Kyber Accelerator on FPGA Using Energy-Efficient LUT-Based Barrett Reduction

Abstract

Contemporary cryptographic algorithms such as RSA and ECC can collapse due to the development of quantum computing. NIST has announced the first standardized postquantum cryptography algorithms that are resilient to the quantum computer’s attack, among which CRYSTALS-KYBER is the only algorithm for public-key encryption. This algorithm uses the number theoretic transform (NTT) to reduce the time complexity of the polynomial multiplication. However, NTT itself is still a big computational bottleneck of the entire algorithm. Also, modulo reduction consumes almost half of the energy of the butterfly unit for processing NTT. This paper presents a Kyber accelerator using LUT-based Barrett reduction, which reduce the energy for modulo operation by 40.6% from the original Barrett reduction. The test system is implemented on the Xilinx Zynq UltraScale+ ZCU104 and shows 113.3x speedup for NTT and 1.78x speedup for Kyber compared to the original CPU.