A Deep Learning Approach for Hardware Trojan Detection in Netlist of Integrated Circuits with Graph Neural Networks

Ensuring the integrity and security of integrated circuits is paramount in today's digital age, as hardware Trojans (HT) pose a significant threat to both critical systems and sensitive data. In this paper, a deep learning approach for HT detection using graph neural networks (GNNs) in list-based designs is presented. GNN’s ability to capture intricate patterns and relationships within the data is utilized by modeling the circuit as a graph, with components as nodes and their connections as edges. The approach focuses on learning the inherent structure of the Netlist, thereby enabling the automatic identification of anomalous and potentially malicious insertions. The proposed GNN model shows on an average accuracy of 99.6%, recall of 95.2%, F-measure of 96.4%, and precision of 100%. The above parameters are obtained across different designs considered in the present paper.