An Efficient Machine Learning Approach for Hardware Trojan Detection

As outsourcing of IC manufacturing has become a global phenomenon, the risk of ICs being infested with Trojans has increased more than ever. In this paper, we propose a circuit gate level netlist based Trojan detection using supervised machine learning approaches. A number of features are extracted from the netlist that delivers a sophisticated dataset for the approximately trained model to deliver a higher true positive detection rate. The netlist is analyzed in 45 nm technology to generate features and Monte Carlo simulation is performed to generate two thousand virtual netlists, including one thousand Trojan infested netlists. We experiment with the s27 benchmark circuit netlist to evaluate our approach. Different types of sequential and combinational types of Trojans from literature are inserted into the netlist to evaluate the proposed approach. The results show significant Trojan detectability in different machine learning approaches.