Paper 2025/1190

Towards AI-driven Optimization of Robust Probing Model-compliant Masked Hardware Gadgets Using Evolutionary Algorithms

Abstract

Side-channel analysis (SCA) is a persistent threat to security-critical systems, enabling attackers to exploit information leakage. To mitigate its harmful impacts, masking serves as a provably secure countermeasure that performs computing on random shares of secret values. As masking complexity, required effort, and cost increase dramatically with design complexity, recent techniques rely on designing and implementing smaller building blocks, so-called “gadgets.” Existing work on optimizing gadgets has primarily focused on latency, area, and power as their objectives. To the best of our knowledge, the most up-to-date ASIC-specific masking gadget optimization frameworks require significant manual effort. This paper is inspired by previous work introducing open-source academic tools to leverage aspects of artificial intelligence (AI) in electronic design automation (EDA) to attempt to optimize and enhance existing gadgets and overall designs. We concentrate on evolutionary algorithms (EA), optimization techniques inspired by biological evolution and natural selection, to find optimal or near-optimal solutions. In this regard, our goal is to improve gadgets in terms of power and area metrics. The primary objective is to demonstrate the effectiveness of our methods by integrating compatible gates from a technology library to generate an optimized and functional design without compromising security. Our results show a significant reduction in power consumption and promising area improvements, with values reduced by 15% in some cases, compared to the naïve synthesis of masked designs. We evaluate our results using industry-standard synthesis and pre-silicon side-channel verification tools.