Hacker News Reader

Comment by lifeisstillgood | Hacker News Reader

I'm sorry I am not sure what you are saying here. It seems to be "this is far more likely to be a test engineers backdoor that was not on the spec" then a Chinese backdoor added at the fab"

with no evidence either way, I am guessing that US intelligence (and others?) are loudly saying this is happening not because they can prove it in silicon but convincing human intelligence has told them

I happen to think that the techniques developed to analysis silicon are going to be useful all round, and that a backdoor in a chip does not make a successful cyber attack in the wild, but if these are real in-the-fab additions then the implications are so large we should be prepared to not label this "it's just a test component" and look at ideas like validating silicon (noted somewherre else below)

View on HN

> with no evidence either way, I am guessing that US intelligence (and others?) are loudly saying this is happening not because they can prove it in silicon but convincing human intelligence has told them

My instincts would be that in the absence of real evidence they are 'loudly saying this is happening' to beat the war drums, declare it as proof a 'cyberwar' is happening, are using it to get more funding and preparing for new draconian measures to control it both domestically and internationally.

You are correct, there is no evidence either way. So one way to look at it is to consider what would have to be true for it to be installed by the 'fab' without the knowledge of the guy who designed the chip, vs installed by the chip designer.

Given what I know of silicon chip manufacturing, and the verification that goes on during, after, and while, manufacturing. I assert it would be extraordinarily difficult for a fab operator (like TSMC) to insert a back door without the designer/manufacturer knowing it.

I also brought up that in my experience adding back doors was certainly done to aid in testability. Sometimes those aids are done in a way that they cannot be used by third parties (bond-out chips) and sometimes they could be (JTAG access) but are obscured in some way.

Backdoor access in the firmware however, is a much easier threat to actualize as it doesn't involve silicon hacking per se. So that is a more credible threat. And I mentioned that we've seen counterfeit versions of 'name brand' products already which would be a fairly straight forward threat.

The NSA has it's own fab resources. That fact alone tells you everything you need to know. The only remaining question is to what extent is it cost effective to still use suspect parts.

Does the NSA have its own fabs to keep back doors out or to keep secrets in?

Both, likely.

Sorry, the NSA has it's own billion dollar fab soitcan build copyrighted Intel clones?

I simply don't get it?

* what happens when Intel release the nextgenration of chips? Apparently Intel needs to rebuild a while new fab plant at x billion - does the NSA? * do they trust the designs made by Intel? If not what do they do ? If Intel is introducing backdoors for the NSA what guarantee is tere those backdoors won't get used y someone else? * if they do trustthe design but don't trust the fab process surely it is better to put armed guard in the fab room or similar checks * and this is only for one generation of one class of chip. Do this for the chips in the CCTV cameras and the door locks and the ...

Well, AFAIK the NSA doesn't publish what their fabs are capable of or what they do with them. Maybe someone here knows better?

I'd assume the NSA's fabrication capability is more on the scale of the pilot plants fabs build at each new process scale. Some universities certainly have fabrication equipment testbeds as well, so the NSA effort may be more that modest scale.

If I were tasked with the problems the NSA faces, I think I'd at least focus in on:

1. CMOS reverse engineering equipment that can shave down dies, image and analyze the structures, etc.

2. Small scale fabrication for extremely sensitive infrastructure. These roles probably aren't performance critical. Eg if you have some microcontroller that plays a role in say nuclear weapon arming protocols, you need that to be pretty much beyond suspicion.

3. Some way of sampling commodity parts for unexpected behavior non-destructively. If this could be done efficiently enough, you could use it in combination with #1 to get reasonable confidence for off the shelf parts.

One thing I'd suspect is that if the NSA did find highly targeted flaws they probably wouldn't disseminate that fact unless absolutely necessary. Keep an adversary using a strategy you know rather than provoking improvement.

Personally I doubt the NSA forces backdoors into commodity chips. In theory there might be some way of introducing a flaw that would cripple specific large computations like crypto-analysis of a particular code, or biasing a particular random number generator. But that just seems too likely to backfire.

I'd always thought it was interesting that the pentium FDIV bug was most easily found by code calculating twin primes. But there may be a mundane explanation for that rather than cloak and dagger stuff.

Not surprising, given how important custom ASICs are for serious codebreaking.

Reply on news.ycombinator.com