Threshold Analysis Using Probabilistic Xgboost Classifier for Hardware Trojan Detection

The fabless nature of integrated circuits manufacturing leaves them vulnerable to modifications by ill-intentioned third party. There arises a necessity for security measures during their manufacturing to protect them from covert modifications known as hardware Trojans. Static analysis of gate-level synthesized integrated circuits can prove helpful in detecting the presence of unwanted circuitry within the host. This paper proposes a static analysis technique of gate-level integrated circuits using supervised probabilistic classifier through effective threshold analysis. New and existing relevant features are extracted that relates to hardware Trojan properties and normalised accordingly. Effective features are selected using their feature importance values. Variance threshold has been used to create a high entropy feature subset to train a supervised model using XGBoost algorithm with relevant hyperparameters. Threshold values of the probabilistic classifier are determined through analysis of threshold obtained using receiver operating characteristic and precision-recall curves. The chosen techniques showcase hardware Trojan detection with high accuracy over gate-level synthesized circuits.