Forget Private AMMs: Scaling DeFi Back Onchain with Parallel AMM

Transcript

Okay. Uh good morning everyone. I'm Dido from the Chris Aldridge and Predator Dow team. I'm super excited to be here. I got a limited time, so I'm going to like dive straight in. Um so today I'm going to share the recent uh progress that our team has made on scaling sequential AMM via parallel architecture. So our target is to scale DeFi back on chain. Let's dive in. So firstly, when we talk about constant product AMM, which is Uniswap or the other AMMs we usually heard, uh that is indeed um we we love it and we think it is the great greatest innovation probably

happening in the DeFi space for the past 5 to 6 years. And uh for the first time, traders get to trade without any intermediaries. And uh indeed, instead they interact with a smart contract via the bonding curve. So every swap moves the reserves along the bonding curve. So the AMM is actually constantly searching for the newer price. And so therefore I would invite you to consider a constant product AMM in this specific lens that it is indeed an on-chain continuous price discovery engine.

And uh we know for most of the short-tail assets, well like uh Bitcoin, Ethereum, stable coins, etc., and actually the most majority of their liquidity and trading volume happening firstly on the centralized exchange. So what we have here is indeed a reactive feedback loop. So we have the in external which at the market price happening on the centralized first reach a new fair price and then there will be the arbitrators who are the informed traders wanting to like capture this price difference for profits. And they want to step in and there will be the competitive arbitrage trades happen. And then

either is arbitrage or is normal trades, it's just trades. So for an AMM, the reserve of updates the price changes and the AMM price actually converges towards the market fair price. Therefore, we want to come down to this conclusion that for mainstream assets, a constant product AMM largely function as a passive price discovery mechanism. And uh the either is MEV or arbitrage trading activities, we we think it's one one of the very important process of correcting the price. We can't eliminate them and we shouldn't.

And there's a very important question we ask here is how fast does the AMM price converge to the fair price? And when we talk about fast, there's a dimension of time. So let's look back at sequential AMM. Trust me, we love Uniswap, but Uniswap does come with a bandwidth constraint by design. Say during the high volatility market times, there will be huge amounts of transaction actually trying to access the same liquidity pool. To give you a concept about how this concentration looks like, I got a number here. So indeed, the leading 0.17% of Uniswap V2 pools perce- possess almost 40% of the total

swaps. That's huge and it's quite concentrated. So there are the hot spots on the Uniswap. It has always been and it will be in the future. So in this case, um there will be a single shared asset reserve state accessed at the same time by huge amounts of transactions. There's one global right lock over there and Uniswap just designed in this way. All swaps must serialize everything like transactions are executed one by one. And then, what does it cost? We got this huge volume transactions that intensified the state contention actually. And then the gas fee spikes, transactions queue up, and there could be transactions spilled across multiple

blocks. And there will be Therefore, there will be an extended stale price time window remaining open for the arbitrators. And for the LPs, that's terrible news. Cuz we have this concept of loss versus rebalancing as known as LVR. Actually, it is one form of MEV that exploit more value than all the other MEV forms combined. So it is a serious condition and the extended time window just make it worse. And this low rich this low risk arbitrage window will be exploited by those arbitrators. And they're actually very well positioned to exploit this slow updates. And that indeed

caused structural cost to the LPs. LP paid the bills. Uh LP paid the bills and arbitrators make the money. So price discovery becomes a very competitive killing in this case on Uniswap. So seeing this bottleneck, the market have their own alternative approach as private AMM. So what does private AMM do? It it stop waiting for price discovery on chain. They moves it off chain. And actually they have their own what they call pricing prediction models collecting all the information like the gathering like order flow signals and market data aggregation and that just push this

uh off-chain information on chain. And then at the end, uh smart contract and blockchain only reduced as the settlement layer. So you can barely call it AMM cuz they don't have the the the constant product anymore. And then, you know, there are plenty of benefits we're going to talk about. It's It's really extremely good using experience for the traders because if there's no near zero slippage and it's extremely fast. Uh and for the market makers, they love it because it it is the it provides extreme capital efficiency for them. But now you know I'm going to talk more about the tradeoffs. The tradeoff here is much bigger than that than that box.

