Energy-Efficient Flexible RNS-CKKS Processor for FHE-Based Privacy-Preserving Computing

This article proposes a total hardware solution for the RNS-CKKS algorithm supporting flexible parameter sets with 2.7-to-13.3-<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mu $ </tex-math></inline-formula>J/boot/slot energy consumption. Targeting fully homomorphic encryption (FHE), we focus on the key-switch operation, which is the most essential, complicated operation for the bootstrapping process to alleviate accumulated error and enable unlimited FHE computations. Introducing novel two-level scheduling schemes for low-latency key-switch operation, the proposed hardware reduces key-switch latency by 47% and increases resource utilization by <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$1.9\times $ </tex-math></inline-formula> compared with a straightforward process. Moreover, with optimized hardware engines dynamically supporting various ciphertext parameter sets, the proposed hardware fabricated in 28-nm CMOS technology enables the most efficient FHE operations with various bit-security levels than conventional hardware, especially <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$4.3\times $ </tex-math></inline-formula> better bootstrapping energy efficiency than state-of-the-art hardware.