CUBA: A Configurable Unified Butterfly Architecture for Lattice-based Cryptosystems

Lattice-Based Cryptography (LBC) is the most promising candidate for developing Post-Quantum Cryptosystems (PQC). A polynomial multiplication is the main performance bottleneck in most LBC schemes. The Number Theoretic Transform (NTT) is a pivotal tool to accelerate polynomial multiplication. This brief presents a novel Configurable Unified Butterfly Architecture (CUBA) for NTT and inverse NTT computations. The CUBA is developed to execute all butterfly operations in NTT/INTT and is programmable for different modulus sizes. It is implemented on AMD Artix-7 and Virtex-7 FPGA platforms. The implementation results are shown for common modulus sizes required in CRYSTALS-Kyber, CRYSTALS-Dilithium, many Fully Homomorphic Encryption (FHE), and newly submitted digital signature schemes. It delivers up to 26× improvement in latency area product with much higher throughput per area, and thus is suitable for many LBC-based PQC schemes.