The ongoing collapse of Moore’s Law is one of the least reported stories in technology today. That old canard says that the number of transistors that’s possible to fit onto a chip will double every two years. Like a shark, our processing industry can only survive in its current form thanks to that constant forward motion — and with the rate of semiconductor advance now slowing, the industry could well slow down with it. Silicon is, at this point, totally incapable of providing any further advancement; a new material is needed, one that can be reliably laid down at scales well below 10 nanometers. For a long time, experts have argued that carbon nanotubes (CNTs) are the most likely answer, and this week IBM announced it expects to have a commercial CNT chip ready by the year 2020.
When it comes to computer chips, size is everything. Our devices, be they smartphones or desktop towers, present a hard physical limit on space; so long as we are still switching physical transistors on and off to do computation, making devices faster means packing more transistors into the same physical space. Engineers have spent the last decade or so forecasting the end of silicon as a metal that could support much further miniaturization; its properties make it inherently difficult to lay down at the scales we’re beginning to require. But carbon nanotubes are just rolled up tubes of graphene, which is only a single atom thick; though engineers have been talking bout graphene as the future of transistors for a while now, IBM is the first to really put itself out there on the issue.
The first CNT transistors were built in the late 1990′s but since then we’ve made little progress in creating chips with billions of those transistors packed densely and — above all — affordably. IBM has already demonstrated the ability to create processors with about 10,000 transistors, but that’s still a long, long way off what we’ll need. The manufacturing process they’ve chosen for this project sees units of six CNTs acting as each transistor. They’re about 30 nanometers long and 1.4 wide, spaced eight nanometers apart — given their calculations, a CNT processor could be six times faster than a modern silicon chip for the same power draw.
The problem is the same as it’s ever been: it’s mechanically very difficult to pack things that closely together without losing quality and accuracy in manufacturing. One possible solution is to use labelled transistors that can be laid down at one scale but which will then self-assemble to pack even more tightly. It would be gut-punchingly expensive to replace all the manufacturing infrastructure that exists around making silicon chips, so much of the funding these days goes to finding ways old manufacturing tech could produce new chip tech.
It’s unclear whether IBM has made a specific breakthrough that led it to this announcement or just a general feeling of progress and meaningful forward movement. Either way, the company is upfront about the fact that if CNT computers don’t manage to make some sort of move by around the year 2020, the window of opportunity may close. Potentially competing technologies are also under development, from quantum computers to optical computers and beyond, and their potential to increase computational power is far greater than CNTs.
Still, carbon nanotubes, and their sister-material graphene, are by far the most likely candidate to replace silicon in the short term. If we don’t have a fundamentally different sort of chip design by the year 2020, we could see the computer hardware industry slow down markedly. And if even the computer business can’t thrive in the new economy, then there it’s official: in the new economy, there is absolutely no such thing as a safe bet.