Reproduce “Trinity: A General Purpose FHE Accelerator” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xianglong Deng, Shengyu Fan, Zhicheng Hu, Zhuoyu Tian, Zihao Yang, Jiangrui Yu, Dingyuan Cao, Dan Meng and 4 more
Trinity is introduced as the first FHE accelerator with a unified hardware architecture, designed to efficiently support both arithmetic (CKKS) and logic (TFHE) FHE schemes, along with their conversion algorithms. The system achieves an average speedup of 1.49x over SHARP for CKKS, 4.23x over Morphling for TFHE, and 13.42x for hybrid FHE applications compared to a combined SHARP+Morphling system, while reducing total circuit area by 15%.
Key insights
- Replacing the Fast Fourier Transform (FFT) with NTT in TFHE allows for significant hardware reuse and unification of core computational primitives across different FHE schemes.
- Configurable Units (CUs) capable of dynamically reconfiguring to perform either NTT or MAC operations are essential for efficiently handling varying polynomial lengths and imbalanced FHE workloads.
- A unified, multi-modal FHE accelerator can achieve superior performance and reduced hardware overhead compared to integrating multiple specialized accelerators.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.