DARPA gives Cal Tech boffin $6m 'to save Moore's Law'

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 …

[Anonymous Coward]

Self-Healing or Self-Defeating

What exactly is the point of achieving atom-scale transistors if the only way to use them is to add stocks of spares and (relatively) hulking great repair factories on board. I can't imagine these nanomachines being smaller than current transistors, so it sounds like one step forward and two steps back....

[Andy Bright]

Re: Self-Healing or Self-Defeating

Nothing defeatist about robot deathbot overlords that can heal themselves..

[Steven Knox]

Good start, but...

A single HEALIC might suffer from a macroscopic failure that the nanomachines cannot repair. So to be safe, systems should have at least two. And of course they should be able to report their status to the software, so the software can be able to prioritize its tasks based on the state of the resources -- probably have to base it on some sort of neural network.

Yes, double HEALICs with a self-modifying neural net might just be the answer...

[Rob Dobs]

should be more elegant than that

I don't think there would be an additional size needed here. The actuators would simply allow redesign of the circuit path. If transistor A has B & C as a path, and C breaks, instead of the being dead the actuator just makes it where A only can go to B, and no longer "knows" about C.

This wouldn't be spare transistors, just the ability to work around dead ones.

Let me just add what utter crap this goal is - THERE IS NO MOORE'S LAW !!!!!

It does not exist and it is not a problem to be solved. This theory really should be called "Moore's Observation", as it has not even held out to be predictive. If you plot the release of CPU's to market, they do not have any designed doubling of transistor count. Competing CPU companies constantly come out with new designs. Transistor count also does not solely define processor performance. Further Memory bottlenecks and Hard Disk speeds are more of a bottle neck currently than transistor count is. Shrinking the die size 120nm to 45nm and smaller, Double Date Rate (using the rising and falling trough of the wavelength) using multiple cores and other technological tricks have enable the forward progress of processing power NOT Moore's Theory/Observation. I could just as easily plot the number of diseases cured by science, find a nice window of time they fit in, and Declare "McCoy's Law" - that the number of diseases will be cured by science will double every X years. or "Loki's Law" that system memory will double every two years. They do not have any predictive capability at all, and are not even sound theories. The idea that the US, humankind or even computing would be harmed by us not making Moore's Law self fulfilling is pure and utter garbage.

REG, please help, can we please stop using this uber-retarded term, or at least have a constant * added to the phrase?

* Not really a Law, or even a proven theory, more like a goal for forward progress.

[Steven Davison]

I for one...

I for one welcome our new silicon based, self altering overlords...

[Glen Forde]

I for one....

Oh no, we don't do that anymore do we. ;o)

[Anonymous Coward]

Well, for 6 million dollars...

... you can create The Six Million Dollar Man! If an electron falls off its path, he can just punch it back on with his slow-motion t-t-t-t-tah punches.

And since we're talking about technological limitations: 640KB was ~never~ enough for Chuck Norris.

[Graham Marsden]

May I be the first to welcome...

... our self-healing, self-replicating, self-improving machine overlords...!

[Anonymous Coward]

Well ...

I for one welcome our self repairing, self modifying, self-improving HEALIC overlords ...

[Andrew Kaluzniacki]

Caltech - one word not two

Fire for Ricketts Hovse

[ian]

Breaking Moore's law?

I was under the impression (mistaken it seems) that the problem with Moore's law had to do with quantum effects. Something about qbits not being the same as bits. No need for self-repair if transistors can be both damaged/undamaged simultaneously...

[Anonymous Coward]

I for one ... erm ...

Oh stuff it! There are no permutations or combinations of "I for one" jokes that haven't been used already. We need a new joke paradigm.

[Camilla Smythe]

No!!!! It's True!!!!

At the sub-nanometre level all sorts of quantum maniacal stuff that makes things not work becomes available to fix it in, near enough, real time.

Consider a patchwork of fields containing the sub-nanometre goop required for digital life. Supply them with 'low tech' but viable sub-micron 'husbandry' sections.

