StarkWare Sessions #9 | zk-Hardware with Omer of Ingonyama

Sometimes, the frontier is at a crypto conference. We’re returning from our adventures in Tel Aviv with nine exclusive interviews with some of the key players in the StarkNet space. Wish you could make it to all the crypto conferences, but don't have the time? Don't worry, Bankless brings the frontier to you. In this episode, we’re speaking with Omer Shlomovits, founder and CEO of Ingonyama, a next-generation semiconductor company, building specialized hardware for Zero-Knowledge Proofs. It’s not often that we talk hardware on Bankless, so when we do... it's time to pay attention. ------ 📣 MetaMask Learn https://bankless.cc/metamaskshow ------ 🚀 JOIN THE NATION: https://newsletter.banklesshq.com/subscribe ------ BANKLESS SPONSOR TOOLS: 🐙KRAKEN | MOST-TRUSTED CRYPTO EXCHANGE https://bankless.cc/kraken 🦄UNISWAP | ON-CHAIN MARKETPLACE https://bankless.cc/uniswap ⚖️ ARBITRUM | SCALING ETHEREUM https://bankless.cc/Arbitrum 🚁 EARNIFI | CLAIM YOUR UNCLAIMED AIRDROPS https://bankless.cc/earnifi 👻 PHANTOM | CROSS-CHAIN WALLET https://bankless.cc/phantom ------ Timestamps: 0:00 Intro 4:00 Ingonyama 7:45 Scaling and Privacy 11:00 Present and Future Computation 13:55 Hardware 19:49 Producing Chips 23:50 Get Involved ----- Resources: Omer Shlomovits https://www.omershlomovits.com/ Ingonyama https://twitter.com/Ingo_zk StarkNet https://starkware.co/starknet/ ----- Not financial or tax advice. This channel is strictly educational and is not investment advice or a solicitation to buy or sell any assets or to make any financial decisions. This video is not tax advice. Talk to your accountant. Do your own research. Disclosure. From time-to-time I may add links in this newsletter to products I use. I may receive commission if you make a purchase through one of these links. Additionally, the Bankless writers hold crypto assets. See our investment disclosures here: https://www.bankless.com/disclosures

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:02David Hoffman?

  Welcome to Bankless, where we explore the frontier of internet money and internet finance. And sometimes, that frontier is at a conference. Like last weekend, where over 1,000 developers, founders, builders, and investors attended the Starkware Sessions in Tel Aviv in order to participate in growing the Starknet ecosystem. This is Bankless' Starkware Sessions series, which are nine bite-sized episodes interviewing the founders, builders, and ecosystem developers of Starknet. Every once in a while, in the crypto world, a conference happens, but not everyone is available to attend. Don't worry, Bankless has your back, because I go to basically every conference that's out on the frontier, and I bring an entire podcast studio in tow with me, in order to make sure that the Bankless Nation stays on the frontier of what's happening in crypto. In this interview,

• 0:46David Hoffman?

  we are talking to Ingonyama, Omer from Ingonyama. Ingonyama is an interesting product. Where most things that we talk about on Bankless or on-chain, we're actually talking about hardware in this episode. ZK proofs are computationally intensive, especially for CPUs. So Ingonyama is working on building chips, which are just ASICs, first FGPAs, and then ASICs later, to help speed up the prover set of ZK proofs. And so, while ZK proofs are computationally intensive, we still want the ability for anyone to run those proofs. And so, the bull case for Ingonyama is to put a bunch of ZK prover chips in all of the hardware and devices that we use in this world. Disclosure, Ryan and I are

• 1:30David Hoffman?

  actual investors into Ingonyama, so we have exposure to what Omer is building here. But the thesis is that self-sovereign ZK technologies are going to be so proliferated, and we need ZK provers to be equally as democratized and accessible, so that individuals can confirm their own ZK proofs, that the fact that ZK rollups and other ZK technologies are valid. When we have chips that are specialized for this, we can extend this out to more and more people. Highly aligned with a bankless philosophy. So that's why we are investors. I hope you enjoy this interview with Omer. But first, a moment to talk to some of these fantastic sponsors that make the show possible. Kraken has been around for almost as long as crypto itself. As crazy as crypto is, Kraken has remained a rock for its clients,

