Skip to content
EthCC - Cannes, France

Jordi Baylina (Polygon Cofounder)_ZisK: proving in real time

Jordi Baylina

0:000:00

Key takeaways

  • Spin-off for Neutrality: Jordi Baylina’s team separated Zisk from Polygon to avoid single-customer dependence and let any chain, including Polygon competitors, adopt the prover stack.
  • Low-Latency First: Zisk is designed around real-time proving; with 64 GPUs it already generates a full Ethereum block proof in under 12 seconds, a ~7× improvement over the 90 seconds required on four GPUs.
  • Full-Stack zkProver: Beyond the zkVM, the stack includes the PIL2 language, ProofMan for proof orchestration, and a recursive “recursor” layer, enabling end-to-end customization without external IP.
  • 128-Bit Security Mandate: The project refuses to drop below 128-bit security, arguing that cryptographic assurances are not a tunable cost lever.
  • RISC-V 64-Bit Advantage: A 64-bit RISC-V ISA lets developers prove native Go or C programs more efficiently and roughly doubles memory-transfer throughput versus 32-bit designs.
  • Open-Source Ethos: All code remains permissively licensed so no chain must pay royalties to publish blocks, preserving Ethereum’s open infrastructure model.

On Strategic Rationale and Governance

Jordi Baylina, Co-founder of Polygon, explained that Polygon now concentrates on an aggregation layer that can source proof technology from many vendors. Spinning Zisk into an independent, community-run startup eliminates perceived vendor lock-in and broadens the addressable market. The move also positions Zisk to commercialize its prover across rival ecosystems without conflict.

On Technical Architecture

Zisk decomposes computation into specialized state machines that communicate through a shared “bus.” Each machine assumes certain operations—such as Keccak hashing, 32- or 64-bit multiplication, or unaligned memory reads—and another machine proves them, with all assertions reconciling to zero on the bus. Proofs are then recursively aggregated in a tree, ensuring that every sub-circuit shares identical public inputs. The 64-bit RISC-V core supports alignment-aware memory and parallel chunking, while PIL2 provides declarative constraint definitions familiar to developers coming from the earlier Hermez zkEVM work.

On Performance Benchmarks

In live demos at ETHCC, the team proved a block in Ethereum’s costliest five-percentile using four different GPU cluster sizes. Latency dropped from 90 seconds on 4 GPUs to 11.8 seconds on 64 GPUs, with further gains expected from load-balancing the witness computation and tuning GPU kernels. The aggregation layer is already negligible in runtime, indicating that remaining bottlenecks are local to the main prover and amenable to parallel scaling.

On Market Applications

Low-latency proofs unlock synchronous cross-chain messaging, zk-based light clients for mobile, and eventual “zk-consensus” at Layer 1 where each block ships with a native proof. For Layer 2 rollups, sub-12-second proofs align with Ethereum’s slot time and could let rollups finalize in a single slot, closing the UX gap with monolithic chains.

Key actionable insights

  • Track Zisk’s roadmap for sub-12 second proofs on commodity 32-GPU clusters; if achieved, real-time zk-based consensus at Layer 1 becomes credible.
  • Monitor adoption of 64-bit RISC-V zkVMs; they may become the default for proving general-purpose code across chains.
  • Watch for ecosystem pressure toward mandatory 128-bit security as projects using lower parameters may face reputational or regulatory risk.