Now, execution layer changes. I've already talked about account abstraction, multidimensional gas, BALs, and ZK-EVMs. I've also talked here about a short-term EVM upgrade that I think will be super-valuable: a vectorized math precompile (basically, do 32-bit or potentially 64-bit operations on lists of numbers at the same time; in principle this could accelerate many hashes, STARK validation, FHE, lattice-based quantum-resistane signatures, and more by 8-64x); think "the GPU for the EVM". https://firefly.social/post/x/2027405623189803453 Today I'll focus on two big things: state tree changes, and VM changes. State tree changes are in this roadmap. VM changes (ie. EVM -> RISC-V or something better) are longer-term and are still more non-consensus, but I have high conviction that it will become "the obvious thing to do" once state tree changes and the long-term state roadmap (see https://ethresear.ch/t/hyper-scaling-state-by-creating-new-forms-of-state/24052 ) are finished, so I'll make my case for it here. What these two have in common is: * They are the big bottlenecks that we have to address if we want efficient proving (tree + VM are like >80%) * They're basically mandatory for various client-side proving use cases * They are "deep" changes that many shrink away from, thinking that it is more "pragmatic" to be incrementalist I'll make the case for both. # Binary trees The state tree change (worked on by @gballet and many others) is https://eips.ethereum.org/EIPS/eip-7864, switching from the current hexary keccak MPT to a binary tree based on a more efficient hash function. This has the following benefits: * 4x shorter Merkle branches (because binary is 32*log(n) and hexary is 512*log(n)/4), which makes client-side branch verification more viable. This makes Helios, PIR and more 4x cheaper by data bandwidth * Proving efficiency. 3-4x comes from shorter Merkle branches. On top of that, the hash function change: either blake3 [perhaps 3x vs keccak] or a Poseidon variant [100x, but more security work to be done] * Client-side proving: if you want ZK applications that compose with the ethereum state, instead of making their own tree like today, then the ethereum state tree needs to be prover-friendly. * Cheaper access for adjacent slots: the binary tree design groups together storage slots into "pages" (eg. 64-256 slots, so 2-8 kB). This allows storage to get the same efficiency benefits as code in terms of loading and editing lots of it at a time, both in raw execution and in the prover. The block header and the first ~1-4 kB of code and storage live in the same page. Many dapps today already load a lot of data from the first few storage slots, so this could save them >10k gas per tx * Reduced variance in access depth (loads from big contracts vs small contracts) * Binary trees are simpler * Opportunity to add any metadata bits we end up needing for state expiry Zooming out a bit, binary trees are an "omnibus" that allows us to take all of our learnings from the past ten years about what makes a good state tree, and actually apply them. # VM changes See also: https://ethereum-magicians.org/t/long-term-l1-execution-layer-proposal-replace-the-evm-with-risc-v/23617 One reason why the protocol gets uglier over time with more special cases is that people have a certain latent fear of "using the EVM". If a wallet feature, privacy protocol, or whatever else can be done without introducing this "big scary EVM thing", there's a noticeable sigh of relief. To me, this is very sad. Ethereum's whole point is its generality, and if the EVM is not good enough to actually meet the needs of that generality, then we should tackle the problem head-on, and make a better VM. This means: * More efficient than EVM in raw execution, to the point where most precompiles become unnecessary * More prover-efficient than EVM (today, provers are written in RISC-V, hence my proposal to just make the new VM be RISC-V) * Client-side-prover friendly. You should be able to, client-side, make ZK-proofs about eg. what happens if your account gets called with a certain piece of data * Maximum simplicity. A RISC-V interpreter is only a couple hundred lines of code, it's what a blockchain VM "should feel like" This is still more speculative and non-consensus. Ethereum would certainly be *fine* if all we do is EVM + GPU. But a better VM can make Ethereum beautiful and great. A possible deployment roadmap is: 1. NewVM (eg. RISC-V) only for precompiles: 80% of today's precompiles, plus many new ones, become blobs of NewVM code 2. Users get the ability to deploy NewVM contracts 3. EVM is retired and turns into a smart contract written in NewVM EVM users experience full backwards compatibility except gas cost changes (which will be overshadowed by the next few years of scaling work). And we get a much more prover-efficient, simpler and cleaner protocol.

YouTube video by Cliff Cash Comedy

YouTube video by Krushed Opiates Music