A Portable Hardware Trojan Detection Using Graph Attention Networks
Han Zhang, Yinhao Zhou, Ying Li · 2023
Among all hardware security threats, the malicious circuits surreptitiously inserted into third-Party Intellectual Property (3PIP) cores, known as Hardware Trojans (HTs), is one of the main concerns. The early discovery of HTs is crucial because any countermeasure after the fabrication process would be expensive or unavailable. Graph Neural Networks (GNN), with the intuitive graph representation of a hardware design, has emerged as a powerful technique to solve this problem. However, most of these networks have two limitations, including the loss of dynamic structural features and the portability issue of using trained models in other designs. To this end, we address such limitations by proposing a new edge-featured Data Flow Graph (DFG) generation method that combines circuit structures with simulation data to establish HTs detection based on Graph Attention Networks (GAT). The solution is utilizing GAT to extract the HTs features from DFG, then identifying the known and unknown HTs hidden in different circuits. We evaluate this methodology on our dataset by expanding Trusthub HTs benchmarks. The results show that our approach can realize the HTs detection with high recall and precision in a short time. Compared with the previous method, our method has more scalable capability and excellent prospects in HTs detection.