Skip to content

Roadmap

Solo build, NLUpdated 2026-04-28

Dates are rough, I might be wrong about all of them.

PhaseDoneHeadline
1Apr 21First on-silicon BN254 Groth16 verify, 988 ms baseline on Cortex-M33
2Apr 23UMAAL inline asm + SRAM placement, 988 → 641 ms at the time (later 551 ms, see row 5.6)
2.5Apr 22BLS12-381 Groth16, same 128 KB SRAM tier, 7.7 s on M33
3Apr 23First STARK verify on $5 silicon, Goldilocks Fibonacci AIR, 74.7 ms at 95-bit
3.3Apr 24-25BabyBear × Quartic cross-ISA experiment, collapsed RV32/M33 gap from 1.51× to 1.04×
4Apr 26First on-device STARK prover, feasibility at ~32-bit security
5Apr 26STARK prover at production security, 148 ms BabyBear+Quartic, 95-bit conjectured
5.5Apr 28Independent Poseidon2-BabyBear audit, 5 findings, 40 tests green
5.6Apr 28Audit-driven self-review uncovered untracked drift; rebench landed BN254 verify at 988 → 551 ms (44% drop), +9 pp vs the previously claimed 35 %
PhaseTargetWhat it is
6May-Jun 2026PQ-Semaphore on RP2350. Replace BN254 Groth16 in the Semaphore path with STARK + audited Poseidon2-BabyBear Merkle membership circuit. Anonymous identity, post-quantum, on-device.
PhaseTargetWhat it is
7Jun-Jul 2026Recursive Groth16 wrapper. Off-chain wrap the on-device STARK in a BN254 Groth16 proof so it can be verified on Ethereum at ~250k gas instead of ~5M. Restores Ethereum-native compatibility for the PQ identity stack.
8Q3-Q4 2026Hardware productization. Custom enclosure, USB-C/BLE, manufacturing partner, FCC/CE certification. Move from “Pico 2 W on a desk” to “shippable device with a price tag”.
9Q4 2026 / Q1 2027Dilithium attestation track. Separate non-anonymous flow for device attestation use cases (firmware signing, secure boot). NIST-blessed PQ signature primitive on the same hardware.
  • Generic SNARK platform which is what every other ZK project ships. zkmcu stays focused on edge devices and identity. Plenty of people doing the platform thing better resourced than me.
  • L1 chain. Not building one. Not interested. Too crowded, too capital-intensive, too long timeline.
  • Token launch before product. Tokens before there’s a working economic loop is how you end up in a courtroom.
  • General-purpose Rust ZK library. Arkworks and Plonky3 cover that surface, I’m not duplicating.
  • Plonky3 API churn. They’re pre-1.0 and refactor frequently. Each refactor potentially breaks our diff tests. Mitigation: vendor the version we audit against, upgrade on a deliberate cadence.
  • Hardware supply chain (RP2350 specifically). Pico 2 W availability has been stable, but if Raspberry Pi changes pricing or pulls the chip, the whole project pivots to alternative MCUs. Mitigation: keep the codebase MCU-agnostic where possible.
  • Recursive proof aggregation centralization. Phase 7 wrapper needs to run somewhere. If the only good wrapper service is run by one company, that’s a failure mode. Mitigation: make the wrapper code public and easy to self-host.

Open issues / PRs at github.com/Niek-Kamer/zkmcu. Grant programs, investors, design partners, all on the contact page.