Paper 2024/973

ICICLE v2: Polynomial API for Coding ZK Provers to Run on Specialized Hardware

Karthik Inbasekar, Ingonyama
Yuval Shekel, Ingonyama
Michael Asa, Ingonyama
Abstract

Polynomials play a central role in cryptography. In the context of Zero Knowledge Proofs (ZKPs), protocols can be exclusively expressed using polynomials, making them a powerful abstraction tool, as demonstrated in most ZK research papers. Our first contribution is a high-level framework that enables practitioners to implement ZKPs in a more natural way, based solely on polynomial primitives. ZK provers are considered computationally intensive algorithms with a high degree of parallelization. These algorithms benefit significantly from hardware acceleration, and deployed ZK systems typically include specialized hardware to optimize the performance of the prover code. Our second contribution is leveraging our polynomial API to abstract away low-level hardware primitives and automate their memory management. This device-agnostic design allows ZK engineers to prototype and build solutions while taking advantage of the performance gains offered by specialized hardware, such as GPUs and FPGAs, without needing to understand the hardware implementation details. Finally, our polynomial API is integrated into version 2 of the ICICLE library and is running in production. This paper also serves as a comprehensive documentation for the ICICLE v2 polynomial API.

Metadata
Available format(s)
PDF
Category
Applications
Publication info
Preprint.
Keywords
ZKPHardware AccelerationPolynomial API
Contact author(s)
karthik @ ingonyama com
yuval shekel @ ingonyama com
miki asa @ ingonyama com
History
2024-06-17: approved
2024-06-16: received
See all versions
Short URL
https://ia.cr/2024/973
License
Creative Commons Attribution-NonCommercial
CC BY-NC

BibTeX

@misc{cryptoeprint:2024/973,
      author = {Karthik Inbasekar and Yuval Shekel and Michael Asa},
      title = {{ICICLE} v2: Polynomial {API} for Coding {ZK} Provers to Run on Specialized Hardware},
      howpublished = {Cryptology {ePrint} Archive, Paper 2024/973},
      year = {2024},
      url = {https://eprint.iacr.org/2024/973}
}
Note: In order to protect the privacy of readers, eprint.iacr.org does not use cookies or embedded third party content.