Hardware/Software Co-Design of a Multi-Mode CRYSTALS-Kyber Accelerator for Quantum-Secure Internet-of-Things Systems

CRYSTALS-Kyber, briefly called Kyber, is an asymmetric encryption algorithm standardized by NIST to maintain data security in the face of the rapid development of quantum computers, which significantly threaten traditional cryptographic systems such as RSA and ECC. Accordingly, multiple hardware studies have been conducted for Kyber to optimize performance and energy efficiency for modern systems such as the Internet of Things. However, most existing Kyber hardware still faces challenges in achieving high speed with high hardware efficiency and mainly focuses on standalone cores without developing practical embedded systems. Therefore, this paper proposes a hardware/software co-design for a multimode CRYSTALS-Kyber accelerator (CKA) to support all operating modes, including key generation, encryption, and decryption, with high speed and low power. Three innovative ideas are proposed for CKA to achieve these goals: resource-sharing modules to reuse almost all main modules in three operating modes, a Unified NTT/INTT module to optimize the speed and area of the most complex and time-intensive module, and pipelined pointwise multiply-accumulate and modified Barrett Reduce to accelerate the complex multi-plication operation. Notably, compared to NTT/NTT works, our Unified NTT/INTT is also 2.14-8.73 times better in area-delay product. The CKA is successfully implemented on the Zynq UltraScale+ MPSoC ZCU102 FPGA, where the power-delay product of CKA is at least 12.3 times better than powerful CPUs. Compared to previous Kyber architectures in three operating modes, the CKA is 1.37-225.26 times and 1.9-166.2 times better in execution time and area-delay product, respectively.