Deep Learning Based Graph Neural Network Technique for Hardware Trojan Detection at Register Transfer Level

The presence of Hardware Trojans (HTs) poses a significant risk to the integrity and security of integrated circuits (ICs), as they can compromise the functionality and reliability of electronic systems. Traditional HT detection techniques rely heavily on reference designs, which limits their applicability and effectiveness. In this paper, a golden reference-free HT detection technique using Graph Neural Network (GNN) at the Register Transfer Level (RTL) is implemented. The proposed methodology leverages the power of GNNs to capture intricate relationships among various components and signals in an RTL design. The graph-based representation enables the GNN to effectively detect anomalous patterns introduced by HTs. In the proposed work, a comprehensive dataset is generated ab-initio comprising both Trojan-free and Trojan-infected designs. It includes 422 circuits which has 262 trojan free and 160 trojan infected circuits, considered for extracting the GNN based model. The proposed model learns to distinguish between normal and Trojan-infected designs, achieving high detection accuracy. The outcome indicates that that the GNN model detects unknown HT with 100 % recall, making it feasible for practical deployment in large scale design.