Technical articles, research findings, and insights from our team.
Achieving performance parity with Plonky3 on Packed Poseidon2, implementing JAX-to-ZKX code generation pipeline, expanding ZKX opcodes for comprehensive prover support, and developing Poseidon2 implementation in pure JAX.
Implementing Poseidon2 in MLIR with performance improvements over Plonky3, and porting WHIR-p3 to Python using JAX for ZK-specific runtime compilation.
Demonstrating how compiler-driven loop fusion achieves 1.5-2× performance improvements over manual GPU programming frameworks like ICICLE, by minimizing memory transfers and kernel launches.
RabbitSNARK achieves up to 2× faster performance than RapidSNARK through compiler-driven optimizations, making it the fastest Groth16 CPU prover with consistent cross-platform performance.