There is light at the end of the tunnel. This is the consensus among Scottish physicists that are working on optical computing. Edinburgh’s Heriot-Watt University under Professor Des Smith is leading the Scottish Collaborative Initiative in Optoelectronic Sciences, SCIOS, which focuses on the construction and operation of prototype digital optical circuitry. SCIOS comprises Edinburgh, St Andrews, Glasgow and Strathclyde universities, alongside Heriot-Watt. The research has been achieved with grants of UKP1m, to be shared among SCIOS members, UKP800,000 for Heriot-Watt from the Science & Engineering Research Council and a $1m grant from Boeing Co. Smith describes the philosophy of the research in terms of the advantages to be gained in processing power by the combination of the device parallelism and the interconnect freedom that optics offer. All involved in this research are at pains to stress that digital all-optical machines are a long way away. The reason that light offers hope is because light beams are not corrupted by interference.
Hybrid machines
Electrical signals interfere with each other. Light beams are infinitely more flexible because light beam can cross light beam and they each continue uninterrupted. Smith explains that the future lies in hybrid machines that exploit both silicon and light. The marriage of optics and electronics is not unique. During the 1960s and 1970s the search was on for a usable optical transistor or optical memory element. Those studied needed KiloWatts of power that posed no threat to the typically milliWatt electronic component. In 1979 a breakthrough occurred with the near simultaneous discovery at Bell Laboratories in New Jersey and Heriot-Watt that low power optical switches could operate in small semiconductor devices. These were termed optically bistable devices as they have two stable outputs for a certain range of inputs. They are now able to operate at microWatt and lower power levels. The Heriot-Watt research is important in that it centres on the practical as well as the theoretical in the construction of devices to illustrate the possibilities of optics in computing. One of the devices on view is the first programmable digital optical processor, dubbed the Optical Cellular Logic Image Processor, or OCLIP for short.
By Noni Stacey
Research physicist Dr Douglas McKnight explained that both a single channel processor and a parallel version incorporating 256 information channels had been constructed. The OCLIP is capable of number recognition, full-addition and subtraction; 16 by 16 parallelism of all components has been demonstrated. It contains five functional modules: an input module, which converts a serial binary signal into a two-dimensional binary optical signal, a spatial light modulator; a two input processing unit that performs some form of combinatorial logic on the inputs; a thresholding unit consisting of an optical interconnect and a variable threshold logic plane; a data synchronisation unit that acts as a temporary memory, and an output stage that converts the two-dimensional binary optical image back into a serial electronic image. All of these modules are controllable from an electronic source. It represents an optical co-processor residing within a host electronic computer that provides the required control and clock signals. Post-graduate researcher Suzanne Wakelin has been working on a hybrid device combining optics with electronics. This attempts to simplify interface difficulties by encapsulating interface and processing in a single device known as a Self-electro-optic Effect Device, SEED. SEEDs cost $10,000 each and provide a 64 by 64 array. They take a week to construct and given the ability to mount the devices properly – commercial mounts are in place at the moment – it will mark another breakthrough. This type of device will be of greatest use in the processing of images, either real or representing numerical data. Professor Smith is not surprised that the computer industry is sceptical about the possibilities of optics in computing. Exist
ing Silicon people are naturally sceptical about a new technology, given the amount of investment in Silicon. Smith lists the advances made by the department since 1989. These include: a plausible, practical architecture; a fan-in and fan-out capability, three port capability; a data rate of 10 to the power of nine bits per second; the development of a digital optical circuit that is compatible with laser sources. On the question of whether optical logic is relatable to a plausible, practical architecture, Heriot-Watt devised lock and clock circulation, which synchronises the flow of optical data – optical cellular logic image processor architecture was formulated between 1988 and 1990. Fan-out and fan-in has been demonstrated and three port capability is available via bistable etalon with absorbed transmission, which isolates input and switch signals, and with two-element standardised input. As for the data rate of 10 to the power of nine bits per second, about two times 10 to the power of six is currently possible using SEEDS.
Breakthrough
The final requirement is that it should be realisable so that digital optical circuit development can proceed that is compatible with laser sources. Smith points to current work to develop switching devices that operate at the wavelength of a suitable laser and vice-versa. He expects to see a breakthrough in this area within two years. The freedom from interference offered by optics is already used in several ways. Sequential optical interconnects are in machines at the board-to-board level, and chip-to-chip fibre connects are in development. Optical fibres are used in the telecommunications area and will become increasingly complex. Interconnection capability relies on holograms and the massive interconnection bandwidth of optics uses holograms to re-route the optical signals. Optics used to access data from optical disks read in parallel offer up to 10,000 channels of information. Neural networks, currently using electronics, feature computers built using the human brain as a template. They use a network of large numbers of simple processor elements together – neurons – that offer thousands of interconnections. These areas will benefit from the use of optics in the future. It is not a case of one technology replacing another, more of realising the potential of optics in relation to Silicon.