So for the traders for the tradeoffs one, we have the the private AMM provides an opaque execution. They're doing everything off chain and it's almost like a black box with closed source models that it is impos- impossible to audit it. And also there are the hidden order flows probably sent privately to the market makers. And there will be censorship mechanisms surfacing as operators can actually filter or reject some of the what they call toxic order flow. So that creating a selective access, it may only whitelist some of the orders and to for execution.

And then the most important thing the the the worst thing, let's say, is the frag- fragmented and isolated liquidity because liquidity is completely completely off chain. So that liquid This kind of liquidity is completely isolated and not composable anymore with any other DeFi building blocks we have in the space that actually make DeFi powerful. So we're asking if private AMM is the industry response, then the DEX aggregators like say Jupiter on Solana actually are becoming the new centralized exchange, probably the new Binance. And what's the point of decentralization of blockchain? So the fundamental ask a question we're asking

here is can we scale price discovery without leaving the chain? We keep everything on chain. Is it possible to do that? We come up with the solution what we call and parallel AMMs. It's actually based on a parallel programming model that we have been working on for a few years. And now we implemented it in the parallel AMMs and to to to have this faster feedback loop on chain. So our ideal case, the target would be starting from the same ideal price, sale price, parallel AMMs would eat much shorter time than sequential AMM to converge to the fair price.

And then uh we we got this reality that our execution speed is two to six times faster than sequential AMM, which translates to 75% down in time in mis- pricing half-life. And before diving into details, I would like to share an inspiration of this parallel AMM design. It's uh our based on our observation of Uniswap V2. The numbers are here. Actually, the majority of transactions make negligible price impact. They don't have price impacts. To give you a number, over 85% of transactions would make a price delta less than like 1,000th. So our approach is that we parallelize the transactions of them. We route We would route the

large transactions to a global pod for the large transactions for the transactions that actually have a price impact. And this one maybe hold probably 80% of total liquidity. And all the small transactions would be routed to the smaller pods for for parallel transaction that and that each of them may hold only 5% of the total liquidity. And we also have that a periodical state synchronization and automatic rebalancing to synchronize the liquid assets the reserves of them across all the pods.

You may ask a question, you got the you got your liquidity splitted. Does that intro- like introduce worse slippage for the traders? Challenge accepted. We have to redesign the whole algorithm. So we have this algorithm with that if a trans- small transaction is sent to a pod, they they we don't quote local reserve. The price is calculated based on the approximate approximate total across all the pods. So are the large transactions. So we have we kind of we have to keep this pricing with a acceptable range of accuracy in our design. I won't get get into the details of this like equation, but you can ask me later.

Let's Let's Let's chat about it. And we have this performance validation here. We got Firstly, let's look at throughput. The execution speed is indeed 1.8 to 5.8 6 x TPS versus sequential AMM. And that translates to 70% reduction in convergence time to the fair price. And for the performance of rye, the price deviation will remain 1.5%. At 90 percentage when we have four sub paths. And that is performance. Just to to prove that we have built up everything and we have stress tested it.

And we were here last year at ECC introducing the underlying programming model, which is Crystallinity. And now we are here. Crystallinity is fully composable with Solidity written in Ethereum client. And we're currently running a validator implementing Crystallinity on testnet. And that enables like parallel execution at the execution layer of Ethereum. So that's what we do on a on a higher level. So what's next? We're going to Yeah. Okay. I'm I'm going to stop here. I'm happy to take what whatever questions you may have.

Okay. Just feel free. Got a question? Sure. Uh by sequential AMM, I assume you mean Uniswap. Have you compared any of these metrics to like a Hyperliquid? Uh not really. We We We We look at more on like Uniswap and constant product AMM only. Yeah. We don't look at it order book thing cuz it's not just comparable to like decentralized order book or whatever order book you're operating. So So that's why I mentioned constant product AMM like repeatedly in these slides. Yeah.

And still Yeah. Cuz that's why Hyperliquid is so competitive. Yeah. I I I totally agree with you. And I think uh the slippage we have right now is is the number is quite close to sequential AMM. And that's I think the best we can achieve. But this is the first few steps. We We We We are just We just built the central building blocks of this parallel AMM the architecture. So we are adding more in mechanism and probably more um I don't know the extensions to it. So like Uniswap the hooks that they have in Uniswap V2 and making everything better make the numbers better, make slippage even much thinner than what Uniswap has.

And that's our goal. Yeah. Thanks for your question. Anyone else? Feel free to chat with me. Thank you. >> [applause]