• 2:13David Hoffman?

  building out a platform designed to stand the test of time. Kraken is all about proof, not promises, which is why they've pioneered a proof of reserves, a system that lets Kraken's customers verify for themselves the solvency and responsibility of the Kraken exchange. And for this reason, and many more, we've chosen Kraken as Bankless's strategic sponsor for 2023. And once you're done using Kraken to get your dirty fiat money into pristine crypto assets, perhaps you should check out Arbitrum, where so much of the frontier of crypto innovation is taking place. Arbitrum just announced their Stylist initiative, which is their brand-new programming environment that lets developers deploy apps using their favorite programming languages like Rust, C, C++ alongside regular old solidity EVM apps. That's can use any coding language they like,

• 2:56David Hoffman?

  and it simultaneously drops Layer 2 fees by an order of magnitude. It's a big deal. So many of your favorite DeFi apps have already deployed to Arbitrum, like Uniswap. But did you know that Uniswap isn't just for ERC2 tokens anymore? Uniswap is also an NFT aggregator too, to make sure that you are getting the best price on your NFC purchases. Uniswap gives NFT collectors and traders powerful tools, all from the cozy, comfy Uniswap interface. By the way, it's also got a native fiat on-ramp. It's also deployed on like every Layer 2, and it's got this nifty cool thing called a universal router that ensures that you always get the best price on your trades. Crypto protocols like Arbitrum and Uniswap are becoming super powerful. So make sure that you're also using a wallet that's equally as powerful, or else you'll be missing out. If you're looking to upgrade your wallet experience, check out Phantom, the number one wallet of Solana,

• 3:40David Hoffman?

  which has recently come to both Ethereum and Polygon. Phantom has native NFT support, giving you your own NFT gallery from inside the wallet. Phantom is both a browser extension and a mobile wallet, and is also the only wallet to span both the Solana and Ethereum ecosystems. Check it out at phantom.app. And now, let's get into the interview. Bankless Nation, once again, we are here at the Starkware Sessions in Tel Aviv, and I'm here with Omer of Ingonyama. And Ingonyama is doing something quite interesting that you might not be familiar with. Omer, can you kind of walk us through why we need Ingonyama?

• 4:11Speaker 01

  So, like every successful technology in history, we can say that there are three pillars to this technology. We have the software, the algorithms, and the hardware. So, zero knowledge, which is a core technology in our space, it's basically a cornerstone for trustless computing, which is what we need for decentralization. It's such, it's just such a technology, and we want to make it successful. We want to make it mainstream. We want to make it at the hands of everybody. So, out of these three pillars, basically what we do is we are taking care of the hardware.

• 4:48David Hoffman?

  And so, this is the hardware, as in, once upon a time, the Bankless Nation will know, I got into the world of Ethereum through GPU mining. It's also the world of ASICs proof of work. How does, and this is the same sort of subject matter, but as it relates to zero knowledge computation. How does this, how is this world similar to the proof-of-work industry? And how is it also different?

• 5:08Speaker 01

  Right, so, dynamics can be very, there are similarities, but I think one step before, we need to understand what exactly is the connection, or what do we need hardware for zero knowledge? And specifically, you know, zero knowledge, it's been around like decades in academia, and only recently, over the last three years, it's doing the transition into industry. And we chose, in the space, with our representative, represented by many companies, what we chose is to focus on very specific type of zero knowledge provers. You might have heard about Snarks, Starks, and companies such as zero knowledge rollups, right? And basically, they have some very interesting properties, this type of zero knowledge systems.

