The Route to Mainstream Adoption: Encrypted Blockchain & FHE Acceleration Hardware

Transcript

[Music] Hello everybody. So hopefully you can hear me. I'm Nick Mu. I'm CEO of company called Optalysis from the UK. Uh we're here today because we believe in the blockchain's potential to transform the way people interact with the web. For it to reach true mainstream adoption, we have to solve the challenges around privacy. At Optalysis, we're building the hardware acceleration layers that solves that problem and unlocks the future of blockchain and web 3.

So one of the key features of blockchain is its transparency. Um this has also become its biggest hurdle. So data that is publicly visible on the chain is no good for commercially uh sensitive data being placed on there. Enterprise organizations cannot deploy applications uh and in DeFi users are losing billions from um attacks such as uh MEV and front running issues, sandwich attacks because the transactions are always public. As Jeremy Alair of Circle put it, the biggest barrier to widespread enterprise adoption of blockchain is privacy.

To on board the next billion users and unlock trillions in value, we have to move beyond a fully transparent model. We need practical onchain privacy. The solution for this is a set of cryptographic schemes known under the banner as fully homorphic encryption or FHE. Where conventional cryptography allows data to be protected whilst at rest or in transit, FHE is fundamentally different. It provides uh the ability to process data while still encrypted. Some believe it's the cryptographic holy grail um and solves the challenges around how data is used uh and made confidential on the blockchain and many

other applications besides in terms of use cases. So we can look at MEV resistance DeFi, truly private dexes with encrypted order books, secure governance, secure onchain gaming with private asset attributes and confidential logic execution, enterprise compliance, so a number of regulations including GDPR and DORA uh protecting commercially sensitive data um which are legally bound these days as well. private digital identities verifying uh without revealing sensitive personal data.

So with the arrival of these solutions and tools for creating and working FHE in blockchain, uh the opportunity to realize the potential of these technologies is now here. However, achieving the full scope of what confidential blockchain has to offer is very much contingent on the ability to handle encrypted operations at scale. And it's only through the orders of magnitude speedups provided by dedicated hardware acceleration that the most transformative FHE powered use cases can drive the next wave of blockchain revolution. And that revolution starts here. Today we're introducing our light locker node. This is the world's first dedicated enterprisegrade FHE solution

for blockchain. It's FPGA based solution purpose-built to execute the core mathematics of FHE at speed. In the case of our LLN 1300 model, we're using AMD V80 FPGAAS designed for high performance and flexibility. These unlock scalable encrypted computation and will be updated regularly in line with our technology roadmap I'll tell you about in a moment. We look to support additional FHE schemes, blockchain infrastructure, additional chains, all the while increasing performance of the actual processing.

We've deployed our own FHEVM architecture from Zamma, starting with our own Ethereum test net that we've now launched uh deployed to the cloud. We put the rest of of ZAM's FHE components in the threshold key management system uh wrapping that up around a co-processor unit. And this means we can benchmark the whole operation from start to finish. So in terms of demonstration, we have a test net open to new users. We've put together some simple applications here. Um so in this case we have two people logged onto our D app on our test net.

One being Alice, one being Bob. Alice's address starting 49. Bob's address starting B 0. Both have encrypted balances. So, first we're going to decrypt those balances. Alice confirms the request. You can see that her balance is 50. We do the same for Bob. Decrypt his balance. Confirm the request. You can see Bob balances. Bob's balance is also 50. Now Bob's going to send 10 tokens to Alice. We encrypt the amount. So that's encrypted with zero knowledge proof. We enter Alice's address. Paste that here.

And we want to transfer the encrypted amount to the receiver. So we hit the transfer, confirm the transaction request. You can see that the encrypted balance has changed. So the transfer has successfully gone through. Now if we refresh Alice's D app, we can see that she now has an encrypted balance. Decrypting that shows the balance is now 60. Go back to Bob's to decrypt his balance. confirm that's 40. So we've successfully transferred uh 10 tokens from Bob to Alice under a fully encrypted scheme. So at no point has the data been exposed through that process.

