UPP-NTT: A Unified Parallel Pipelined NTT Accelerator for CRYSTALS-Kyber Post-Quantum Cryptography
Ramyavani P, Tanay Reddy, Archith Casheekar, Amit Kumar Panda, G. Ashwini Kumari, Subhendu Kumar Sahoo · IEEE Access · 2026
In lattice-based post-quantum cryptographic schemes, the Number Theoretic Transform (NTT) is a core computational operation that enables efficient polynomial multiplication in algorithms such as CRYSTALS-Kyber. This work presents a Unified Parallel Pipelined NTT (UPP-NTT) accelerator optimized for CRYSTALS-Kyber. The proposed architecture integrates four parallel pipelined radix-2 butterfly functional units (BFUs) within a single reconfigurable datapath that supports both Forward NTT (FNTT) and Inverse NTT (INTT) operations. A shift-and-add-based Barrett reduction unit reduces dependence on DSP slices, while a conflict-free single-memory organization with a dedicated address generator enables deterministic access to coefficients and twiddle factors. These design choices result in a fully pipelined and parallelized architecture well suited for post-quantum computation. Implemented on a Virtex-7 FPGA, the proposed UPP-NTT completes one forward and one inverse NTT operation in 470 clock cycles at 166 MHz, corresponding to a latency of 2.8μs. Compared with the Flexible Parallel NTT (FP-NTT) and Unified NTT (Unif-NTT) architectures, the proposed design achieves a 3–3.6× reduction in execution time and up to 3.8× and 4.5× improvements in throughput per area, respectively. Furthermore, when evaluated against the same FP-NTT and Unif-NTT baselines, the architecture demonstrates approximately a 60% reduction in average area–time product (Avg-ATP), confirming its scalability and efficiency for FPGA-based post-quantum cryptographic accelerators.