• 5:53Speaker 01

  One of them is that the proof is very small. And the other thing is that it's publicly verifiable, okay? So, before we even go into explaining what zero knowledge is, and why do we need it, just in terms of like, the analogy here, to prove a fork, what we see is that there is Kind of like a trade off because we gain all of this like very nice properties. But we need to pay somewhere and where we need to pay is by computing this proof. So it roughly takes seven orders of magnitude more than the computation itself to actually compute the proof. Once you have the proof, you have a lot of power, right? But to get to this point, you need to spend a lot of money and compute power. This is where hardware can um can make a big difference.

• 6:36Speaker 01

  You know, it's still going to be seven orders of magnitude, more computation. But the fact that you or what you feel in terms of um uh the cost. And uh and and the time that it takes you, we aim to make it as uh as simple as possible, obviously. But also as close as it's possible to the user experience and developer experience that we have by just running the computation. Now, what it means is that, you know, as I said, trustless computing is zero knowledge can uh can mean many things. There are definitely few applications that we see today that are touching or try to replace exactly uh the old mechanism of proof of work by something that's based on zero knowledge.

• 7:19Speaker 01

  There are huge advantages of doing so, but this is far from being the only use case. So I think that right now we are talking about level that is way uh lower, somewhat lower than uh the actual proof of work uh type of uh or mining that we had in the space. But it's definitely involved. It has some kind of uh relations to it. For example, you can build something similar to a proof of work mechanism based on zero knowledge. Of course, with many benefits that I I can go over.

• 7:48David Hoffman?

  So it's been part of the trajectory of much of Ethereum to actually get rid of hardware. Like we want to get proof of state because we don't have, we don't want to have proof of work miners. But there's a dynamic here about zero knowledge computations, which also is part of the broad Ethereum roadmap where we want like ZK rollups with zero knowledge proofs and other reasons to apply zero knowledge technology to what we're doing here. Which means that are we committing to some sort of hardware layer to the Ethereum future? Because of just the the commitment to zero knowledge?

• 8:23Speaker 01

  Yeah, so that's a good question. And I think that to understand it or to answer it first, I do need to explain why ZK, Zero Knowledge, is useful in the context of Ethereum. So let me just take two like big ideas. One is scaling, the other is privacy. Now, in scaling, what Zero Knowledge basically is doing is kind of compressing, right? I mean, just like trying to simplify it a bit. We try to compress many transactions that we just don't have enough room to put them on chain. So we put only the proof. And it's very fiber computation. Everything should be fine. In a way, we are saving a lot of money by doing so, right?

• 9:05Speaker 01

  The end goal is to have it in a full decentralized manner, right? You don't want people to actually or to create this kind of market around or this kind of dynamics where you actually need or it's open only for certain type of actors. You want to enable anybody to obviously verify a proof. This is like definitely super cheap and efficient. As I said, it's the properties of these like snark stocks that we chose. But you also want the proving and computing the proof to be done basically by anyone on even commodity hardware or close to it such as GPUs. So this is just, when it comes to scaling, I don't think that we are committing in a way to a future where we actually need to pay more

• 9:50Speaker 01

  or to consume much more electricity. I think it would save us much more than than in terms of when we try to compare it to what we know until now. Privacy, this is, you know, this is like an application. This is like another thing that we can gain that was not accessible to us before. Privacy at scale, privacy on chain, privacy as a layer two. I think it's kind of an application that Zero Knowledge is bringing. And of course, you need to consider the cost of such an application. Personally, I think it's a rightful cause and would pave for, you know, a future that is really great. And it's worth to kind of put some effort into it in terms of computation.

• 10:35Speaker 01

  Long-term thinking. Again, this hardware is going to be in your, I don't know, your iPhone. So part of the chip would run ZK. It would be seamless, right? Uh, it would be very efficient and so on. So again, I don't think it's going to go into committing to, uh, where we were going for, where we were coming from and, and, uh, you know, rowing in this proof of fork to proof of stake, um, transition.

• 11:00David Hoffman?