Give the 'farmers' the information required to build, monitor and maintain their fields and you have a quadzillion core processor.

Hook the fields back into the 'farmers' and say goodbye to human domination of Planet Oith.

Well that assumes you hook them back into two stroke powered fartbots.

Be afraid, be very afraid.

[Anonymous Coward]

Redundancy?

Why not just design the circuits with redundancy built in? I'm thinking here of the kind of redundancy found in information theory and its applications, such as telecommunications. Come to think of it, that's already being done with memory devices that use error correction codes, isn't it?

Okay, maybe I'm not at all up to date with this stuff.

[Glen]

@ac 1549

i can mostly see this affecting yield. A process decrease that is beyond the current realms of profitability could be made viable if the yield were increased.

the question is, does the potential die shrink/speed increase outweigh the increase in overhead? (routing will be a nightmare)

*points to the number of transistors required to implement pipelines/superscalerblah*

slighty OT , this could be used in an fpga type scenario where whole areas of the chip could be reconfigured *on the fly*

no fpu? no problem! but give me a nanosecond to change these banks of sram... what dya mean you want to keep those?

[W.S. Burroughs]

& another thing...

How will the system deal with failures in the sensing & switching circuits. My dear old friend Control is going down his familiar rabbit-hole.

[Dave]

It's What?

Self Healing Information Technology?

[Anonymous Coward]

Re: I for one ... erm ...

"Oh stuff it! There are no permutations or combinations of "I for one" jokes that haven't been used already. We need a new joke paradigm."

I for one will welcome our new joke paradigm.

[darrin allen]

Moore's law and implications with nano-technology

I think nano tech will be a big asset to Moore's law in the future.

I am interested in PCB's and microelectronic devices. I hope to grad work in the future

[Anonymous Coward]

Bloody silly.

I also agree that this is just going to end up chasing its own tail down a diminishing returns alley of massive overhead and circuitry bloat. Moore's Law isn't going to last forever, let's just get used to the idea and start designing the new types of architectures that are going to take us forward from where we are today in computing performance.

[Andrew Bassi]

Moore's Law ....

@Rob Dob

Not looking to rain on your parade (you clearly dislike the idea of 'Moore's Law').

Moore's Law isn't about increasing processing power or speed whatsoever. Moore's Law simply states that :

the number of transistors that can be placed inexpensively on an integrated circuit has increased exponentially, doubling approximately every two years.

Which is (within acceptable limits) true.

Whereas there may not be a designed 'doubling' of transistors, plotting the number of transistors on microprocessors over the last 51 years since Moore coined the term shows that it holds true.

As with almost all scientific 'laws' it is an observation. Until it is proven to not produce an accurate prediction, it stands.

[goggyturk]

Iconoclast?

Iconoclast death-boffins...? I can just see those rabid geeks rampaging through the Vatican, smashing up statues of the Virgin Mary and burning the sacramental tables as they go. The Swiss Guard must be bricking themselves.

This sounds like the bastard offspring of Issac Newton and Richard Dawkins.

[Alistair]

Six million dollar man

You can make me one if you want - just give me six million dollars and I'll make up any crap that you want me to make up.

Yes redundancy as we call it is already used in memory chip manufacture. During factory test, the broken memory lines/towers/arrays are swapped out for spare ones by setting one-time-programmable switches on the chip. It increases the yield (percentage of chips on a wafer that work) so even though it decreases the number of chips per wafer (because each one is a big bigger than it would be without redundancy), it maximizes the number of working chips produced for a given wafer cost.

Mines the one with the matching cleanroom booties, ta.

[Ray]

Uh oh...

In 2029, Cyberdyne Systems 101 reroutes YOU.

[Robert Forsyth]

Where's Ivor Catt

http://en.wikipedia.org/wiki/Ivor_Catt

http://en.wikipedia.org/wiki/Wafer_scale_integration

[Chris]

@ Steven Knox comment

Fantastic - thanks for that one.