Paper 2023/532

HLG: A framework for computing graphs in Residue Number System and its application in Fully Homomorphic Encryption

Shuang Wu, Huawei International (Singapore)
Chunhuan Zhao, Huawei Technologies (China)
Ye Yuan, Huawei International (Singapore)
Shuzhou Sun, Huawei Technologies (China)
Jie Li, Huawei Technologies (China)
Yamin Liu, Huawei International (Singapore)
Abstract

Implementation of Fully Homomorphic Encryption (FHE) is challenging. Especially when considering hardware acceleration, the major performance bottleneck is data transfer. Here we propose an algebraic framework called Heterogenous Lattice Graph (HLG) to build and process computing graphs in Residue Number System (RNS), which is the basis of high performance implementation of mainstream FHE algorithms. There are three main design goals for HLG framework: • Design a dedicated IR (HLG IR) for RNS system, here splitting and combination of data placeholders has practical implications in an algebraic sense. Existing IRs cannot efficiently support these operations. • Lower the technical barriers for both crypto researchers and hardware engineers by decoupling front-end cryptographic algorithms from the back-end hardware platforms. The algorithms and solutions built on HLG framework can be written once and run everywhere. Researchers and engineers don’t need to understand each other. • Try to reduce the cost of data transfer between CPU and GPU/FPGA/dedicated hardware, by providing the intermediate representation (IR) of the computing graph for hardware compute engine, which allows task scheduling without help from CPU. We have implemented CKKS algorithm based on HLG framework, together with a compute engine for multiple CPU cores. Experiment shows that we can outperform SEAL v3 Library in several use cases in multi-threading scenarios.

Note: This paper includes: 1. Introduction of functionalities and design rationale of HLG framework 2. User manual on usage and interfaces for HLG framework and HLG-CKKS library.

Metadata
Available format(s)
PDF
Category
Implementation
Publication info
Preprint.
Keywords
Fully Homomorphic EncryptionResidue Number SystemComputing GraphHeterogenous ComputingIntermediate Representation
Contact author(s)
wu shuang @ huawei com
zhaochunhuan @ huawei com
yuanye44 @ huawei com
sunshuzhou @ huawei com
lijie303 @ huawei com
liuyamin3 @ huawei com
History
2023-04-13: approved
2023-04-12: received
See all versions
Short URL
https://ia.cr/2023/532
License
Creative Commons Attribution
CC BY

BibTeX

@misc{cryptoeprint:2023/532,
      author = {Shuang Wu and Chunhuan Zhao and Ye Yuan and Shuzhou Sun and Jie Li and Yamin Liu},
      title = {{HLG}: A framework for computing graphs in Residue Number System and its application in Fully Homomorphic Encryption},
      howpublished = {Cryptology {ePrint} Archive, Paper 2023/532},
      year = {2023},
      url = {https://eprint.iacr.org/2023/532}
}
Note: In order to protect the privacy of readers, eprint.iacr.org does not use cookies or embedded third party content.