https://aurivelle.github.io/tags/arm-cortex-m4/ARM Cortex-M4Hugo Blox Builder (https://hugoblox.com)en-usThu, 02 Jul 2026 00:00:00 +0000https://aurivelle.github.io/media/icon_hu_982c5d63a71b2961.pnghttps://aurivelle.github.io/tags/arm-cortex-m4/https://aurivelle.github.io/projects/m4optimization/Thu, 02 Jul 2026 00:00:00 +0000https://aurivelle.github.io/projects/m4optimization/<p>This project is the final project for <strong>Cryptography Engineering (2025 Fall)</strong> at National Taiwan University, lectured by Matthias J. Kannwischer. It focuses on aggressive performance optimization of two modern cryptographic primitives on resource-constrained embedded systems.</p> <p>The project implements highly optimized versions of:</p> <ul> <li><strong>ML-KEM (FIPS 203):</strong> NIST&rsquo;s lattice-based post-quantum key encapsulation mechanism</li> <li><strong>X25519 (ECDH25519):</strong> Elliptic Curve Diffie-Hellman key exchange</li> </ul> <p><strong>Target Platform:</strong> ARM Cortex-M4 microcontroller<br> <strong>Constraints:</strong> Constant-time execution, strict code-size budgets for NTT routines</p> <h3 id="key-optimizations">Key Optimizations</h3> <h4 id="x25519-ecdh25519">X25519 (ECDH25519)</h4> <ul> <li><strong>Finite Field Arithmetic (mod p = 2^255 - 19):</strong> Hand-optimized assembly for <code>fe25519_mul</code> (field multiplication) and <code>fe25519_sqr</code> (field squaring), achieving 81-85% cycle reduction.</li> <li><strong>Constant-Time Scalar Multiplication:</strong> No secret-dependent branches or table indexing to prevent timing side-channel attacks.</li> <li><strong>Fixed-Base Acceleration:</strong> Window method with precomputed tables using constant-time selection, reducing fixed-base scalar multiplication by up to <strong>94%</strong> against the course baseline.</li> </ul> <h4 id="ml-kem-fips-203">ML-KEM (FIPS 203)</h4> <ul> <li><strong>NTT/iNTT Kernels:</strong> Optimized Number-Theoretic Transform operations with reduced memory traffic, layer fusion, and loop unrolling, achieving <strong>2.85-3.79x speedup</strong>.</li> <li><strong>Polynomial Arithmetic (mod q = 3329):</strong> Leveraged ARMv7E-M DSP instructions (SIMD 16-bit operations) for packed polynomial addition and subtraction, achieving <strong>2-2.2x speedup</strong>.</li> <li><strong>Code-Size Budgeting:</strong> NTT routines implemented under strict per-function size constraints while maintaining performance.</li> <li><strong>Course-Baseline Result:</strong> Reduced the cycle count of ML-KEM <code>poly_ntt</code> by up to <strong>62%</strong> under the same benchmarking setup.</li> </ul> <p>This project demonstrates deep understanding of low-level optimization, cryptographic algorithm engineering, and secure implementation practices on embedded ARM platforms.</p> <ul> <li><strong> </strong></li> </ul>