DNA-carbon nanotube microprocessors — small hope for a big shift?
Moore’s Law on a new curve
Comment We may have a long way to go before Moore's Law, which calls for the doubling of transistor counts — and therefore computing capacity — every two years or so, runs completely out of gas on current electron beam and optical lithography techniques.
The question is, will the IT industry, which is predicated on the idea that computing gets cheaper over time and we use more of it all the time, be able to afford Moore's Law?
Just as we have hit the thermal ceiling for server processors — somewhere around 3GHz for the x64 architecture — well ahead of where many had predicted (remember the 10GHz Xeon that never happened and never will?) — we may hit a budgetary ceiling for future wafer baking technologies long before we get to the theoretical physical limits of lithography.
Being in the chip-making business is a rough proposition. Each shrinkage of technology shows the same kind of exponential growth in terms of the cost it takes to develop and implement a particular chip making process as the aggregate compute performance of a microprocessor grows, thanks to the addition of more transistors that implement more cores, memory controllers, and other gadgets onto a given piece of silicon.
Talking telephone numbers
The numbers are staggering. IBM has pumped $1.5bn into its chip factories in East Fishkill, New York, to eventually upgrade them so they can spit out chips based on 32 nanometer and 22 nanometer processes — processes that its Power8 and Power9 server processors as well as its future mainframe engines all depend. GlobalFoundries, the wafer baker spinout of Advanced Micro Devices, is sinking $4.2bn into a nearby foundry in upstate New York.
Intel said earlier this year it would spend $7bn to upgrade its fabs in the States, like this investment is no big deal. But it is a big deal, and even Intel can't fight the economics of Moore's Second Law, which says it gets more and more expensive to keep pushing chip technology.
Everybody is looking for a way out of this Moore's Law conundrum, where processing capacity goes up exponentially and so does cost. All the computing power in the world doesn't mean much if you can't afford it.
And the odds favor that something radically different - a new kind of computing — will eventually emerge, just because the best minds in the semiconductor business have admitted they are pushing up against some pretty hard limits.
Which is why there has been a certain amount of excitement about a bit of research that pairs two very different technologies in a way that might — perhaps a decade from now, when Moore's Law hits the wall — give us a new way to make processors for PCs and servers.
Chip geeks at IBM's Almaden Research Center in San Jose, California, working with scientists from the California Institute of Technology, are domesticating virus DNA to create what could be the future of computing.
Synthetic viral DNA has been domesticated by IBM and Caltech researchers to be a kind of coral reef upon which other kinds of future semiconductor technologies such as carbon nanotubes, silicon nanowires, and other nanoparticles can be grown. The DNA sequences are mini scaffolds or circuit boards, depending on how you want to think about it, says IBM.
The neat bit is that the DNA origami technique allows for these DNA reefs to be placed on silicon wafers that are created using current semiconductor manufacturing techniques, and in an organized fashion that may make the production of chips based on carbon nanotubes or silicon nanowires economically feasible.
Right now, the cost of progressively shrinking the wires etched onto silicon is getting higher and higher with each successive shrink, and the lithographic techniques used today are going to hit a wall when they try to go below the 22 nanometers.
While few doubt that Moore's Law will carry on below 22nm, exactly how this will be done is still open for debate. Moving from light to x-ray lithography would require a vast retooling of the semiconductor industry, something that chip makers already hammered by competition and the global recession very much want to avoid.
IBM and Caltech outline the combined DNA origami and semiconductor manufacturing techniques in the journal Nature Nanotechnology. Paul Rothemund of Caltech's departments of Computer Science, Bioengineering, and Computation and Neural Systems, is the lead researcher on the self-assembling DNA scaffolding idea. IBM helped figure out how to make the DNA scaffolds stick to silicon wafers, and Gregory Wallraff is the lead on that effort.
The technique they came up with uses electron-beam lithography or optical lithography to etch DNA binding sites onto silicon dioxide or diamond wafers. While the Caltech team can mix the synthetic viral DNA with different lengths of shorter molecules called oligonucleotide strands that bend the DNA into different shapes — triangles, squares, stars, and so forth — all from a simple solution of slime, the DNA origami, as Rothemund calls these shapes, spread themselves randomly around a bit of silicon wafer.
IBM has figured out how to punch little holes into the wafer at regular intervals, and based on the properties of each DNA origami bit, that allows the DNA shapes to be herded up and spaced evenly at resolutions as small as 6nm.
Speaking to Reuters, IBM researcher Spike Narayan said that this was the first time that biological molecules had been used to help with semiconductor manufacturing. (Excepting all the people in the bunny suits who want lunch hours and health benefits, I presume.)
He also raised the prospect that the DNA origami scaffolding technique might allow for future chips to be made using a million dollars worth of polymers and DNA sequences instead of hundreds of millions of dollars in etching and doping gear.
We're a long way from that point, though — if this technology even gets out of the lab and into a testbed, much less full-scale production. But once it does, we all know that resistance is futile. And that will be the case because of the downshift in the cost of making chips that such a strange process holds.
Such a hybrid DNA-carbon nanotube chip would put the IT industry on a whole new Moore's Law curve. Which is what it really needs. There has been much progress — a stunning amount, in fact — but no radical shift. ®
Sponsored: Benefits from the lessons learned in HPC