Latest Developments in Open Silicon

Transcript

[music] [music] >> Hello. Um so uh recently Vitalik coined this term crops, sensor resistant, capture resistant, open source, private, secure. And Ethereum excels in these properties, but we often forget that the hardware upon which that software runs does not. It's oligarch dominated, closed source, opaque, vulnerable.

Uh it's funny that the upper layer gets all the attention and the place where the problem actually is is kind of ignored. You know, maybe it's too hard, too opaque. Um I would point out as an old person, I can tell you open source software was not always a thing. When commercial software started, it was overwhelmingly proprietary. In fact, the term open source software was was coined in 1998. For me, that's recent. Uh today vast majority of commercial software is majority open source. Uh but not so with chips. Chips have are still dominated by closed source design.

And among chips, secure elements are the closed of the closed. Uh a lot of people aren't aware, you know, you see EAL certification under common criteria. Uh that not only tolerates but actually requires security. This is a quote directly from the uh certification document says this is in contrast to Kirchhoff's principle. They say in multiple places in the document, we do not follow Kirchhoff's principles. You are not allowed to disclose things. [snorts] Uh it seems very odd that then this forms the fundamental root of trust for the entire wallet system from which we drive inputs into uh into Ethereum. You know, in the normal world, you'd have uh if there's a vulnerability, you'd go

into CVE database. Uh everybody would work collaboratively on it, know about it. Uh in secure elements, typically it's a customer discloses it, they're reminded that they have an NDA obligation. So, as an end user, you know their vulnerabilities, you don't know how many they are, when and if they're fixed, how many people know about them, who those people are. You know none of this. That's a strange foundation for uh for root of trust. Um and it's not just me who has thinking this. You know Vitalik started talking about this more than a year ago, um probably more eloquently than I'm able to do so and you know basically saying open and

verifiable hardware is a precondition. You build trust from the bottom up. So, it makes no sense to have trustworthy software on top of untrustworthy hardware. Uh and this is not just hypothetical. Many things have happened in the world where hardware bugs cause major loss, major problems. Uh I am going to by the way publish these, so I'm I'm flipping through, but you can you can read offline. Um I'll just jump down to the last one and point out that because this is kind of relevant to wallets a lot.

Um a lot of wallets say they're open source, but that's just their own code. Uh there would be microcode, security setup code and things that come from the chip vendor and those are generally closed source, not revealed even to that device vendor. So, this is a case where the firmware uh the the vendor selling the device didn't have access to the firmware in which there was a vulnerability. So, if this is such a problem, why don't we just solve the problem? Well, uh there's a lot of problems with solving the problem.

Uh I won't go through all of these. I'm just going to deep dive on a a few of them. Uh one of them is certification. Uh I would remind people that uh you know today we take for granted um that whoops. Today we take for granted that software libraries, algorithms should be open source. Of course, it's very natural that, for example, NIST PQC algorithms, everybody worldwide can can hack on them, try to break them, and so on.

We all think of this as very natural. The world did not start that way. That's a relatively new development. Uh I personally had copied some of these documents in bold, like the GSMA 5 software uh cipher. I personally had copies of that. I was required to lock in a safe anytime I wasn't actually reading it. And so that was what people leaned towards in algorithms initially. Well, those were reverse engineered because they had been developed in secrecy, they weren't that strong. Over and over this happened uh until people started to realize, well, maybe this is a stupid idea, right? The So a lot of younger people don't know it it wasn't always this way. You know, we started with obscurity.

So why do we still have security in Silicon? Well, a big reason is certification, regulatory capture. Uh whatever the shortcomings of the certification process, it is baked into regulation. It is baked into marketing. Uh you're not actually prohibited from certifying open source, but there there's a point system. Knowledge of the target of evaluation, access to the target of evaluation are major point deductions. It's essentially you you can't certify open source. Well, logically there should be some world in which there's a credit for open source. I always think of this like if you do mountain climbing, you're advised, don't try to find that one infallible anchor. You'll never find it.

You should have multiple anchors that look pretty good, right? And this mathematics of diversity is is used in all kinds of industries. It's used in your cell phone to have multiple antennas and so on. Um so obscurity versus transparency, perhaps uh there is some valid place for each one. Uh but AI is hugely disruptive and disrupts that balance. Uh on the demand side, I think all of us, whether uh enterprise or personally, are a little bit freaked out by the power of AI and then you know what is the chip really doing? Who has our data?

