Self-healing chips survive repeated LASER BLASTS Modern high-speed integrated circuits can be fragile things. Even a single fault can often render them completely inoperable. But a team of researchers at the California Institute of Technology (Caltech) says it has developed an "immune system" for chips that can allow circuits to route around problems and keep working in the …
Their approach of cycling through all the states to find the most appropriate won't scale particularly well. log2(262144) is only 18. There are 18 individual stateful objects on that chip, and the maximum re-routing time was 0.2s. Even the relatively modest CPUs on board the Curiosity rover sport over ten million transistors. Quite a lot of states.
True. However those same 10 million transistors can probably be replicated several times over onto a suitable sheet of silicon these days. If "sensors" can be replaced by "testing units", maybe you can have a 32-core CPU with instructions replicated N times for redundancy and an automatic failover amongst various CPU cores and sub-components of cores, all in one package?
Now that could be useful on Earth, let alone Mars.
What if it actually worked on a computers innards?
We'd (in theory) never need to replace that hardware. I do wonder if the company would live long enough go out of business due to having saturated it's market, or if it would be destroyed by the rest of the industry taking it out in self preservation.
Routing around physical damage from a laser is what I understand from TFA. Modern processors (or SoC) are tiny, and tend to be surrounded by laser reflecting heatsinks, gold, silver and tin contact material. You'd have to be pretty accurate with your laser to cause the self healing to initiate, assuming that the stuff you lasered through, to get to the chip, was also capable of self healing.........
At Ferranti in the 70's & 80's EMP proof processors were developed for use in computers, comms equipment & targeting devices. Those babies could carry on during & after a nuclear airblast. However it was accepted that if the equipment was close enough to be physically damaged....
win lose
___ ___
lose win
My guess is that the damaged chip uses less power because it has fewer components to consume power than an undamaged chip does. Of course, this also means the chip is less powerful. That should be expected; all those parts are there for a reason, one would think, so destroying a few must result in a performance trade-off somewhere.
This assumes a chip is homogeneous in design and the loss of any given unit would fine. So redundant units would probably be necessary in many cases. At which point one wonders if it would be better to just have redundant chips and the self healing ASIC(s) in separate chips at mission critical areas.
As others pointed out the self healing ASIC is a single point of failure. Perhaps it would be better to have this ASIC's functionality spread throughout the chip. But again this would lead to considerable complexity. Chip size, heat, and manufacturing defects would probably rise. i.e. bigger target, more cooling, but at least the chip could 'fix' it's own defects right?
This isn't really healing at all. It's just internal failover whose premise works best on simple chip designs. As I suggested before an external failover mechanism is probably more ideal for more complex setups.
What we need is real self healing via nanomachines and temporary failover. Don't see that being around the corner any time soon. Or IC's made out of organics.
Well, not so much. Take an x86 as an example.
A quad core I5 has multiple floating point units. It, also, has more than one cache. Well, not more than one L2, no. But Intel can, an do, disable some of it - to make an I3. So, yes. We could route around these (eventually) defective parts.
No, it wouldn't survive a (real live) laser blast - but would keep ticking even with half L2 and/or one less floating point unit.
Could be handy in servers and hardware in hard to get places...
Exactly my thoughts. A complex system might need several healing units placed in different locations. While I agree that in the short term, more fuctional units might be more effective than healing units, longer term and for systems that stretch over geography or even in space, then the multi nodal healing units might be necessary for 100% uptime. Think of the internet and routers. There is enough redundancy in the system that 1 or several boxes fail, it routes around. However in a more "mission critical" system, redundancy may not be practical. Healing systems need HW + SW to achieve optimum results.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
