Out of the Lab: Really Cool Chips
Niobium is a refractory metal that can be patterned in much the same way as silicon. When it is cooled down to a few degrees above absolute zero (-273.15 degrees Celsius) -- so cold that atoms almost stop moving altogether -- it becomes superconducting, which means that the electrical resistance of the metal virtually disappears. With nothing to slow them down, electrons can whizz round at close to the speed of light, performing digital functions with unprecedented accuracy.
It is generally accepted in the industry that after the next step up to 40-Gbit/s optical systems, electronics will run out of steam and be superseded by all-optical signal processing. But superconducting electronics could -- if it becomes a commercial reality -- compete with all-optical systems.
"The key to trigger the availability of products is financing," says Darren Brock, director of advanced development at Hypres. With appropriate financing, Brock feels, superconducting ICs could be available in as little as two to three years.
Brock is lead author of an article in the December issue of IEEE Spectrum, which reviews the start-of-the-art in superconducting electronics and describes how the barriers to commercialization are gradually being broken down.
Hypres already sells devices in small quantities to non-telecom manufacturers. It offers simple circuits operating at over 100 GHz as well as more complex circuits that do digital signal processing at 25 GHz.
But this just scratches the surface of what could be achieved. Right now Hypres can only squeeze 10,000 to 20,000 Josephson junctions -- the superconducting equivalent of a transistor -- on to a chip. That's because the side of each element measures 3 micrometers, which is about ten times the state-of-the-art in silicon electronics. It should be straightforward, however, to adapt silicon process technology to make smaller Josephson junctions, and this would result in a hike in speed. Complex circuits that now run at 25 GHz could step up to 80 to 100 GHz, while simple circuits that now do 100 GHz could step up to 700 GHz, claims Brock. "We have already verified this in the lab," he claims.
Not surprisingly, the need to house chips inside a freezer was a big impediment to getting these things out of the lab in the first place. It was one of the key reasons why IBM Corp. (NYSE: IBM) wound up its superconductivity program in 1983. Hypres was formed by a group of ex-IBM believers who thought there was still some mileage in the technology.
Since then, there have been big advances in so-called cryogenic cooling, says Brock. For $20,000 it is now possible to buy an off-the-shelf cooler that sits in a 19-inch equipment rack. That's a huge advance on what was available twenty years ago.
However, Brock reckons that with custom designs and volume manufacturing it will be possible to reduce prices to $2000 per cooler, and to shrink the box down to nine inches high. What's more, one cooler can easily accommodate 20 or so chips, bringing the cost per chip down to something more reasonable, he says.
"We need to dispel cryophobia," says Brock. Cold is good because it reduces the noise in the electronics. Ordinary silicon circuits can benefit from this, too. "I think people should seriously look at cooling as a faster route to achieving high speed."
As an example of what can be achieved, he cites Kryotech Inc., a company that takes off-the-shelf PC processors and cools them down to achieve better performance. "Kryotech had the AMD Thunderbird running at 1.5 GHz 18 months ago -- 50 per cent better than it's rated."
-- Pauline Rigby, senior editor, Light Reading http://www.lightreading.com