Analysis and Hardware Optimization of Lattice Post-Quantum Cryptography Workloads

The mathematical constructs, nature of computations and challenges in optimizing lattice post-quantum cryptographic algorithms on modern many-core processors are discussed in this paper. Identification of time-consuming functions and subsequent hardware optimization using vector units and hardware accelerators of one of the candidates, CRYSTALS-Kyber, leads to performance improvement of around 52% for its SHA3 variant and 83% for its AES variant. Detailed Cycles-per-Instruction (CPI) stack breakdown before and after optimization indicates a CPI of around 0.5 and dominance of load/store operations in these workloads.