Engineers Build "Self-Healing" Chips Capable of Repairing Themselves

hypnosec writes "A Team of researchers and engineers at California Institute of Technology (Caltech) has developed 'self-healing' chips (PDF) that can heal themselves within a few microseconds. The team tested their work by damaging amplifiers in several places using high-powered lasers. In less than a second the chips were able to develop work-arounds thereby healing themselves."

Comments

More accurate to say "More resilient chips"?

[Looker_Device]

Not to be too pedantic about it, but I'm very touchy about biological metaphors being inappropriately applied to technology (lets we forget how amazingly complex evolved biology really is compared to even our most advanced tech). FTFA, it sounds like they don't really "heal," they just reroute around the damage. But the damage is still there. It's more analogous to network packets being rerouted around a bad server than a biological entity actually replacing damaged cells.

Re:More accurate to say "More resilient chips"?

[sensationull]

Thank you, that was what I was about to say, massively redundant, cool but it does not actually repair itself back to the way it was before, as it 'heals' it uses up that ability.

Re:More accurate to say "More resilient chips"?

[N!k0N]

Evolution's had a damn long time to get the "rebuilding cells" part down -- we're just at the "stop the bleeding" phase with the chips. Once they can rebuild their structure, we're in for trouble...

Re:More accurate to say "More resilient chips"?

[bluefoxlucid]

Usefulness for a purpose has disconnected but relevant bearing on how it works. Operational details specify what problems the technology can solve; there is overlap for bare function, but the details are important. In this case you could have a chip with an inherent structure that would self-regrow when current is applied if damage is not extensive and materials were present (i.e. cracks self-heal infinitely, uses electricity); or a blue-goo type chip that contains a reservoir of consumable raw material to accomplish same purpose (consumes goo and electricity); or redundant schematics back-ups and FPGA (limited surface area to damage; potential routing problems); or redundant schematics with blue-goo (same, different implementation, in practice one will have an advantage of weight and effectiveness over the other but basic concerns are identical); or a simply redundant chip (prone to manufacturing deviations reducing its effectiveness; certain failures may be fatal; you need 100% additional capacity plus routing for each 100% redundancy; redundancy is per redundant component--100% blue-goo could repair the same 1% of surface through 100 failures, but 100% pure redundancy can recover from the same 1% surface failure 1 time).

This is not really "self-healing" but "redundant". Of the three above, you have these considerations:

Stable reflowing crystal: Can't heal from large damage; small damage (microcracks, minor fissures, electrical/mechanical stress) should heal. Chip normal lifecycle is extended indefinitely.

Blue-goo self-repair: Can heal from larger damage for a limited supply. Works off total damage: Eventual cumulative loss over 100% of chip area with 100% repair capacity will repair itself. Beyond capacity, repairs cease. Most likely, blue-goo would have transport issues into very tiny micro-fissures and cracks--cracks embedded deep inside substrate are not physically reachable and won't repair, yet may affect electrical properties of the substrate and thus impact chip performance.

Redundancy: Absolute healing from any type of failure. Redundancy takes up additional space for each redundant copy of each component: if you have 2 redundant copies of an ALU, you need space for 3 ALU total. If you lose one or two, it's functional. If you have one redundant copy of every component (100% coverage) and you lose two of any one, you lose the whole package at what may be 10% or 1% or more or less utilization. Contrast with blue-goo, where 10% loss means 10% loss of capacity to self-heal. Likely a high-stress area will require more specific redundancy; and bad luck could render a chip useless even with all that unutilized self-repair (redundant) capacity.

The holy grail would be a stable reflowing crystal with blue-goo that can repair the crystal. Tiny cracks would self-heal, while larger damage would trigger self-repair. Redundancy is worthless: if you have 3 ALU, you should be parallel processing instead of idling 2 of them for fail-over. Redundancy only seems desirable because we don't have a better method like blue-goo or stable reflowing crystal.

Re:More accurate to say "More resilient chips"?

[jones_supa]

Not to be too pedantic about it, but I'm very touchy about biological metaphors being inappropriately applied to technology (lets we forget how amazingly complex evolved biology really is compared to even our most advanced tech). FTFA, it sounds like they don't really "heal," they just reroute around the damage.

Some of the biological processes also route around the damage, the brain being a good example.

Re:More accurate to say "More resilient chips"?

[drinkypoo]

It seems dumb to me to have unused functional units lying around. But if a chip could detect that a functional unit has failed (by cross-testing?) and then degrade performance by not using it, while continuing to operate so that I can at least get useful fault information, that would be a massive win.

That BS again....

[gweihir]

They are NOT "self-healing". That would mean they can get back to their original state after damage. What these things have is a high level of redundancy. But whenever they suffer damage, the redundancy gets less and eventually they fail. Calling this "self-healing" is a direct lie.

Re:That BS again....

[webmistressrachel]

| You can't repair the building blocks in electronics.

Yet.

Re:That BS again....

[ledow]

Agreed.

But still has interesting implications for, say, radiation-hardened hardware like space-travel. Of course, it's nothing they don't already have in terms of the overall process, but having it on-chip is yet-another factor that has to experience corruption before you need to replace the hardware.

Another nice step, but nothing miraculous.

Re:another non-story

[fuzzyfuzzyfungus]

Man, it makes me sick that people haven't taken the obvious step of giving the intricate metal layers and zones of dopant concentration on a silicon wafer the same modularity as 3.5 inch HDDs with hot-swap connectors... Scientists are so lazy.

Heck, why do we get worked up about integrated circuits at all? I saw Bell Labs demonstrate the same concept with discrete transistors before 1950, and they were basically just ripping off vacuum tubes...

Real Genius?

[luke923]

After reading CalTech and high-powered lasers, I could only think of a ragtag team of students like Mitch Taylor, Chris Knight, and Lazlo Hollyfeld implanting a two-way transceiver into Kent's dental work in order to thwart Hathaway's plans to embezzle funds from the DoD.

Re:another non-story

[bluefoxlucid]

RAID is only 15 years old? It came about in like 1998?