AegisGAN: A Generative Adversarial Network Approach for Efficient and Adaptive Hardware Trojan Detection

Detecting Hardware Trojans (HTs) in integrated circuits presents a significant challenge due to their stealthy activation mechanisms, which rely on rare signal combinations. Traditional detection methods, such as side-channel analysis and statistical test pattern generation, struggle with scalability, high computational costs, and the reliance on “golden reference” circuits, making them unsuitable for modern complex hardware designs. To address these limitations, we introduce AegisGAN, a generative adversarial network (GAN)-based framework that not only generates optimized input patterns to activate rare signals associated with HTs but also eliminates the need for a “golden reference” design. Unlike existing methods, AegisGAN adapts to varying circuit complexities through an innovative adaptive training mechanism, which enhances its flexibility and robustness. Our approach significantly reduces the computational overhead, offering a more scalable, efficient, and accurate solution for HTs detection. Experimental results on standard benchmarks show that AegisGAN achieves 86.6% rare nets coverage and reduces test length by 4.2<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">x</sup>, demonstrating its superiority over traditional methods in real-world IC security applications.