A Configurable, Parallel and High-Performance NTT/INTT Accelerator for Post-Quantum Cryptosystems

For the post-quantum cryptosystems such as the CRYSTALS-Kyber, the number theoretic transform (NTT) and the inverse number theoretic transform (INTT) are usually adopted as a necessary approach to calculate polynomial multiplications. This paper presents an NTT/INTT accelerator with configurable number of processing element (PE). Varied parameter set (N, q) is supported in the accelerator. A storage-saving scheme for twiddle factors memory is proposed, requiring N/2 storage space only. Implementations under different parameter sets (N, q) on FPGA are described, showing that about $\mathbf{2 5 \%} \sim \mathbf{6 0 \%}$ area-time reduction over many other existing works can be achieved.