High Throughput Arithmetic Computing Unit for BFV Homomorphic Encryption

초록

Homomorphic Encryption (HE) enables secure computations on encrypted data, which is<br /> crucial for cloud and edge computing. The BFV scheme, widely used for integer arithmetic, faces performance bottlenecks in polynomial multiplication, especially in tensor operations. The Residue Number System (RNS) helps address this, leading to the BEHZ and HPS BFV variants. The Halevi-Polyakov-Shoup (HPS) variant simplifies implementation but still struggles with the overhead of modular arithmetic. We propose an Arithmetic Computing Unit (ACU) optimized for key BFV operations, including modular addition, multiplication, NTT/INTT, and MAC to improve efficiency. Implemented on a Xilinx Alveo U250 FPGA, our design achieves up to 2.3x higher throughput, 2.2x less latency, and 9.4x better BRAM efficiency than existing solutions, demonstrating FPGA acceleration's for homomorphic encryption.

키워드