So on a real system any volume of transactions um for any volume of transactions you'll have multiple transactions coming in at random times. So the key figure here is system latency for a blockchain end user. That's the time lag between clicking their mouse and a transaction completing for a blockchain infrastructure operator. It's the time for the chain to reach consensus. That's in reality made up of several different contributions. Primarily, it's a combination of the single transaction latency um and the throughput of the system.

That's how many transactions per second can be performed on a sustained basis. So user perceived or system latency is much more important than single transaction latency that's sometimes quoted. To optimize that we have to care about throughput and latency. So it's worth noting that on an engineering level single transaction latency and throughput are somewhat conflicting. This graph shows the performance of the light locker node against a roughly equivalent GPUbased system based on Nvidia H100s. This gives a fair comparison on the basis of cost and power consumption. Here we're looking at how long it takes to execute a batch of transactions on each system. So the light locker is the gray line. The GPU

system is the green. As you can see, the two systems have comparable initial latencies around 4 to 500 milliseconds. But beyond the first 20 or so transactions, you can see that the light locker node rapidly outstrips the GPUbased system in terms of actual results. And the story gets better as the throughput advantage becomes more prominent. You can see here that by the time the systems have been running for 10 seconds, our node has already processed around twice as many as the GPUbased system. So to help you understand what configuration you're going to need when you've developed a tool such as this, you can select your smart contract uh upload your own smart contract to our

code, our tool. uh set your target performance and the network architecture. We put in the smart contract, the uh function we want to process, the performance level required, the network architecture, in this case an optimistic um rollup and put this number of standby nodes in. And you'll see that the with the model of the light locker node we have here, you need two servers per co-processor uh to get performance over 200 transactions per second. So in this case, we need two of our servers.

So you can use that tool uh on our desk. Um we you'll talk about at our stand at the in the Wagme area in a moment. Um what we are doing here is we're making uh FHE practical for the blockchain ecosystem now but to power a global encrypted internet and handle immense workloads in the future uh looking at further applications such as web 3 and other uh and secure AI applications at a global scale we have to push beyond the limits of where electronics gets us the ultimate bottleneck becomes the energy and the time it takes to move electrons and that's why a core development at optalis is uh is not using electrons, we use photons. So we compute with light.

So by doing that our future optical accelerators unlock further orders of magnitude gains in speed and power efficiency. This isn't just about faster FHE. It's about building a fundamentally new class of compute infrastructure capable of powering a truly secure and private global web. So the route to mainstream blockchain adaption becomes clear begins with the need for confidentiality enabled by FHE. If we come back to our friend Jeremy Aair again at Circle who says no business in their right mind is going to build onchain unless they have totally secure confidential transactions. So if we want society at large to adapt

blockchain technology, we have to preserve the privacy of the data. It's made practical today with our gen one light locker node based on FPGAAS and we scale that into high performance uh future technology with our phetonic solutions pioneering advances in optical computing. So if you're building the future of blockchain and need practical high-speed confidentiality confidentiality then come and talk to us. You can visit us at our stand in the Wagme Boulevard. use the tool that we went through uh live chat through use cases and also catch us at the Phoenix encryption day on Thursday at the hotel grade Albon. Uh you can also scan the QR code here uh

find out more about us and light locker node and download documentation. Thank you very much. [Applause] Any questions? Any questions guys? Yeah. Uh my my question is how much is the poson photon chips are expensive? I mean in can normal users can afford it.

So price price to be determined. We these are still under development at the moment. We are not aiming them to be prohibitively expensive though and in fact part of the features of Phutonix is the ability to actually use processes that aren't necessarily expensive uh to build with. So this is going to be uh comparable in terms of GPU costs. What we're aiming for is widespread adoption of this technology. So more news to come on that. All right. Thank you. Okay, thank you.

[Applause] [Music]