HTrans: Transformer-based Method for Hardware Trojan Detection and Localization

Hardware Trojan (HT) is a malicious code intentionally inserted into the original circuit design to modify the original function, leak information or decrease the performance. Circuit fabrications have increased Third-Party Intellectual Property (3PIP) usage with market pressure and the increasing global economy. Consequently, hardware may become vulnerable to a wide range of attacks at some stage of the manufacturing process, making detecting HT a necessary procedure. HT detection in the early stage is crucial because removing HT and re-designing the circuit later or after fabrication could be expensive. In this work, we propose a novel Transformer-based Method for pre-silicon HT detection and localization called HTrans. We innovatively use Graph Convolutional Network (GCN) as a preprocessing stage before the Transformer, giving our model the scalability to any design size. Experiments on the Trusthub benchmark show that our model achieves an average of 96.7% F1 score on HT detection and 91.7% accuracy on HT localization. In addition, HTrans can quickly complete the detection on the Register Transfer Level (RTL) within a second.