The Year Ahead: Gazing into the chip crystal ball
But the contents of the CD were electrifying: hundreds and hundreds of technical papers given at the IEDM in Washington earlier in December detailing the state of the art in semiconductor research.
Some of the ideas will come to nothing: intriguing experiments into blind alleys. Others will turn into useful but mundane aspects of bread and butter chip design.
Among these papers, however, are the ideas that will fuel the fundamental technologies of high tech in five years' time. It's a real ghost of Christmas future. Here are some of the highlights:
• Free Space Optical Chip-To-Chip Interconnects. All the processing speed improvements in the world aren't any good if you can't move data between chips fast enough. But fast buses have lots of problems--noise, radio frequency interference, short maximum distances, impedance matching and so on. One answer is to use light instead of electricity. Sadik Esener from the University of California in San Diego reports on practical ways to bond optical components to standard chip packages, and on demonstration systems that run at up to 2.5 gigabit per second (gbps), linking two chips a foot apart and using only a few milliwatts of power. Error rates are fast approaching those of electrical systems, and he reports densities of around 1000 independent signals per square centimetre of chip area. So far, the chips use infrared: he doesn't mention this, but if they move to visible light the inside of our PCs will look spectacular in the dark.
• Microelectronics meets Molecular and Neurobiology. Science fiction might be strong on machine-human cyborginalia, but the practical side of linking biological systems to silicon devices is a long way from reality. Peter Fromherz of the Max Planck Institute for Biochemistry has experimentally grown a network of rat neurons on a silicon matrix and created an electronic interface to them. He says that this is the first example of a defined neuroelectronic circuit on a semiconductor chip, while warning that visionary dreams of bioelectronic neurocomputers and microelectronic neuroprostheses obscure the numerous practical difficulties still to be overcome. But he clearly has those visionary dreams, and he's doing something about them.
• Compared to the Bladerunner world conjured up by the above, a paper from seven Dutch engineers at Philips about a "record high 150GHz fmax realised at 0.18 micron gate length" seems as interesting as tapioca. However, the significance is that the exceptionally fast transistor described doesn't need exotic materials, design tricks or special fabrication techniques--it's all in standard CMOS. This eases the creation of single-chip radio devices working at the very high frequencies needed for very high bandwidths, and also demonstrates one of the perennially forgotten aspects about high technology: old technology can often be developed far beyond its original limits to compete with--and often beat--new ideas also intended to work beyond those limits.
• Quantum-level devices also get a strong showing. A particularly intriguing paper from three researchers at NTT points out that if you have devices that use small numbers of electrons--from one to ten, for example--then they are naturally good at storing numbers from one to ten. You don't need to resort to binary. And to prove it, the NTT engineers built a test circuit where single-electron devices interfaced with more traditional transistors. Applications proposed include analogue to digital conversion and circuits that add numbers without converting them to binary; the benefits are hugely simplified and speeded up circuits. Perhaps the most impressive aspect of the paper is the blase way the engineers discuss the manipulation of single electrons to practical purposes--mundane engineering with subatomic particles, one by one.
• And then there's wireless. Fujitsu researchers predict fourth generation (4G) cellular systems in 2007-2010, with downlink speeds of 100mbps and uplink of 30mbps. It'll cost the same to own and use as cellphones do today, and will need base station antennae that can send sharp beams of radio waves, tracking users as they move around, in order to cope with large numbers of calls. Hugely powerful DSPs will be required to implement the system--future handsets won't have much more than an aerial, a user interface and a single chip--and the researchers point out that these will require very high clock speeds and produce massive amounts of heat. Nonetheless, they're optimistic that the many problems will be overcome, although they warn that this will need much closer co-operation between the people who design the overall system and those who make the chips than has happened to date.
• Finally, technology comes home. Johan Danneels from Alcatel writes that the skills needed to produce good home networking devices are considerably greater than those needed for business and commercial products. The home user doesn't have an IT staff, a huge budget or any time to waste: successful home networking chips have to do absolutely everything themselves and at a very low cost. Because the home user will also want to integrate voice, video, security, music and everything else onto their system, the design has to have lots of very different functionality in hardware and software--different flavours of wireless, DSP, network bridging, firewalls and so on. It's a very exacting challenge, especially for something that most people think is trivial. Danneels thinks that no one company is likely to be able to do it all themselves, so being prepared to co-operate with others from the word go is a vital part of the design strategy.
There's tons more on that CD, and heaven alone knows if any one person can digest it all before the next one comes out in a year's time. But I'm going to try--you don't turn down gifts from the future lightly.
Staff writer Rupert Goodwins reported from London.