  So, um, the idea that, that I have for the, the future of Ethereum and actually even baked into the Ethereum roadmap is to ZK the whole entire blockchain. But then there's like applications that can use ZK tough and there's other layer twos that can use ZK stuff. All of this, all, anytime a zero knowledge is executed upon, that's a computation that needs to be run. Where are we today? Like, how is that computation being done today? And with the, the future that you have in mind, how will it be done in the future?

• 11:30Speaker 01

  Right. Um, so yeah, that's true. ZK, um, is very powerful and it can find, uh, use cases in many places. And obviously all of layer two, ZK rollups. And I think also at some point, um, optimistics or optimistic rollups would, um, would start using this technology to some extent. Uh, so we are talking, if you try to sketch like a graph of the number of proofs per unit of time. So what I think would see is that now it's growing and it's keep going, uh, which is, which is very interesting. And, and at some point, you know, it will consume kind of a lot of, a lot of electricity. As you said, any application that is building, being built on top of a rollup eventually

• 12:12Speaker 01

  would require to, um, verify it on chain and it means another proof that needs to run. Not only that, you want, you know, for, um, uh, you want to often to run the same proof more than once so that in case something happens to one of the prover, you have like a backup, right? Um, so it can, it can scale definitely. There are other type of, um, uh, ideas that are counter to that scaling, which are, for example, are kind of packing more than one proof, uh, together. So because basically you are proving a computation, you can also prove. Another proof, right? So you can just stack them up in some kind of recursive manner. And this kind of saves you a lot of the trouble.

• 12:56Speaker 01

  This is a lot of what we see today in ZK rollups. Um, how they moved like StarkWare, for example, moved into uh using recursion, which allows you not only to stack proofs one onto another and eventually end up just with one proof. It also makes these proofs way smaller, right? Because you can just like aggregate them together. And so this is kind of counter trend to the growing need in um in proofs, right? What will happen is that with the introduction of hardware, getting these proofs gonna be the the the the threshold or the barrier to get this like very complex proof uh in computation will just get lower and lower. And usually, just like with other inventions, what it means is that there would be more and more use cases using these proofs.

• 13:39Speaker 01

  So I think that overall, uh we're just seeing increase in the number of proofs, but definitely it would be uh contained. It would be contained in the hardware level. It would be contained in the algorithmic level in a way that will be manageable. And for us as a community, it would uh eventually be a net positive.

• 13:57David Hoffman?

  So to use a metaphor, would you say it's fair to say that Ingo Yama is trying to build the engines to run zero knowledge computation? Like you're building the combustion engine for zero knowledge?

• 14:10Speaker 01

  Uh, yeah, kind of. I mean, we are building computers. Like, you know, you can look at the uh traditional computer architecture and something is missing. I mean, something that like, you know, a finite field arithmetic, some discrete math, like the stuff that that are fundamental in the computer uh architecture is is not really accelerated. It's not used in in AI, network, storage, acceleration. But it is needed in cryptography, even not just zero knowledge. Folliomorphic encryption, um and and even, you know, lattices and such, like basic, the basics of of post quantum cryptography. Everything is built on the same math. We are just trying to bring a modern type of computer architecture that will support natively this type of um uh computations.

• 14:51David Hoffman?

  So is the idea that what you just said that we have this new thing in this world called zero knowledge proofs, they're computationally intensive, is a future state of the world, one where there is a new component, a new hardware component in perhaps every single device out there, my phone, my computer, my laptop, that is specialized to run a zero knowledge proof? Is that is that kind of the future that we're looking at?

• 15:16Speaker 01

  Yeah, absolutely. I mean, and we also can can tell, you know, it's going to start from more of the server side, like the data centers. This is where it's easier to kind of take a whole new computer architecture and just like push it and then get some get some optimizations in running ZK provers. But, you know, phones, they are like very, the phone factor very small, it's going to be very hard. But let's say decentralized identity, all right, this is something that we know several like phone vendors, manufacturers are looking very deeply into. So this will require some of these functions, like so many pickles that are relevant to their border, like ZK computation, but they can also fit inside like chips that are being manufactured by others. Also, you know, when it comes

• 16:00Speaker 01

  to privacy, eventually you need it in your hand, you need you need full control. So that's definitely going to be like in every one of our hands in phones and so on. Yeah.

