Crypto Café - Bardia Taghavi - A Unified High-Performance NTT Architecture for ML-KEM and ML-DSA скачать в хорошем качестве

Crypto Café - Bardia Taghavi - A Unified High-Performance NTT Architecture for ML-KEM and ML-DSA

February 3, 2026, 10:00 am Science Building (SE-43), room 271 Speaker: Bardia Taghavi, Ph.D. student, Computer Engineering, Florida Atlantic University Title: Two Birds, One Stone: A Unified High-Performance NTT Architecture for ML-KEM and ML-DSA Abstract: With the recent NIST standardization of ML-KEM (Kyber) and ML-DSA (Dilithium), the focus of Post-Quantum Cryptography (PQC) research has shifted from algorithm selection to efficient hardware implementation. While both schemes rely on the hardness of the module-lattice hardness assumptions and utilize the Number Theoretic Transform (NTT) for fast polynomial multiplication, they operate over distinct algebraic rings (Rq with q=3329 for Kyber vs. q=8,380,417 for Dilithium). This discrepancy typically forces hardware designers to implement separate, redundant accelerators for each scheme, leading to suboptimal area efficiency. In this talk, I will present a novel, high-performance unified NTT architecture capable of accelerating both ML-KEM and ML-DSA within a single, reconfigurable hardware block. We will examine the mathematical structures that allow for a "dual-mode" Butterfly Unit (BFU) design, utilizing configurable modular reduction techniques (such as Barrett reduction or Montgomery multiplication) that adapt to either modulus at runtime.