A power traces based hardware trojan detection using deep artificial neural network

Abstract

To establish trust and security in integrated circuits manufacturing and by considering the third-party vendors, a novel hardware trojan detection method employing a deep artificial neural network is proposed in this work. The power consumption traces are extracted as features from the ISCAS’89 benchmark circuits. The proposed deep artificial neural network proves to be efficient with good performance and minimal loss. The ANN model developed behaves ideally for the s444 benchmark circuit with an accuracy of 100% and a negligible model loss of 0.0074. From the experiments conducted independently for various benchmark circuits, this proposed neural network model outperforms the existing power-related hardware trojan detection methods by achieving an overall accuracy of 95.76%, recall of 94.24%, and precision of 97.13%.