Tuesday, January 25, 2022

Carbon Nanotube-Based Microchips: A (Very Short) Primer

As Moore's Law runs its course those looking forward to continued improvements in computing power necessarily turn their attention in other directions--not least, a turn away from silicon to other materials as the material from which we make our integrated circuits. For decades one of the more promising possibilities has been the use of carbon nanotubes.

Carbon nanotubes are one of many materials formed from an arrangement of carbon atoms in hexagonal (six-sided) structures, themselves arranged in a lattice, with every one of the six atoms connected to and forming part of the adjacent structures of the same type. Left as a flat sheet, the resulting material is called graphene. With a nanotube what happens is that the "sheet" is bent, and its opposite, horizontal ends joined, to form a hollow cylinder.

These nanotubes' ultrathin bodies and smooth walls enable charges to flow through them more rapidly, and at a lower supply voltage, than is the case with silicon, making them potentially faster and more energy-efficient. Their smoother, more energy-efficient structures may also make them a more suitable material than silicon for denser "3-D" chip designs--which would not just put more transistors on a single 2-D surface the way conventional chips do, but layer those 2-D surfaces on top of one another to cram more computing power into a given space. And altogether the combination of attributes has led to speculation about their permitting a thousandfold gain in computing speed over silicon, equivalent to about ten more doublings and an additional couple of decades of Moore's Law-like progress, which as a practical matter that would convert today's personal computers into machines comparable to today's supercomputers.

Of course, in considering all this one has to note that there has been a long record of great expectations and great disappointments in regard to the mass-production of carbon nanotubes, not least because of the old "works well enough in the lab but not ready for real life" problem of consistently getting the required quality at a competitive price when producing at the relevant scale. (Back in 2014 IBM said they would have them in 2020. Well, 2020 has come and gone, with the arrival of the technology, like that of so many others, deferred indefinitely into the future.) Still, it is one thing to acknowledge that the technology has been slower to emerge than hoped, another to write it off--and it may well be that the belated arrival of carbon nanotube-based chips, and the boost they deliver to the continued progress of computing power, will be what opens the way to the next great round of advance in the development of an artificial intelligence sector that it may turn out has been held back from realizing its promises only by the limits of the hardware.

No comments:

Subscribe Now: Feed Icon