Physics has come to the rescue of supercomputer researchers yet again. The workers at the hairy end of computer design have just about squeezed the last drop out of existing technologies to get the present top performance of around a billion sums a second from their machines. They have tried linking lots of CPUs together – up to a quarter of a million – with lots of different architectures – the binary n-cube, loosely and tightly coupled, the hypercube, crossbar switching, radial connections, multiple buses, shared memory, multi-stage networks, but they have tended to lose on the swings most of what they gained on the roundabouts
A distressing number of programs and real-world problems turn out to be very difficult or impossible to divide up so that the separate parts can be run on different processors. Other exotic architectures such as neural networks that try to mimic the way a brain works, dataflow machines, applicative language engines, or graph reduction, have yet to demonstrate their full potential. However, most observers believe that at best they will only show dramatic improvements for a limited range of applications or, like RISC technology, show a useful but hardly earthshaking incremental increase in performance and flexibility. So the desperate designers, frantically searching for orders of magnitude improvements rather than incremental ones, have seized on the recent, much hyped, advances in superconducting technology as a heaven-sent answer to their problems. (After all, when valve computers grew to the size of small factories and used as much electricity as a small town, the physicists obligingly invented the transistor. And the integrated circuit. Then VLSI. As every circuit engineer’s granny used to say: Necessity is the mother of invention. Necessity is just cutting up a bit of fine time.) Superconducting materials allow electricity to pass with no resistance – a current in a loop of wire will flow around it forever. It can be used to make very powerful magnets, transmit electrical power without losses and, in the field of electronics, supersensitive detectors called Squids, microwave signal processing, and all sorts of good things. Unfortunately, up until now superconductors only worked at a few degrees above Absolute Zero – minus 273 degrees Centigrade – and had to be expensively cooled with liquid Helium. Now a range of materials have been discovered that superconduct at higher and higher temperatures. IBM researchers in Zurich started the bandwaggon rolling by reporting superconduction at a record 28oK (Kelvin – degress above absolute zero) but this was quickly beaten by researchers in Japan, the US, China and Europe. The record for the highest temperature was broken almost weekly as researchers tested the new ceramics made from compounds of Barium, Strontium, Calcium, Oxygen and Copper, progress has been so startling that some researchers believe that a room temperature superconductor is just around the corner. The developments have been hyped as the most important discovery of the century and were even reported in the daily newspapers – albeit with a level of accuracy and imagination usually reserved for political reporting. The computer designers are excited because of a British invention, Brian Josephson’s Josephson Junction device, which switches thousands of times faster than the fastest existing device used by chipmakers, and uses around five thousand times less power. Unfortunately the Junctions use superconducting technology, which caused insurmountable problems at a few degrees above absolute zero – IBM spent many years and hundreds of millions of dollars trying to perfect the technology before abandoning it in 1983. IBM’s faint heart brought research outside Japan almost to a standstill. However, small research projects kept going in Europe, the Japanese kept working at it and the Pentagon has recently stepped up funding in the area – partly out of fear of a Japanese breakthrough that would give them a big lead in supercomputers and partly to try and develo
p Josephson Junction-based signal-processing technology for space- based microwave systems and imaging radars for the Strategic Defense Initiative. When IBM shut down its research effort, there was nowhere in the US that had the clean room facilities to fabricate Josephson devices. Now, thanks to the Pentagon prodding and funding, Westinghouse, TRW and Hypres all have the capability, but they seem to be concentrating most of their efforts on military projects. In Japan there are a number of development projects working with Josephson technology as part of the government supercomputer project, plus a large effort by Nippon Telegraph and Telephone Corp, which refuse to discuss it. NEC is working on memory while Fujitsu and Hitachi are working on logic – the latter has developed a 4-by-4-bit multiplier and a 3,264 gate array.
Heath Robinson cooling
The work at the goverment’s Electrotechnical Laboratory is even more ambitious: 2-bit slice arithmetic-logic unit building blocks, a multiplier, latches, registers and program counters – all the building blocks needed for a complete computer built with Josephson technology. As these all work at a few degrees above absolute zero, does discovery of the new materials that work at much higher temperatures mean that all the effort has been wasted? Or perhaps the new materials will simply increase Japan’s evident lead in the technology. But whoever wins in the end, the prize for the first commercial product on the market using Josephson Junction technology has gone to a tiny US company called Hypres. To give a taste of what we can expect in the future, Hypres put a single Josephson Junction circuit into the measurement head of a digital oscilloscope, giving it a performance that far outstrips any of its competitors. The tiny chip is cooled by a Heath Robinson device: a hosepipe that sprays it with liquid Helium from a thermos flask under the machine. But they have shown that the technology is reliable enough to be used in a commercial machine, just as new discoveries open the way for devices that work at room temperature. And when they put hundreds of thousands of them on a single chip to make a computer, it will make the present generation of supercomputers look as fast and as brainy as a dinosaur. Meanwhile, while development goes on at labs all over Japan, Hypres is still trying to find a partner to finance the development of a Josephson Junction computer.