Can Darpa fix the Cyber Security ‘Problem From Hell?’ | Hacker News Reader

Can Darpa fix the Cyber Security ‘Problem From Hell?’

wired.com

20 pointscanistr4 comments

I attended a presentation a while back by a guy who was working on thermal fingerprinting/watermarking for ASI[CP]s. It was targeted at design/soft-IP producers, to detect if a particular chip was using their SoC modules without a proper licence, by observing particular thermal signatures each IP block generated. The cool bit was how it integrated deeply into the actual block layout, to make it harder to remove.

It wouldn't stop the problem of a malicious manufacturer with sufficient resources from adding nasty things, but it could probably raise the bar a bit.

This is reminiscent of the problem laid out in Ken Thompson's "Reflections on Trusting Trust."

http://cm.bell-labs.com/who/ken/trust.html

TL;DR version - Foreign governments can inject malware at the chip level, we don't have a second source.

I think the short answer to this question is 'no.' The more interesting question is can you write software that defends itself against a chip trying to subvert it? Possibly. Some of the original triple redundancy with voting research that was done for the space program might be useful here. Basically run three chips in parallel and insure they all do the same thing. This would have addressed the SCADA malware as it affected the operation of a single PLC but wasn't synchronized to other PLCs. So using a voting scheme for 2 out of 3 you would have had reliable operation.

I think one of the winners would be Intel here. They have the most fab capacity in the 48 contiguous states. With government contracts mandating the chips be built on Intel processes in the U.S. of A. they could at least protect the supply chain some what and act quickly when they suspected a problem.

Redundancy and majority-vote decisions are not much use if you're relying on 3 instances of the same subverted hardware.

(A similar problem applies to non-malicious uses, where multiple systems (even with N-version programming, clean room implementations, etc) interpret the specification incorrectly and the majority vote the wrong way.)

Second sourcing and strict requirements for different hardware implementations would cut a lot of the risk, especially if, as you say, one of them is a trusted, domestic manufacturer.

The level of global production processes for high technology used in weapons would make for an interesting additional dynamic to military foreign policy. You can't really go attacking the guy who makes the widgets for your laser guided bombs unless you've got either a really big stockpile (and risk economic problems and obsolescence), or you can build them domestically if necessary.

The statistic quoted in the article:

"The Pentagon now buys 1 percent of all the world’s integrated circuit production"

is pretty scary. I can't seem to find any reasonable source for the total IC production available. http://dx.doi.org/10.1126/science.1200970 looks interesting, but I don't have access, so I've gone with the Ars summary on it: http://arstechnica.com/science/news/2011/02/adding-up-the-wo...

Assuming they're in it purely for GPU power, 1% of the estimated 10E12 MIPS is 100 Billion MIPS for the US defence establishment. Of course, 90% of it will be custom fabbed stuff for embedded systems, but the sheer scale of it is mind-blowing.

Reply on news.ycombinator.com