Skip to content

Implementation and Evaluation of NTT Arithmetic for ML-KEM on a CGLA

Sep 2026 · arXiv.org · Vol abs/2609.28904 · 0 citations · 25 references
Computer Science

TL;DR

An eight-stage cyclic radix-2 driver over the ML-KEM modulus is implemented by splitting each twiddle into 8-bit and 4-bit parts before modular reduction, differs from the standardized seven-layer incomplete negacyclic NTT and does not implement the full ML-KEM polynomial multiplication path.

Abstract

FIPS 203 standardizes ML-KEM for post-quantum key establishment. Its polynomial multiplication relies on NTT butterflies with exact modular arithmetic over q = 3329. Dedicated NTT accelerators minimize latency with fixed modular arithmetic and stage schedules. A CPU-Grounded Linear Array (CGLA) reuses one programmable linear datapath across several workloads. Mapping the transform to this datapath requires exact FP32 reconstruction and explicit stage transitions across the ARM-to-CGLA interface. We implement an eight-stage cyclic radix-2 driver over the ML-KEM modulus by splitting each twiddle into 8-bit and 4-bit parts before modular reduction. This driver differs from the standardized seven-layer incomplete negacyclic NTT and does not implement the full ML-KEM polynomial multiplication path. The arithmetic sequence keeps every integer below 2^24. One 41-PE call fuses the first two radix-2 stages, and six 47-PE calls execute the remaining stages while scattering outputs into next-stage records. Across four cohorts, 105 FPGA runs match all 817152 output coefficients. At batch size 64, measured FPGA end-to-end latency is 27.5 us per NTT, and the ASIC projection is 6.74 us. With PE gating, projected ASIC system energy is 10.1 uJ per NTT at batch size 8 and 58.9 uJ at batch size 64.

View source

Similar papers

Conference Aug 2026

A One-Bitstream KV260 Accelerator for ML-KEM-512 and ML-DSA-44 with Split KEM Datapaths and Sign-Path Early Rejection

Post-quantum hardware implementation is moving from accelerating a single primitive or arithmetic kernel to executing multiple standardized primitives in one security flow. Existing FPGA studies have optimized Kyber/ML-KEM, Dilithium/ML-DSA, and number theoretic transform (NTT) kernels separately, but many results do n...

Chao-Zheng Li, Ding Deng, Lu Li et al. · 0 citations
Preprint Oct 2026

System-Level Optimization Beyond Cryptographic Kernels: An ML-KEM Case Study on Arm Cortex-M7

Recent work on embedded post-quantum cryptography has focused primarily on instruction-level optimization, including arithmetic-kernel improvements, assembly tuning, register allocation, and instruction scheduling. Using the Module-Lattice-Based Key-Encapsulation Mechanism (ML-KEM) on an Arm Cortex-M7 as a case study,...

Mahmoud Abdelhafeez Sayed, Mostafa M. I. Taha, Gurp Nijjer · 0 citations
Conference Aug 2026

SAPA: A Programmable Shared Polynomial Arithmetic Accelerator with Semantic Handoff Validation for Staged ML-KEM/ML-DSA Flows

ML-KEM and ML-DSA are dominated by the same modular polynomial operators, which makes a shared FPGA datapath attractive. However, an accelerator that runs both primitives back to back has a failure mode that arithmetic checking alone cannot catch: a command can be locally correct yet bound to the wrong cross-command st...

Lei Peng, Ya-Lin Wang, Yun-Bo Jia et al. · 0 citations
Conference Aug 2026

A Highly Flexible and Reconfigurable Instruction-Driven Hardware Acceleration System for ML-DSA

The rapid advancement of quantum computing severely threatens conventional public-key infrastructure, promoting the migration to Post-Quantum Cryptography (PQC). With NIST standardizing Module-Lattice-Based Digital Signature Algorithm (ML-DSA) as FIPS 204, the high computational complexity of lattice-based cryptography...

Xiao Liu, Qing-Xin Xie, Hui-Hong Zhang et al. · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.