A XGBoost based Hybrid Detection Scheme for Gate-Level Hardware Trojan

Abstract

In recent years, with the prosperity and globalization of the semiconductor industry, hardware security issues have become another major problem besides software security. To reduce the development costs, integrated circuits are often outsourced to various external vendors. This provides malicious third-party vendors a good chance to implant Trojans in the circuit design and manufacture phases. It is crucial to find effective solution to hardware Trojan detection. This paper proposes a hybrid hardware Trojan detection method for gate-level netlist based on XGBoost algorithm. The detection method affirms each network as a node and aims to detect all Trojan networks existing in the circuit. Firstly, according to the difference between the features of the Trojan circuit and the normal circuit, the effective features of the circuit are extracted. Then the XGBoost algorithm is applied to train the optimum feature set and classify the circuit into suspicious Trojan circuits and normal circuits. To obtain a higher accuracy, the dynamic detection method is applied to make up for the shortcomings of the static detection. The experimental results on the Trust-HUB benchmarks show that the detection method can achieve accurate detection of hardware Trojan circuits. The detection method can finally achieve 90.9% average True Positive Rate (TPR) and 99.0% average True Negative Rate (TNR).