Autonomous Hardware-Trojan Generation Method Using Reinforcement Learning for Random Forest-Based Hardware-Trojan Detection

In recent years, the risk of inserting malicious circuits called hardware Trojan (HT) in the IC design process has become a serious concern. Although machine learning methods are effective in detecting HTs, their performance is not guaranteed for more difficult HTs. In order to make practical use of HT detection methods, they must be able to adapt to the daily updates of hard-to-detect HTs, and it is quite necessary to efficiently generate them. In this paper, we propose an autonomous HT generation method using reinforcement learning for random forest-based HT detection. The proposed method uses reinforcement learning to generate HTs efficiently and autonomously. The reinforcement learning algorithm used repeatedly generates new HTs and observes the HT detection accuracy, and learns a policy for modifying parameters during HT composition in order to minimize the HT detection accuracy. From the evaluation experiments, we confirmed that the proposed method generates HTs that are difficult to detect.