whit@milton.acs.washington.edu (John Whitmore) (01/09/90)
According to Electronic Engineering Times (25 Dec '89), a group funded by Japan's MITI has built a four-chip computer in a totally non-semiconductor electronic technology, i.e. superconducting Josephson junction switching, with niobium devices. The speed/power product for this technology (previously explored in lead superconductors by IBM) is very good; the Japanese group claims 6.2 milliwatts for a 1 MFLOP machine. RAM speed is about 500 ps (that's PICOSECONDS, meaning thousandths of nanoseconds), and ROM speed 350 ps. The total instruction set is 128 of 'em, and by some standards, this is a RISC machine. The other two chips were the ALU/latch and the program sequencer (it's Harvard architecture, by the way; 10 bit instructions and four bit data). The "1 MFLOP" figure is probably for 32-bit floating point words, but I can't be sure from the article. Two interesting notes: the four-chip set is from a research group with only five people; the announcement coincides with their request for continued research funding (I think they'll get it). One use immediately springs to mind; space applications are power-stingy, and require hardness against radiation (solar protons as well as cosmic rays) which is difficult to achieve in silicon; superconductors should function well in space, and they'd be easier by far to cool (with a well built parasol, any old rock in space will achieve 3.2 Kelvin temperatures, well under the superconducting threshold temperature). Maybe the next outer-planet probes will carry a descendant of this machine. Of course, the computer only works in liquid helium, so a significant barrier exists to connecting it with the outside world. Also, the logic depends on magnetic field sensing; a tenth of a percent or so of Earth's magnetic field leaking into the area of the chip could erase EVERYTHING, so some state of the art magnetic shields are a necessity (not difficult to build, though, given superconductors to build 'em from). The group's spokesman, Susumu Takada, says the junctions are simpler to build than transistors; they haven't tried making anything smaller than 3 microns yet, and the group hopes to build much faster devices (to get reliable operation without crosstalk, the picosecond speed of the ALU's components has been deliberately slowed). The future looks promising for this technology. I am known for my brilliance, John Whitmore by those who do not know me well.
henry@utzoo.uucp (Henry Spencer) (01/10/90)
In article <1309@milton.acs.washington.edu> whit@milton.acs.washington.edu (John Whitmore) writes: >superconductors should function well in space, and they'd be easier >by far to cool (with a well built parasol, any old rock in space >will achieve 3.2 Kelvin temperatures, well under the superconducting >threshold temperature)... Unfortunately, not so. That parasol is not magic; if it keeps the sunlight off, it will in turn get warm and start to radiate heat into your computer. A well-built parasol in near-Earth space will give you liquid-nitrogen temperatures, 70K or thereabouts, but you'll have to be a lot further out from the Sun to get liquid-helium temperatures. Satellites that need liquid-helium temperatures for sensors, like the current Cosmic Background Explorer, carry large tanks of liquid helium and plenty of insulation... and when the tanks boil dry, that's the end of the sensors' useful life. -- 1972: Saturn V #15 flight-ready| Henry Spencer at U of Toronto Zoology 1990: birds nesting in engines | uunet!attcan!utzoo!henry henry@zoo.toronto.edu
mark@mips.COM (Mark G. Johnson) (01/10/90)
In article <1309@milton.acs.washington.edu> whit@milton.acs.washington.edu (John Whitmore) writes: > > ... about an EE TIMES article covering a Japanese 4-chip computer > built of Josephson Junctions .... > >The speed/power product for this technology is very good; the >Japanese group claims 6.2 milliwatts for a 1 MFLOP machine. > > ...it's Harvard architecture, by the way; 10 bit instructions >and four bit data ... Contained within the 4 chips are a total of 26,000 Josephson junctions, which implement approximately 8,500 logic gates (2100 gates/chip). This is why it's a 4-bit computer - they didn't have very high LSI density. Power is supplied to the devices in AC form; it's two out-of-phase sine waves at 1.02 Gigahertz, with peak-to-peak amplitudes of 100 millivolts. To supply 6.2 milliwatts of power (RMS) to the chips, these sine waves must deliver 175 milliamps, 88mA apiece. Pity the poor devil who has to provide a scaled-up computer (say, a 16-bit machine) with a pair of 3 GHz sinewaves deivering a quarter of an amp :-(. The logic levels in this JJ technology are 0.0025 volts apart. Works fine on-chip, but it may present difficulties communicating off-chip to, say, ECL caches. You gotta have a voltage gain of about 200x to get the amplitude up to 500mV for ECL. This may prove difficult to do quickly; consider the necessary gain-bandwidth product. -- -- Mark Johnson MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086 (408) 991-0208 mark@mips.com {or ...!decwrl!mips!mark}