Optoelectronic computing has moved a step nearer with the award of a US patent to three scientists at the University of California at Davis for Optimul: An Optical Interconnect for Multiprocessor Systems. According to Microbytes Daily, the scientists believe that their patent holds the solution of two of the problems that have crimped development of hybrid systems that combine electronic and optical technologies. The key to Optimul is the free space use of a laser – Modular Technology Ltd founder John McNulty pioneered communication by modulating a laser beam in free space more than a decade ago – and a thin film technology designed to eliminate the memory access bottleneck. The approach is claimed to solve memory contention problems in shared memory systems, and network bottlenecks in separate memory systems. With sufficient funding, a crude working prototype of Optimul could be built in about six months, says Steven Kowel, a materials scientist on the Davis team. Despite optimistic noises from Plessey Co and Edinburgh University here, and AT&T Bell Laboratories in the US, Kowel avers that designers that have tried to build optical computers have encountered startlingly difficult hardware problems, Kowel says, with the result that most of them remain primitive and expensive, which has led researchers to examine hybrid systems that handle logic, memory and input-output electronically and use optical fibres simply to create much faster interconnections between the Silicon elements. Researchers have also tried to replace the fibre links with a technique using holographics, but the Davis team believes that Optimul, which would coat memory chips with a thin polymer film treated for sensitivity to electrical charge and to light, producing something rather like a liquid crystal diode, is less complex than using holographics. A laser beam illuminates the coated memory chip and broadcasts a picture of the entire state of the memory contents; the broadcast can be picked up by multiple processors at once, the beam being demodulated back to electric charges and stored as such, creating a double parallelism, across processors, and across bits in memory. The idea is that in a tightly coupled or shared memory system, a wireless broadcast would solve the problem of contention for memory access, a problem, Kowel points out, that even defeated the Cray Research Inc designers of the X-MP. With loosely coupled systems where there is no central memory, a bottleneck is created by network interconnections, so that even with high bandwidth fibre optic channels, a limited number of data pins feeds memory contents to the optical channel: Optimul would solve that problem by broadcasting rather than channelling the data. Writing to memory would still have to be done via electronics, so that at least in its initial form, a data bus would remain part of the overall system.
