High-Performance Implementation of Crystals-Dilithium using DSP-Free NTTs on FPGA

With the release of the NIST post-quantum cryptography (PQC) standard ML-DSA (Dilithium), achieving high-performance signature algorithms on resource-constrained hardware devices, such as low-end FPGAs, has become a research trend. However, most of the existing solutions either rely on scarce Digital Signal Processing (DSP) or use logic-only implementations that suffer from large area and poor timing. To address the limitations, we propose a high-performance deeply pipelined modular-multiplier and a dilithium implementation. First, we propose an optimized encoding scheme that combines Multiple Constant Multiplication (MCM) with Canonical Signed Digit (CSD) encoding to eliminate DSP dependency and to maximize resource reuse; Secondly,we derive the accurate upper bound via modulus properties for a compact, single-cycle parallel reduction circuit and remove carry-chain delays. Lastly, we integrate an improved fully pipelined Radix-2 Multipath Delay Commutator NTT (MDC-NTT) to boost frequency. Evaluated On Xilinx Artix-7, compared to DSP-based schemes, our DSP-free NTT multiplier achieves a peak frequency of 186 MHz and improves the average ATP by over 40%. Compared to SOTA DSP-free design, our design boosts 29.2% frequency along with comparable ATP.