search

scroll-tech / zkevm-circuits

MIT License
919 stars 387 forks source link

Short-Mid term features #1097

Open lispc opened 9 months ago

lispc commented 9 months ago

Describe the feature you would like

  1. (P0) upgrade halo2-libs and remove mix usage of challenge and halo2-libs "static" circuits. (Agg circuit / sig circuit / ecc circuit needs to be updated a lot. For agg circuit, there will be no "rlc keccak" module, we will use halo2-libs "static" keccak. For ecc/sig circuit, we will change rlc_input, rlc_ouput, .. lookup to many rows of MemoryOp input/output byte/word lookup)
  2. (P0) chunking or continuation. (Circuit should never fail for any type any size tx). This can be completed either by splitting circuits vertically(then we can increase max k to 22 even 23, then split block into txs) or horizontally(chunking of pse, FSM of zksync-era).
  3. (P0) use blob to reduce l2 tx gas fee
  4. (P0) cancun new opcodes
  5. (P1) Reduce usage of challenge: remove word_rlc and keccak_rlc.
  6. (P1) access of l1 block header
  7. (P1) Refactor sub circuit: we may need to add standalone "Arith" sub circuit and "Memory limb <=> bytes" sub circuit. Arith circuit is also useful for support "heavy computing l3s" like mimc / mixer with a lot of poseidon or mimc hash, a lot of mulmod addmod ops. Arith circuit can also be used to contrain limbs <=> bytes.
  8. (P2) changes needed for "smaller field". We should not assume word being hi-lo, instead, use more general type abstraction.
  9. (P2) exploration of spliting sub circuits. So if some circuits are not used, (eg sha256 / ecc), they will not incur proving overhead.
  10. (P2) more efficient micro-op vm. and refactor precompiles

Additional context

No response

0xShankar commented 2 weeks ago

"I've been following the updates on zkEVM circuits closely, and I’m impressed by the progress on enhancing circuit performance, especially with improvements like the RecursionCircuit and BatchCircuit. One area I believe could further streamline development is focusing on modularizing common components across circuits. This would simplify maintenance and make it easier to introduce optimizations in a single place, benefiting the broader zkEVM ecosystem.

Also, introducing automated testing tools specifically for the prover-side circuits could reduce errors during integration and ensure consistency. I’d love to contribute by helping develop a testing framework to track performance across multiple zkRollups, as well as look into improving gadget modularity.

What’s the current focus for the next quarter, and how can I best assist in accelerating the zkEVM circuit optimizations?"