Thoughts on ZKVM and ZK Verification

Author: Miles Source: X, @Miles082510

Yesterday, I discussed the issue of ZK acceleration with a domestic ZK chip manufacturer. At present, the L2 based on ZK only needs a few hundred GPU machines to maintain daily business, and the future development of Asic will be even more exciting. So, where is the current acceleration demand? Which parts are worth accelerating? Is the bottleneck of the ZK track in proof generation or verification? What impact will these issues have on the ecosystem?

Based on this tweet, I would like to share some of my own understanding, welcome corrections and discussions!

Past ZK technology

Early ZK solutions were mainly based on circuit development. Developers must build circuits using a specific language, which is complex and expensive. @StarkWareLtd is leading the way in this space, building circuits in Cairo with STARKs, leveraging the recursive proof advantages of STARKs. However, this approach is limited in that it can only aggregate STARK proofs from a single proof system or machine.

Current ZK Technology

Now, with the advent of the general-purpose ZKVM, programmers can develop applications by simply writing Rust code without having to learn a new circuit language. Representatives in this area include @RiscZero, @SuccinctLabs' SP1, @NexusLabsHQ, and @lita_xyz and jolt from @a16zcrypto.

While there are many ZKVM projects, I believe that all future proof aggregation systems will rely on RISC-V ZKVMs or Rust-based ZKVMs. By using RISC-V proofs, we can easily merge different proof systems and simplify the complex verification process.

Problems and Challenges

The cost of verifying proofs on Ethereum is very high and does not support large volumes. For example, a 1GB proof is not practical to verify directly on Ethereum, although it can be compressed through recursive proofs, but this is also expensive and time-consuming. If you run a ZK rollup, the cost may be as high as millions of dollars per year.

In addition, although @alignedlayer can verify 2500 hash values ​​per second, Ethereum cannot handle such a large volume.

Solution

There are two ways to achieve fast and low-cost verification on Ethereum:

• Proof Aggregation: A way to improve verification efficiency.

• ZK Verification Layer: Implementing a low-cost, scalable ZK verification layer on @eigenlayer is what @alignedlayer is currently doing.

Discussion and Views

Regarding the bottleneck of the ZK track, @Ozhar (from @zkSync) believes that only 5% of the cost is related to verification, and 95% is related to the hardware cost of generating proofs. @heslinkim (from @gevulot_network) holds a different view, believing that the real bottleneck lies in proof generation and hardware.

In fact, the field of ZK hardware and proof generation is also developing rapidly, and projects such as @cysic_xyz, @Ingo_zk and @lagrangedev are actively promoting related technologies.