pipette
ENEnglish

ZKPU: Zero-Knowledge Processing Unit for Heterogeneous Workloads

Jonathan Ku, Changchun Zhou, Jiawen Wu, Sangouk Jeon, Cong Guo, Yuzhe Fu, Bowen Duan, Dongsuk Jeon, JV Rajendran, Yiran Chen, Hai Li

PreprintDice ser un gran avanceUso en el mundo real

En palabras de los autores

Zero-knowledge proofs (ZKPs) have become a fundamental building block for privacy-preserving computation, blockchain systems, and verifiable machine learning. Modern ZKP protocols span two distinct families: curve-based protocols, which rely on large finite-field arithmetic, and hash-based protocols, which operate over native 64-bit fields. Existing hardware accelerators are specialized for one protocol family because they are built on operand-scanning Montgomery multiplication, resulting in fixed-width datapaths, poor utilization across heterogeneous operand widths, and limited scalability. This paper presents ZKPU, a Zero-Knowledge Processing Unit for heterogeneous ZKP workloads. ZKPU introduces a new compute paradigm by co-designing finely integrated product scanning (FIPS) Montgomery multiplication with a reconfigurable Twin Systolic Fabric (TSF), enabling modular multiplication at arbitrary widths in 64-bit increments. Combined with a fine-grained Deque buffer, on-chip twiddle-factor generator, and memory-efficient MSM preprocessing strategy, ZKPU provides a unified architecture for both curve-based and hash-based protocols. Evaluation using Groth16 and Plonky2 shows that ZKPU achieves up to speedup for NTT, for MSM, and approximately end-to-end application speedup over state-of-the-art accelerators, while being the first architecture to efficiently support both protocol families on a single reconfigurable hardware platform.

Resultado principalEl resumen no menciona limitaciones.

Apareció: miércoles, 21 de octubre. arXiv. Preprint, todavía sin revisión por pares.