Feeds

back to article DARPA shells out $194m for 'phase 6' of STARnet chip project

War tech agency DARPA is not happy with the pace of progress in semiconductors, so it has been funding primary research through a program called Semiconductor Technology Advanced Research Network - or STARnet for short, for the past several years. And it has now announced that it is kicking in another $194m over the next five …

COMMENTS

This topic is closed for new posts.
Gold badge
Thumb Up

Mad enough to make some sort of progress.

But you have to wonder why did we not hear of the worldshaking discoveries of the previous 5?

Perhaps because they didn't?

But that said let's not forget that one DARPA AI project made savings that paid for all DARPA's AI research for the last several decades.

I for one welcome our new nanomaterial based spintronicly actuated swarming overlords.

1
0

Is it just me......

......or do those logos down the side of the article look like they came straight out of a sci-fi zombie apocalypse game ?

0
0
Silver badge

Re: Is it just me......

What, things like Tyrell Corp, Weylan Yutani, Ono Sendai, Omni Consumer Products, Cyberdyne Systems Corporation, General Forge and Foundry, General Products...

0
0
Silver badge
Terminator

Re: Is it just me......

I was surprised that Aperture Science didn't have their logo up there...

3
0
Silver badge

Re: Is it just me......

D'Oh! I knew I missed an important one!

2
0
Gold badge

Re: Is it just me......

Tempel I.G. ?

1
0
Silver badge

Whatever happened to...

Using genetic algorithms to design better, faster and more efficient chips?

0
0
Anonymous Coward

Re: Whatever happened to...

They tried it but realised they needed better, faster and more efficient chips to run the genetic algorithms. :)

4
0
Silver badge

Re: Whatever happened to...

They seem to be researching the building blocks of chips, rather than optimising chip layout and scheduling- which appears to be the focus of most recent papers on GA for chip design.

2
0
Anonymous Coward

Re: Whatever happened to...

"Using genetic algorithms to design better, faster and more efficient chips?"

The designs "worked", but why they worked wasn't clear, and often they "worked" only because of a quirk of the implementation.

There was one case where genetic algorithms designed a chip that had a block of gates in the middle that didn't seem to do anything - it wasn't connected to anything else in the part. However, when it was deleted the part stopped working. MUCH analysis and head-scratching later, they came to the conclusion the gates were introducing noise onto the power bus, and that noise was needed elsewhere in the chip. However, shift just one aspect of the fabrication - use thinner gate material, or do a geometry shrink - and the design no longer worked.

"Designs" like that aren't terribly useful - if a genetic algorithm gives you something where you don't know what aspects you can change without breaking it, it stops being engineering and becomes magic.

2
0
Boffin

Re: Whatever happened to...

My team did a bit of work running GAs and other non-linears on supercomputers back in the mid-90s for, of all things, marketing research. We got results very much like the chip layouts you mentioned - the numbers worked, but no one could explain why, much of it seemed counter-intuitive, and worse we couldn't explain why a given campaign or marketing idea was really better...we could just show that the results DID come out better. But in the end, the difference wasn't huge, at that time GAs and hardware were expensive, and in the end our clients needs for clarity trumped a few percentage points of results for the money involved - unless it was a BIG campaign. So mostly, we went back to linear models, which we could always explain...nice to see it wasn't just us!

1
0
Silver badge
Pirate

Uh,Oh!

Is that DARPA icon an inverted pentagram?

Maybe all those conspiracy theories about Skull and Bones etc. are right.

<starts incantation whilst sharpening magic sword>

2
0
Bronze badge
Stop

Funky?

The Center for Systems on Nanoscale Information Fabrics... wants to move from a deterministic to statistical computing and communications.

No, just NO.

1
0
Devil

Re: Funky?

"Excuse me, we've just lost contact with the Phobos labs"...

2
0
Gold badge
FAIL

Re: Funky?

"wants to move from a deterministic to statistical computing and communications."

Well you could argue that the brain is a "statistical computing" mechanism producing results based on a complex statistically weighted output from a neural web.

Of course would you trust such a system to calculate your paycheck is a different question.

But how does that acronym work exactly?

1
0
Terminator

Love it !

"And now we know who we have to go back in time to get, Jeffrey ®"

That.is.all

3
0
This topic is closed for new posts.