SRAM PUF Reliability Podcast - Intrinsic ID

Transcript

Milan Lazich: Thanks for joining us for another edition of Intrinsic ID’s podcast series, You Can Never Be Too Careful. I’m Milan Lazich and I’m joined by three other members of the intrinsic ID team today. On the broadcast are Pim Tuyls, co-founder and CEO of Intrinsic ID; Natalie Bijnens of Intrinsic ID’s Product Management team; and Geert-Jan Schrijen, Intrinsic ID’s CTO. Today we’ll be discussing PUF from the perspective of reliability, which is obviously an important topic when you’re talking about security.

Pim, we’re going to start with you. I saw a blog you published a while back on the reliability of SRAM PUF and in it you mentioned that reliability is something you get asked about from time to time. Why do people ask about PUF’s reliability? What are they worried about?

Pim Tuyls: So it’s indeed a question, Milan, that comes up on a regular basis. And the main worry is that since people know that a PUF is noisy, that the key that is constructed out of the PUF would not be exactly the same all the time. That’s the real worry. And that is amplified by the fact that in security, if one bit is wrong in the key, then the system does not work anymore. We would not be able to decrypt, not be able to reliably encrypt or sign messages or verify messages, and that’s clearly a problem. In security one bit off means the whole system will fail. And that’s of course, unacceptable for all the smart devices that we are in. So everybody wants to make sure that if we use this technology, is it really true that the key is always the same? And in fact, that’s exactly what we designed our system for, to make sure that it is always the same. As I usually say, to customers and partners, it will be the same in Alaska, in Phoenix, in Holland, and in 25 years from now. Always exactly the same key. And that’s because of a number of technologies that we have embedded into our IP products. One being error correction codes. So we have very sophisticated error correction codes being used in the product. You can compare the two error correction codes being used in CDs and DVDs and blue ray discs and memories to make sure that the data that is being read is always the same.

And we use these techniques now in a completely different but much more powerful way than they have ever been used before. Because we correct up to 25 percent of errors. And that is basically unseen – there are very few systems out there that can correct so many errors. And that’s because we have a very sophisticated scheme for doing that. We use a concatenation of three very specialized codes. On top of that we have techniques like soft decision information to further improve the error correction capability. And we have even included burst protection, making sure that if a number of bits next to each other fail, that then still exactly the same key will be reconstructed. So that’s an integral part of the solution, is the error correction.

But we also understand the physics of the SRAM, in particular the aging part of the SRAM. And an anti-aging measure is therefore embedded. And that’s something that Geert-Jan, our CTO, can talk about, on how that works to keep the noise even lower over time.

Geert-Jan Schrijen: Yes, there are actually several failure mechanisms in silicon that can impact the silicon over time, and these effects we have studied in depth only in the past. And it turns out that there’s in particular one effect, sort of a burn-in effect, that could cause the silicon fingerprint that we read out from the SRAM to drift away over time. Now it turns out that there’s a very good countermeasure against this burn-in effect. And this consists of writing back values into the SRAM. And by doing that, we can actually improve the quality of the PUF over time.

And this anti-aging mechanism is actually part of all our IP products. So it’s guaranteed to give longer lifetime of the PUF than the SRAM itself is.

Milan Lazich: Nathalie. Let’s talk about when devices go into space. What are the special problems that devices face in that environment?

Nathalie Bijnens: So when they go in space, you have more radiation than you have here on Earth. And then you see the nice thing of having error correction. Because we have that error correction we can correct for bit errors, and that’s an advantage that we have. And that other PUFs might not have. If you don’t have error correction, if one bit fails, your system fails.

And, let me go further into detail. So let me go to the lower technology nodes, so the more advanced technology nodes. Components are becoming smaller and the impact of radiation is harder. It’s more damaging for these technology nodes. But SRAM PUF still works. For example, for a 7-nanometer technology node, that’s not a problem. If you look, for example, at OTP, then, OTP is less reliable for the lower technology nodes. So what’s typically done, by the military, for example, for components in general, also for OTP, you go to the higher technology nodes or you stick with the larger nodes. Or, as Geert-Jan was mentioning, the SRAM PUF is more reliable than the SRAM itself. So, especially in space, components will fail first before the SRAM PUF is failing.

Milan Lazich: Let’s talk a little bit about testing. Geert-Jan, what kind of processes do we follow to ensure that the technology is reliable?

Geert-Jan Schrijen: Yeah, so for testing the algorithms in the logic and making sure that the error correction capabilities are what we expect, we do lots of simulations, simulated data. This is actually the only way to properly test this given the complex nature of the error correction code. But the second very important part of our testing has to do with the physical unclonable function itself, so the SRAM PUF. We have, over the past years, we have done a lot of qualifications on different technology nodes, on chips from different foundries, with different process optimizations, on memories from different compilers settings. And to guarantee that all the variation that we see is within the bounds that our IP is able to work with.

And actually here SRAM PUF has a big advantage because you are able to test, to qualify, the SRAM cells for PUF behavior actually on existing chips. You do not have to design in your PUF and make a new chip before you can test it. So that has been quite well to our advantage. So we have put lots of devices we could get our hands on in climate chambers to test on full pitches, different temperatures also accelerated aging experiments. And furthermore, we have done a lot of work with customers to get to actually larger number of devices. Customers typically have capabilities of testing on wafer level, which also adds to our database. So, overall, we have been able to characterize the behavior of SRAM PUF on a very large variation of conditions and devices and have been able to verify that our IP works on all of these variations, in all the circumstances.

Milan Lazich: Thank you all for joining the podcast today. Pim, do you have any final thoughts?

Pim Tuyls: Yes, so I think SRAM PUF is very successful. So far SRAM PUF has now been used in more than 170 million devices. And that is probably also largely due to the fact that it has superior characteristics when you compare it with other PUFs. When we did research in our team, we have studied many PUFs, from optical, acoustic PUFs to all kind of silicon PUFs. And we have chosen to go with the SRAM PUF because of its superior characteristics in terms of reliability, in terms of entropy, but also in terms of scalability. SRAM is basically everywhere in every node in every process, there is SRAM, and it makes it very easily accessible. And as Geert-Jan already hinted to, even if you want to evaluate or test the SRAM PUF, it’s very easy to do and it can be done at a very low cost. And in the coming quarters you will see many updates. You might have seen some in the past weeks, about customers that are implementing SRAM PUF. And the number of 170 million is expected to grow now very quickly, to several hundreds of million devices with SRAM PUF in the market.

Milan Lazich: Very good. Well, once again, thanks to our guests today – Pim Tuyls, Nathalie Bijnens and Geert-Jan Schrijen. Thank you for joining us on today’s broadcast of You Can Never Be Too Careful. And we hope you join us for our next podcast. Thank you.