Um, but also on the capability side, fundamentally AI diminishes the effectiveness of secrecy. AI increases the effectiveness of transparency. Uh, I would point in the references to some papers by an economist, Peter Swire, who kind of tried to mathematically model as economists do, when does disclosure help security? You read through that, you can see that AI fundamentally shifts all of the inputs into his mathematical formulas. Uh, on this slide I would just point out the the part in red that we're talking about not only the thing under evaluation being open, but also the evaluation itself and that's tremendously valuable. That means that

evaluators can duplicate, challenge, improve upon other evaluations. Um, uh, apart from just security audit, open source also enables new techniques in terms of uh, in terms of security. The disclosure of the design would allow you to do uh, empirical or logical formal verification of the execution. Uh, if you go one step down to the net list, the kind of sea of gates um you can instead of just randomly, you know, trying to inject a fault, aiming a laser around a chip, you can, for example, say what if I could flip any one of these billion gates? Would any of them result in the disclosure of

a secret? You can start to do uh, injection attacks kind of virtually. Um, and we're doing this. We're uh, disclosing to a small set uh, a actually net list level um, which requires some redaction before you do that, but uh, it it's powerful. So um, another observation here is academia versus industry. I've been in this open silicon community a little little while and I've noticed a kind of uh, divide in the community. Uh, open permissions chip design flows like academic flows A are not very production worthy.

Production worthy flows like what I'm used to are not very open and permissionless. So an academic would think, oh let's start with these open flows and try to make them better. Okay? Somebody like me from industry would think, let's start with a real flow that we know can produce chips. Let's try to make it a little more open. Uh these are fundamentally different points of view and I I see people having trouble seeing eye to eye on this. Kind of a divide in the community you could say. From my point of view neither one is right or wrong. I think they serve different purposes and they're both individually useful.

Um I always use the analogy of of a 3D printer versus a LEGO factory. Uh either one of these can make an ABS plastic block, right? But the 3D printer does not allow you to be in the LEGO brick business. Uh and most importantly these things are so different that they're never going to become the same by gradual changes. And I see some people think, oh we'll just unify everything. Everything will be open and it'll be producible and so on. Uh I don't think so. I think these are fundamentally different flows for different purposes. And I noticed that all of the people who think it's that simple have never run a chip business.

Okay? Um last on inspectability. One of the most obvious you know things you would come to when you think about open source silicon is inspectability. With with software you can build it yourself. Open source tools you can match the end product to the source. With chips there is some design material at the front end but then you go through closed source tools, closed source IP, closed source manufacturing. In the end you get a little black rectangle. You have no idea if it actually matches the source.

So if you stop by our little booth over there in the Jasmine Hall we can demonstrate an infrared inspection of the finished goods. So you're trying to then correspond uh what you have in your hand back to the source. Uh so I would draw it on this like this where we use the close source flow and then we publish image of the layout. Now you can look at the chip. Again we'll demonstrate it over there and you can see that they match, right? So this is a huge improvement but it's not really trustless because we're the one providing that that image.

So how do you improve on that? Academics would start with the point of view, oh let's just make everything open. We'll have an open factory, open tools, blah blah blah. Problem solved, right? Um I think that isn't super practical. One is that it assumes that you can create an open flow. You know in major nation states, China, US, India, all the EU are trying to do that and failing. Nobody is close to TSMC. So it's an ambitious undertaking.

Also the those design materials like those photographic plates, that's not the bottom of the tech stack. There's also then materials, process, machines. There's a paper from 2013 about doping concentration can undermine a TRNG. So but most problematic I think even if you could do that thing which I think is impossible you would only have one factory. So this is not capture resistant. That is capture. And in business foundry portability is is everything. You can't have you know just to be dependent on one foundry. So I think there's a way forward to go ahead and use the close source flow but

export artifacts from that that can be then verified with with open source tools. Obviously creating those artifacts is much more complex than verifying them. I think the open source tools can do the verification. So I'll briefly jump around some of the happenings in the world. So risk five instruction set architecture is proliferating. Most implementations are closed source, but open source implementations are getting better. Archip uses the VexRisc. If you look at things like peripheral IP, there's a ton of you can find a million different UART Hard to find one that actually works well, but you know, there's there's IP out there that's usable.

In cryptographic engines, there's open Titan derived engines like Pavona and Caliptra. I think very good. There's this little Sec Works engine. There's our engine that's in open source and then there's Tropic Square. Bus fabric kind of like the nervous system of the chip. There's some I think producible bus fabrics. But everything I've talked about so far is like this is our chip layout. Everything I've talked about so far is the colored part of that, like 25% of the area. People forget about the hard parts, the memories, the IOs, the PLL, the power management, all these kind of things.

There are some nascent open source efforts in those, but it's fairly sad, I would say. The difficult parts is lagging far behind the logic. I'll skip over, you know, EDA flow and open manufacturing. Like I mentioned, there are some. There are a bunch of stewardship organization advocacy organizations. I want to shout out to a couple that I met here. Foss Asia, they're running the open hardware hub here on the level three. Also ran into opnsilicon.sg, a Singapore group. They're running an event in November.