• 16:08David Hoffman?

  Yeah. So I know one of the big conversations that's going out here at the StarkWare Sessions is account abstraction and how account abstraction uses the secure enclave of a phone to store data, to unlock, you know, assurances about a private key. And so there's already the conversation happening about how does our physical hardware impact the technology that we have in the virtual space, right? Like, even though we're all love to be in the metaverse, the physical nature of our hardware is still very, very important. And so maybe another to use another metaphor is like we have this secure enclave to store our private keys, and that's a physical component of our devices that help us do our crypto things. And then what you're perhaps envisioning is that there's also going to

• 16:50David Hoffman?

  be another additional part of a hardware that is meant to do zero knowledge computation to allow for zero knowledge computation to be done by the individual on their device. So is that all line up with you?

• 17:02Speaker 01

  Yeah, I think I think it is. Like, when I look at the future in a world where ZK is cheap, I mean, there's no reason why you cannot just Verify any computation you do. I mean, the entire, everything you are on the phone, including communication with secure enclave and so on, can be verifiable, right? You can also output a proof that the computation was done correctly. It's even more important once you outsource the computation to someone else. I don't know, to the cloud. Everything can be verifiable and it should, right? So I think, yeah, I think we are talking about kind of a new, modern type of architecture that will also include a component that is specialized for this type of cryptography.

• 17:45David Hoffman?

  Yeah, and, and if we don't have this kind of research and this kind of like R and D effort that you're, that you're, you're doing, then we probably ultimately outsource that proof generation to a third party, which is not exactly what we're going for in this world of self-sovereign like technology. Is that, is that a fair assumption?

• 18:03Speaker 01

  Yeah, it is. I mean, there's a ongoing research on how you can actually separate the compute-intensive part from the privacy-preserving part of the computation. It's not easy, there are a few ideas. And it's, you know, right now it seems very far away. So I think definitely a way to go is to try and get everything done on the hardware locally. And it's a future, it will take a lot of time, but we'll get there. Meanwhile, there are many use cases and applications that can still run only on data centers by miners and, you know, some people and corporates that have access to this like, to this like specialized hardware. That's, that's for sure. But yeah, we are going into the, this like future

• 18:46Speaker 01

  where there's ASICs that are like different chips that can be used in any type of computing device and can bring you this technology to the level that again, user experience is going to be completely seamless. You won't feel that you are computing a proof.

• 19:02David Hoffman?

  So maybe we can really judge about how trustless and self-sovereign the world is with how many, how far your chips have proliferated across the world.

• 19:12Speaker 01

  I mean, it's one way to look at it. Again, remember that right now, even the rollups, the Zicke rollups, all of them are more or less centralized.

• 19:22David Hoffman?

  Sure.

• 19:23Speaker 01

  So there's still a far way to go. We still need to, and you know, we are a hardware place. So we are long-term thinkers. This is how we are building the company. Yeah, I mean, at some point, I do hope that many of us would have this freedom of just, you know, verifiable and private type of computation, privacy preserving computation, verifiable computation. It will take time until it will get basically to be democratized and be by the hands of anybody.

• 19:50David Hoffman?

  Sure. So the topic of hardware is not one that we've actually talked a lot about on Bankless. And so this is probably striking the imaginations of a set of Bankless listeners that doesn't previously get a bunch of attention. Walk us through the roadmap for like how a chip actually gets produced? And what is the roadmap for Ingonyama?

