Performance Enhancement of Unsupervised Hardware Trojan Detection Algorithm using Clustering-based Local Outlier Factor Technique for Design Security

Internet of Things (IoT) has become extremely prominent for industrial applications and stealthy modification deliberately done by insertion of Hardware Trojans has increased widely due to globalization of Integrated Circuit (IC) production. In the proposed work, Hardware Trojan is detected at the gate level by considering netlist of the desired circuits. To mitigate with golden model dependencies, proposed work is based on unsupervised detection of Hardware Trojans which automatically extracts useful features without providing clear desired outcomes. The relevant features from feature dataset are selected using eXtreme Gradient Boosting (XGBoost) algorithm. Average True Positive Rate (TPR) is improved about 30% by using Clustering-based local outlier factor (CBLOF) algorithm when compared to local outlier factor algorithm. The simulation is employed on Trust-HUB circuits and achieves an average of 99.83% True Negative Rate (TNR) and 99.72% accuracy which shows the efficiency of the detection method even without labelling data.