PQ Transaction Signatures #003
Transcript
- Renaud-ZKNOX
gm, gm
- Fab
Bonjour
- matteo vicari
Gm
- Renaud-ZKNOX
Whose notetaker is it ? Could you share the note after the meeting ?
- Fireflies.ai Notetaker Andriia
Andriian invited Fireflies.ai here to record & take notes. View Security & Privacy info: https://fireflies.ai/policy Type: '/ff leave' - Remove Fireflies View Realtime notes here: https://app.fireflies.ai/live/01KJWFJGA8M4F7QC5JHE6XRR6M?ref=live_chat
- Gottfried Herold
quick question: Do what extend do these precompiles utilize hardwiring the prime you work with?
- Stefa
The way I always envisaged a development like a NumPy precompile would be to support a large set of vectorised — or, more broadly, tensorised — operations. It should be a universal extension to make SIMD programming feasible on the EVM, and this would feed into a lot of new potential AI-related use cases for Ethereum. The bigger vision is so much more interesting.
- Danno Ferrin - Tectonic.xyz
How does the precompile plan on addressing the various bit widths of canonical public key and signaure format (3-bit, 4-bit, 10-bit, 14-bit,20-bit) and the huffman-style encoding of falcon signatures?
- Ak
ML/finance primitives would benefit a lot from numpy precompile
- Stefa
Replying to "ML/finance primitive..." From a general-purpose one, I fully agree. From a specific one for selected vectorial operations, not so much I believe.
- Stefa
Reacted to "ML/finance primitive..." with 👍
- Simon ZKNOX
We have been helped by Yoon working on the NTT EIP :-)
- Danno Ferrin - Tectonic.xyz
After we finish this discussion I’d like to go down that tangent of not carrying the publick key in the transaction. That has some intersting potential but we would lose some key properties of current transacitons
- Danno Ferrin - Tectonic.xyz
Why ML-DSA65 and not 44?
- Antonio Sanso
Reacted to "We have been helped ..." with 👍
- Renaud-ZKNOX
if you have a precise algorithm/use case i am interested into benchmarking it
- Renaud-ZKNOX
we plan to bench poseidon/ntt/falcon and maybe large ecc (secp384r1)
- Antonio Sanso
Reacted to "we plan to bench pos..." with 🔥
- Renaud-ZKNOX
but please give concrete example, eager to add it to our use cases
- Danno Ferrin - Tectonic.xyz
Replying to "we plan to bench pos..." ECC has an expiration date, I hope that is just for reference and not inclusion
- Renaud-ZKNOX
you can use create2 on your AA account
- Renaud-ZKNOX
Répondre à "Why ML-DSA65 and n..." because it seems that web2 is adopting it rather than 44
- Gottfried Herold
Wouldn't those "users" be wallet providers?
- Danno Ferrin - Tectonic.xyz
I’ve got a hard stop in 30 seconds. I’d love to stay longer
- Gottfried Herold
Those should be somewhat tech-savvy, no?
- Antonio Sanso
Reacted to "I’ve got a hard stop..." with 👍
- Renaud-ZKNOX
Users will use audited library.
- Stefano Gogioso
Reacted to "Users will use audit..." with 👍
- Renaud-ZKNOX
So this is not letting average joe vibecoding their wallet. Some natural candidates would rise.
- Renaud-ZKNOX
see u all
Call summary
Targets
- •Next call (2 weeks): Account Abstraction focus with Frame Transaction team participation - 01:06:00
Decisions
- •ZKnox team prioritizing NVMPI precompile development over NTT-specific implementation - 00:44:36
- •ML-DSA-65 selected as reference over ML-DSA-44 based on Web2 adoption trends - 00:46:27
Highlights
- Ntt Precompile Proposal:
- ·EIP-7885 NTT precompile: 4 opcodes for polynomial multiplication, supports lattice schemes and Stark verification - 00:11:48
- ·Falcon verification benchmarked at 400K gas with NTT precompile (down from 1.5M pure Solidity) - 00:18:30
- ·Shake precompile combo could achieve ~2K gas for Falcon; calldata cost remains dominant - 00:19:15
- ·NTT precompile hardwires 5-6 standard primes; generic primes require expensive lookup table transmission - 00:24:18
- Implementation Challenges:
- ·14-bit Dilithium/Falcon encoding requires expensive bit manipulation in EVM (~150K gas decompression cost) - 00:31:41
- ·Under 200K gas computational floor with NTT+Shake; monolithic NIST precompile significantly cheaper - 00:39:00
- ·Data carriage cost dominates: 500-2000 bytes for PQ signatures (5-10x current ECDSA) - 00:57:10
- Account Abstraction Synergy:
- ·Native AA with Create2 enables deterministic address derivation from public key hash - 00:53:38
- ·AA allows on-chain public key storage, enabling key rotation and reducing calldata transmission - 00:55:35
- ·Frame transactions provide signature modularity but need separate composable signature mechanisms - 01:01:48
- Numpy Precompile Alternative:
- ·NVMPI precompile proposed: vectorized modular arithmetic, single opcode with operation parameter - 00:20:10
- ·NVMPI more generic (benefits ML/finance/Poseidon) but higher gas cost than NTT - 00:22:02
- ·NVMPI uses Barrett multiplication with generic moduli up to 2^32; constants computed on-the-fly - 00:46:27
Action Items
- •Renaud/Simon (ZKnox team) - Benchmark NVMPI precompile with Falcon/Poseidon/large ECC and provide gas cost breakdown - 00:48:00
- •Community (post examples to team) - Collect concrete ML/finance vectorized operation use cases for NVMPI scope definition - 00:50:15
- •Antonio (next call organizer) - Coordinate with Frame Transaction/AA teams on signature separation and composability requirements - 01:02:22
Key decisions
ZKnox team prioritizing NVMPI precompile development over NTT-specific implementation
Prioritization of NVMPI precompile developmentML-DSA-65 selected as reference over ML-DSA-44 based on Web2 adoption trends
Selection of ML-DSA-65 as reference
AI Disclaimer: Some content or metadata on EIPsInsight may be AI-inferred or automatically compiled. If you find any discrepancy, please contact us at dev@avarch.org.