Hardware Acceleration of NTT-Based Polynomial Multiplication in CRYSTALS-Kyber

CRYSTALS-Kyber is a promising post-quantum encryption candidate and has been selected for standardization. However, its operational efficiency faces challenges due to complex and time-consuming polynomial multiplication, which can be accelerated using number-theoretic transform (NTT). In this work, we propose a novel approach of hardware acceleration for NTT-based polynomial multiplication in CRYSTALS-Kyber on FPGA. Our approach leverages pipeline technology and optimized butterfly operation units with Montgomery and Barrett reductions, significantly improving computational efficiency. By running eight butterfly units in parallel, we achieve remarkably shorter computation cycles for key operations such as NTT, INTT, and PWM. Moreover, we introduce a dedicated Kyber algorithm PWM unit and optimized multi-RAM channel storage, greatly boosting memory access efficiency. This comprehensive optimization results in superior energy efficiency, surpassing other existing schemes and propelling the practical application of CRYSTALS-Kyber to new levels of efficiency.