There's a bunch of kind of sovereignty driven efforts, you know, like I've mentioned China, India, EU, US trying to create self-sufficiency in fabs. One group that that tends to hang out sometimes, they're not here at Devcon, but they attend a lot of the Ethereum events. ML Pseudo, Quintus Kilbourne, you may know know them. I like what they do. They do some really insightful education and presentation of the problem and solutions and so on. And then of course Vitalik has been blogging about this and talking about it.

Given all this going on though, it's kind of odd that there aren't more examples in the wild. If you go to chips in production in the wild, I only know of these three examples, Tropic Square, Nuvoton, and Crossbar. So fairly slim pickings given all the activity and discussion of open silicon. So to introduce our company, we make a very advanced microcontroller. There's no chip like this in open source. It's the only one. Very high power microcontroller with a security engine.

You can hear more about it at our booth. We produce several packages, several hardware form factors. I would call attention to this Ostro card, which is I would say the only open source card in the world because it's from silicon up. We were in the Devcon 34 badge in Las Vegas this year. That led to a ton of actual open source disclosures of vulnerabilities that were fixed in the boot code. So you could see the open source project process working. Also Bow Chip has a reproducible side channel board.

This is the beginning of the uncommon criteria process of trying to do open source certification. So I cite in the references all of the the GitHub for these. Also I'll mention some of the things going on here at Devcon. Or like I mentioned at our booth and on this level we can show optical inspection. In this room at 2:00 p.m. Nicholas Bach will be talking about the universal signer. It's a clear signing wallet that runs on our chip and other chips.

There's a hackathon that we're sponsoring. You can hear about it at our booth running through mid December and launching a MPC wallet on on that card as well. So I'll close by talking about what I think is the future of open silicon. Open source evangelists are actually a small market. Crypto itself is a small market, and open Silicon is a manufacturing business with expensive tooling. This implies that for Open Silicon to thrive, it has to be the best solution for a broad class of of problems, not just freedom of evangelists.

It has to also be outside of crypto. It has to be broad market. And it has to preserve competitive advantage. People have to make money or they they won't do it. But I think the time is right for this to happen. And again, AI is the disruptor. So AI design flows can use radical disclosure, disclosure that would be useless to a human because it's far too detailed. The software engineers don't how to know how to read the chip design materials. Those levels of disclosure are extremely useful in in a AI coding flow.

I cite an example here or something just just, you know, disclosing the design of the chip enables radical innovation. Traditional software flows on a closed source chip rely on a kind of long, laborious, error-prone flow. The ancillary materials are very all the, you know, very accurate documentation is and app notes and so on. That's very difficult to create. And that kind of favors the incumbents for whom that kind of forms a competitive mode.

But disclosure enables a new AI-native design flow. And I think that it will become the path of least resistance to get to wherever you want to go, not just for security. It's you know, broader than that. Whether you care about openness for its own sake or not, it's just a better design flow. So is it evangelism or good business? Well, obviously there's enthusiasm for verifiable Silicon among those who value autonomy and so on. But I think it's a green field business opportunity. Market stats are already shifting. Risk five is growing much faster than the general MCU market. What we're targeting is the broad MCU market, this $35 $40 billion market with a high

CAGR. Uh security MCU is a huge uh huge fraction of that. Uh that's the kind of market that can support chip of manufacturing. Crypto itself is just a tiny tiny part of that. Um also a number of people ask, well, if you disclose the chip, doesn't that mean you're giving up a competitive advantage? Isn't this a isn't this a business problem? Uh but it's a manufactured good. I I use the example, you look at the shirt you're wearing, it has a little tag on there and it tells you the fibers that are in there. If you know how to sew, you could probably make a reasonable replica of that shirt, but you can't be in the shirt business unless you have huge

amount of capital, deep industry knowledge, you know, processes and all these things. So the level of disclosure that's useful to you to know how to launder your shirts that's useful to the user is useless in helping them be in the shirt business. Chip design material is very much like that. The chip design materials you can use in an AI design floor very helpful, but they do not decrease the competitive mode. This is completely different from software where if you can see it, you can build it and you can make infinite identical copies. So I'll stop with that and uh we'll publish this with a bunch of references and so on you can read over. So thank you.

