A New Hardware Trojan Detection Technique using Class Weighted XGBoost Classifier

The involvement of external vendors during the different phases of the integrated circuits (IC) allows the adversaries to insert the hardware Trojan (HT). Existing HT detection techniques fail to detect Trojans accurately, time-consuming, does not handle class imbalance problem efficiently, use tree-based feature importance, and possess large false positive and negative rates. Therefore, to overcome these problems, this paper proposes a new class weighted Extreme Gradient Boosting (CW-XGB) based HT detection technique, which detects Trojans from gate-level netlist using the best set of SCOAP feature values. A weighting scheme is proposed in the CW-XGB model to tackle class imbalance problem by assigning higher weights to minority Trojan-inserted class, thus removing the need to apply the oversampling/synthetic data algorithm. Moreover, the proposed model automatically addresses the overfitting issue by incorporating the regularization in the loss function. Besides, we also employ early stopping to avoid the over-training of the proposed model. Further, a new feature selection method is proposed, which selects the best set of SCOAP features based on permutation feature importance values, and it also avoids the continuous retraining of the model. Finally, a new HT detection algorithm is proposed that accurately detects HT from gate-level netlist using the best set of features and proposed CW-XGB model. Experimental results on Trust-Hub benchmarks shows that the proposed HT detection technique provides on an average 99%, 98.86%, and 99% accuracy, precision, and recall, respectively. Further, it provides on an average 98.76% and 98.73% F-measure and ROC-AUC score, respectively.