order to exploit the potential of wafer-level systems for providing compact distributed computing systems, mapping of a very high perfromance communication network into the scaled multi-wafer environment will be essential. Both free-space and planar integrated optics can play an important role in establishing this interconnection medium, potentially providing an environment of homogenious connectivity from wafer to wafer. For planar optical interconnections, the connectivity that can be provided is largely a function of the routing constraints imposed by the medium itself and the system configuration, noise and power budget. The set of trade-offs are significantly different than those both for long distance lightwave systems and for circuit board electrical interconnections. In this paper, some basic issues related to routing of planar optical interconnections in manhattan-like routing configurations are reviewed with emphasis on the layout of large arrays of waveguides for parallel signal transmission. Present experimental work to help clarify constraints on optical interconnection layout using dense arrays of polysilyne thin film optical waveguides is described.

We give a brief description of a freespace beam steering crossbar switch design using acousto-optic deflectors, and report the results of an experiment to accurately measure beam deflection time for the beam width required to support our design. A beam diameter of lmm to allow reasonable separation of 32 potential targets along an axis was steerable in 130 ns to 569 ns, depending on beam proximity to the piezoelectric transducer. All experimental components and instru-mentation used are commercially available.

A novel photonic interconnect architecture is proposed which can provide extremely high dimensionality. The proposed architecture, which resembles a collapsed network, avoids the difficulties associated with the use of optical crosspoints. This is accomplished by providing a dedicated path for all input- and output-port connections on a common transmission medium. This eliminates the restrictions imposed by 2x2 switching elements in classical space-division switching architectures.

Controlling system clock skew at the desired system cycle time is the major challenge in the design of a clock distribution network. Optical techniques offer a potential alternative to conventional electrical means for distributing clock signals. For optical clock distribution, the limiting factors are the level of optical power available to the photodetectors and the timing uncertainty (skew) in optical receivers. This paper presents an analysis for receiver skew in terms of the RMS timing jitter, static skew and photodetector-induced skew. Based on this analysis, an optical fanout model is developed relating system skew to the received optical power for various Si and GaAs receiver configurations. This model predicts that, for high optical fanout levels (in the order of 100), system clock skew in the range of 100 to 200ps is achievable using avalanche photo-diode receivers. Skew measured for an experimental system was 120 ps, correlating well with the analytical model.

Optical computer interconnects appear very attractive when integration of state of the art technology of quantum well GaAs/GaA1As lasers is considered. These ultralow threshold lasers provide the very high transmission rates and the inherent simplicity required for such systems. A detailed design is presented for a 5 Gbit s-1 transmission rate, suppression of pattern effects, and a system power supply of approximately 25 mW per laser. Existing experimental data show that little extrapolation is required to reach that kind of performance from state of the art technology.

Capability for the lowest cost is the goal of contemporary communications managers. With all of the competitive pressures that modern businesses are experiencing these days, communications needs must be met with the most information carrying capacity for the lowest cost. Optical fiber communication systems meet these requirements while providing reliability, system integrity, and potential future upgradability. Consequently, optical fiber is finding numerous applications in addition to its traditional telephony plant. Fiber based systems are meeting these requirements in building networks and computer interconnects at a lower cost than copper based systems. A fiber type being chosen by industry to meet these needs in standard systems such as FDDI, is multimode fiber. Multimode fiber systems offer cost advantages over single-mode fiber through lower fiber connection costs. Also, system designers can gain savings by using low cost, high reliability, wide spectral width sources such as LEDs instead of lasers and by operating at higher bit rates than used for multimode systems in the past. However, in order to maximize the cost savings while ensuring the system will operate as intended, the chromatic dispersion of the fiber must be taken into account. This paper explains how to do that and shows how to calculate multimode chromatic dispersion for each of the standard fiber sizes (50 μm, 62.5 μm, 85 μm, and 100μm core diameter).

Integrated optical matrix switches can process optical signals whose data rate exceeds 100 Gbit/s, which might be very useful in a multi-computer environment. This paper describes a hybrid design which uses combinations of the conventional switch and a new 2-stage switch for crosstalk suppression in the integrated optical rectangular crossbar matrix switches. Using the hybrid design, we believe that low-crosstalk integrated optical crossbar switches can be fabricated with much relief to the stringent process control.

Magneto-optic garnets are efficient Faraday rotators at 1300nm and are thus of interest as the basis for a non-mechanical bypass switch. Recent improvements in bismuth-substituted rare earth iron garnet (BRIG) technology have removed limitations which made garnets less than ideal for use in multimode fiber systems such as LAN's. We report recent progress in our laboratory to increase Faraday rotation, lower optical absorption, and to decrease magnetic drive field in BRIG garnets. These improvements allow a simple implementation of an optical bypass switch suitable for multimode use. We report test results from a switch which incorporates features which reduce parts count and simplify magnet design. The results include insertion loss of 2.47dB and crosstalk of -23dB. The switch uses a single garnet and is amenable to a single beamsplitter implementation.