Hunting for Hardware Trojan in Gate Netlist: A Stacking Ensemble Learning Perspective
Hong Liang, Ge Zhu, Jing Zhou, Xuefei Li, Ziyi Chen, Wei Hu · 2023
As hardware designs become more complex and incorporate a wider variety of third-party IP cores, the risk of hardware Trojan insertion increases. While several techniques have been introduced to detect hardware Trojans, machine learning-based detection methods rely on a proper feature selection algorithm and learning model. This paper presents a hardware Trojans detection method for gate-level netlists using the stacking ensemble learning method. The study analyzes the essential attributes and common structures of hardware Trojans and proposes inv_x as inverter features to detect ring oscillator circuits. By incorporating the proposed features with the existing attributes, the feature set comprises 56 features. Thus, a hybrid feature selection method that employs Random Forest (RF) and Recursive Feature Elimination with Cross-Validation (RFECV) is proposed to identify the optimal feature set To improve Trojan detection’s accuracy, a stacking ensemble learning model is developed by integrating multiple machine learning classifiers. Finally, we utilize multidimensional data visualization and distance measurement to analyze the detection results. Experimental evaluations using Trust-Hub benchmarks show promising Trojan detection results with the average TPR of 94.15%, the average Fscore of 95.36%, and the average Precision of 96.93%. These results demonstrate a significant improvement over existing hardware Trojan detection methods.