DARPA gives Cal Tech boffin $6m 'to save Moore's Law'
Invulno-chips that can 'self-modify' in microseconds
Maverick Pentagon deathboffins aim to prevent processing progress grinding to a crunching halt in the next few years by developing "self healing" integrated circuits, able to repair themselves in the event of damage or failed components.
The idea here is that as more and more teenier and teenier transistors are packed into an integrated circuit, particularly when turning the overclock knob up at the same time, it becomes extremely hard to make sure that there will be none with flaws.
"As transistors approach atomic dimensions and run at very high frequencies, even very fine-scale variations within seemingly identical transistors can make a large difference in performance," says Dr Ali Hajimiri of Cal Tech. "Some circuits may run faster, some slower. Some may actually fail."
And that simply won't do, because it means that Moore's Law - on which the entire fabric of the universe has now become dependent - might stop working. Quite apart from devastating the underpinnings of the IT industry, and even more those of IT journalism, this would have other consequences. In particular, the US war machine's continual push for faster, better, madder electronics could be stymied (let's not forget, after all, that integrated circuits were first developed for use in missiles).
Our old friends, the paradigm-punishing boffinry iconoclasts of DARPA, aren't having any of that. That's why they've just hired Cal Tech's Hajimiri to sort things out, for $6m. According to the Cal Tech statement:
DARPA's Self-HEALing mixed-signal Integrated Circuits, or HEALICs, program is designed to enable the continuation of the Moore's scaling law, which predicts an exponential increase in the number of transistors that can be placed on an integrated circuit (one that performs multiple functions), in the face of inevitable imperfections in those transistors.
Hajimiri's solution is to employ sensors that can detect the conditions within a circuit - and determine, for example, that a particular transistor is not working up to par, or at all - along with actuators that can then modify the system. For example, the actuators could swap functional transistors for failing ones, or add "helper" transistors that would boost the functional capability of a transistor running at sub-optimal speeds. All of these modifications ideally would be made within thousandths to millionths of a second, effectively fixing failing circuits on the fly.
"In a few years self-healing circuits will continue Moore's scaling law by making integrated circuits resemble living organisms in their ability to self-heal and adjust to changes in the environment," Hajimiri says.
Nice try, lads. Saving Moore's Law - good one. But it's quite plain that you're creating indestructible computers which can't be damaged or wear out, and which can heal themselves - probably even modify, redesign and improve themselves - millions of times a second.
I think we all know where this is going. ®
Sponsored: Hyper-scale data management