GATrojan: An Efficient Gate-level Hardware Trojan Detection Approach Using Graph Attention Networks
Sen Wang, Yijun Cui, Shichao Yu, Chongyan Gu, Chenghua Wang, Weiqiang Liu · 2024
With the globalization of the semiconductor industry, the utilization of third-party Intellectual Property(3PIP) cores has become prevalent. However, this widespread adoption of 3PIP cores has risen the risk of Hardware Trojan(HT). This paper proposes a gate-level HT detection method based on Graph Attention Networks(GAT) named GATrojan. It addresses the challenges arising from the utilization of 3PIP cores in the semiconductor industry. GATrojan can detect Hardware Trojans (HTs) at the gate-level without relying on golden reference model circuits. By leveraging GAT and supervised learning, circuit representations can be effectively learned using directed acyclic graph(DAG). To validate the proposed detection methods, GATrojan used different benchmarks to do the validation. Experimental results show that on different Trust-Hub HT benchmarks, GATrojan achieved ${9 2. 1 7 \%}$ F1 score, ${9 9. 4 2 \%}$ TNR, and ${9 6. 0 4 \%}$ TPR.