>> All right, ladies and gentlemen, come on, let's give Dr. Mark a big round of applause. Thank you, Doctor. I love the analogies as well. That was very nice uh helping us understand, especially for people like me. I'm not from a tech background, but understanding now I understand, right, a 3D printer versus a LEGO building machine. Okay, uh Doctor, I think we have some time for Q&A. Do you have time for Q&A? Yep, yep, yep, yep, yep. Would you guys like to ask some questions? You already have been submitting some of your questions on the QR code over there. Thank you very much. Doctor, we've got some great questions coming in. Let's see how many we can tackle in a few minutes, okay? First one right at the top there, quick one. This one got a lot of upvotes. Did you meet the freedom

factory people making the PQ one, the open hardware, open software post quantum hardware wallet? Yes, so you can go ahead to the microphone over there at the podium. >> Yeah. >> That's fine. >> Put that on my to-do list. >> Okay, [laughter] on your on your to-do list, right? >> yeah. Um the second one, you know, can we call it open source as a design? >> Okay, hold on. One second, one second. I'll answer one at a time so that everybody can follow along. Okay, the first one, it's on your to-do list. Let's look look at the second one, okay? Let's let me read it out first for our friends online as well. Can we genuinely call silicon open source when the design is open but the fabrication process remains controlled by a few corporations? Let's let's answer to the audience yeah.

>> Yeah, I that's um you know, like I talked about it would be great if you could have everything open and make a business of it, but I think we have to take what we can get. It's not great that TSMC is so far ahead of everybody in the world that there's kind of like one factory serving the world. That's not ideal, but I think that's what we have to deal with. >> what we can get when we have it, right? Yeah. Okay, great. Let's let me just click that as answer. Let's move on to a few more. Okay, why can't open source be certified similar to something like ISO standards for various industries? Seems surprising that nobody has done this.

>> No, absolutely. I think it should be and can be. Like I said, there's a community developing around this term uncommon criteria. It's just the regulatory capture of the closed source method is very strong market and regulation. >> Yeah, so there are challenges, but it's nice to know that you showed us quite a lot of uh companies and groups of peoples working working on it, right? It is a problem that they're working on. That's good. How can we protect Oh, this is good. How can we protect ourselves from hardware hijacking like the recent cold wallet spyware injections? Now, this caused quite a stir online. I'm sure you've seen it. I'm sure all of you are aware of it. What can we do as regular >> Yeah, that's that's kind of like a popular topic now. So what we're showing over at this Jasmine concourse, you can

examine at least chip level in injection of a of a Trojan and get some confidence there. But at this level, device level, I think you have to rely on diversity. You know, you can use all the common techniques, but you also have to have diversity. Like we're floating an MPC wallet, for example. >> Right. >> Decentralized entropy generation, decentralized signing. You would have to get multiple Trojans in multiple places >> Right. >> to So I think, you know, architecturally >> Okay. Security and diversity, that's that's a nice thought to think about. I I like the fact that everybody online became an expert in opening their own hardware wallets and they were cracking it open and stuff like that. That was quite cool to see, but let's move on.

Okay. Does working on Open Silicon require expensive equipment? What's the best part to get into this? >> Oh, absolutely not. I think one of the best ways to participate For example, you can stop by again. I keep pitching them but to work on We have a one of our partners actually supports a dev kit. Developing applications and tools around this open source chip and dev kit is extremely useful. You can do that on your own. Like our hackathon that we're sponsoring, you know, participate in that. You don't need anything but a PC.

And I think creating AI native design tools is actually super useful. >> fantastic. And you know, there are a lot of actually some tables at the Jasmine Hall where you can check out some of this happening in action. Next question. What will drive the most demand for Open Silicon in the next 10 years? BCI hardware perhaps? >> I I You know, as I said at the end of my talk, I I feel like you have to just look at the general MCU market. And that especially security MCU, but not limited to that. I think you will see open AI native design flows based on disclosed designs being the fastest way to design with a microcontroller. It's that simple.

>> Right, right. The The of design is changing with AI. It's not easy for anyone to copy verifiable hardware, but what if they're a big manufacturing company? I think you may have touched on this just a little bit. >> Well, yeah, I mean, I we did have some discussion about open sourcing parts of our design, and we came to the conclusion that this is the the way of the future. Yeah, other we disclose it under a license where other people can use it commercially. We kind of welcome that because it makes it For example, you look at these open Titan derived engines. The more people use them, the more valuable they are.

The more valuable they are on our chip. >> Exactly. I think last question to wrap it up. Do you envision a future, Dr. Mark, where everyone has to build their own chips and hardware at home if we truly want to be safe? >> No, but I would say, you know, like this inspection that we're demonstrating, it's very accessible and cheap. I don't think you would build it in your house, but if there's a maker community in every city, maybe every electronics store has an inspection set up like this, you know, you can start to get it more accessible. >> Awesome. Be a part of a community that cares about your security and privacy.