• 20:08Speaker 01

  Right. So the first thing to understand about us is that we are not religious about hardware. Meaning that, you know, there are a few types of specialized hardware with differences. I can name a few, FPGA's, GPUs, and of course ASICs, like this specialized chips. For us, it's about finding where is the pain and then solving it. And when I say solve it, it's about catering for the little guy. Again, we want to make the user experience seamless. That's the goal, that's the ideal. And what it means is that right now, in terms of roadmap,

• 20:46Speaker 01

  we are doing a lot of R and D internally to kind of build our hardware and our IP to run everything. What we want to do is basically run everything inside of this specialized hardware. And then we try to feed the right solution to the right problem, right? So you can think about, let's take Starkware, right? We are in a Starkware event. Let's take Starkware, for example. So for Starkware, right now, the prover is centralized. It's one problem. It's not the most important problem they have. They have also a problem with witness generation, with sequencer. They pay much more on getting the transaction on-chain for fees than they pay on running the prover. So when we're talking with Starkware and we want to develop hardware,

• 21:30Speaker 01

  what we basically try to do at this point is just to move the bottleneck from the prover to somewhere else in the system, okay? Now, at some point, these other places in the system would also have their bottleneck removed. The only bottleneck that I predict will remain would be the hardware for ZK provers and maybe, you know, a bit beyond. But right now, it's just about being efficient enough and cost-effective enough to move this bottleneck someplace else in the system. That's one example. Obviously, we have projects where we want to optimize on the throughput, right? We want to run as many proofs at the same time. And we don't really care about the cost, let's say. There are other projects where we care just about the form factor. We have a project, by the way, for Filecoin,

• 22:13Speaker 01

  where it's mostly about just replacing GPUs with FPGAs that are tiny, right? It's like small, it fits a data center. So it really depends. Overall, our vision is, and I think this is the long-term solution, is to have these dedicated chips. And to go to this place, we need a lot of, we first need to come up with an architecture, something very similar to a GPU, if you try to think about it, like many, many cores that run in parallel, but are suitable to the type of computation that you can see in Zero Knowledge. And once we have this figure out, remember that ZK, and this is probably a question many ask, is it's a fast-moving space, right? Everything is kind of changing. How do we de-risk it when we build something that is meant to last for many years to come?

• 22:55Speaker 01

  And one way that we try to do it is by basically billing it in a programmable way. That's one thing. And the other thing is to battle-test it all the time, right? So every time there's a new ZK protocol primitive coming out from academia even, we try to kind of retim it and see if we can support it and to what extent. And then we fine-tune our design. Hopefully, we'll be able to go to a tape-out, which is... To your question, kind of the no return point, where you kind of say, okay, this goes and when it gets back, it's in silicon, then it goes into, I don't know, miners and whoever needs it. So hopefully we'll be able to go to this, to tape out soon, meaning when I'm saying soon, it still mean like, you know, can be in a couple of years.

• 23:39Speaker 01

  And it's really gonna depend on how this market is gonna evolve. We first want to target rollups, ZK rollups, and therefore we are very much dependent on their decentralization efforts and we try our best also to, to help with what we can.

• 23:52David Hoffman?

  Omar, if somebody is interested by what you are doing, what you've talked about here today, how can they join? How can they help? What, who do you need help from? Or how can people get involved?

• 24:03Speaker 01

  Right, so at this point, we do mostly R and D. My team, they are a bunch of very talented cryptographers, mathematicians, and hardware guys, like chip designers, architects, and so on. We are trying to build an ecosystem in software. What we're gonna do soon is gonna open source much of what we do. Hardware is traditionally a very closed source type of environment. This is not good for our space. We try to break this narrative, we try to go and be and walk in the open, be transparent, and also put some code and IP out there for people to use and play. Feedback is critical from developers mostly, that are gonna build on ZK. Any type of application, everything is applicable.

• 24:47Speaker 01

  You can find us in Twitter, @ingoZK. We have a website, we launched not long ago, ingonyama.com. We have a GitHub with all of our stuff open. We have a Discord that soon we're gonna launch it officially. Right now, if you are working on some technical problem and want to join the Discord, meet the team, this is a very good place. Reach out to me and I'll be able to connect you with the team.

• 25:10David Hoffman?

  Omar, thank you so much.

• 25:12Speaker 01

  Thank you.