An Efficient Reverse Engineering Hardware Trojan Detector Using Histogram of Oriented Gradients

作者：Nasr Abdurrahman A^*; Abdulmageed Mohamed Z

来源：Journal of Electronic Testing-Theory and Applications, 2017, 33(1): 93-105.

DOI：10.1007/s10836-016-5631-z

• Hardware Trojan
• Automatic feature selection
• Histogram of oriented gradient
• AdaBoost
• Classifier cascade

The problem of hardware Trojan is certainly serious, complex and tricky. Therefore, hardware Trojan (HT) detection is difficult, time and effort consuming and challenging due to non-trivial threats that compromise the security of integrated circuits (IC). The problem becomes more serious with the extensive outsourcing of ICs design and fabrication by untrusted foundries. Recently, hardware Trojan detection has gained insight and interest from different researches and industries in order to detect Trojan horses in the fabrication phase of an IC. In that phase HT detection requires expensive testing techniques. Hardware Trojans in the fabrication phase show that the detection process may be either side-channel based (non-destructive) or reverse engineering based (destructive). The destructive approach consists mainly of three steps: (1) decapsulation, (2) delayering and (3) imaging for layout identification. This paper presents a new approach for automating the third step, namely, the layout identification of the underlying circuit. The proposed approach automatically extracts and describes the features of circuit layout by making use of histogram of oriented gradient. Features descriptors obtained from such histogram of oriented gradient are then fed to a machine learning classifier represented by a decision tree that aims at learning the pattern of malicious ICs in order to differentiate them from benign ones. In addition, the classification result is enhanced by utilizing AdaBoost learning algorithm to produce a strong meta-classifier in a chain of cascaded stages. Based on that scheme, a composite classification model is built up to provide an Automatic Hardware Trojan Detection and description Tool (AHTDT). The proposed model has been tested (on noisy and clean data) and evaluated using ISCAS89 benchmark dataset. Such benchmark is emphasized deliberately to show different Trojan examples -namely, Trojan insertion, Trojan deletion and Trojan parametric- inside hardware circuits. Model simulation and evaluation results have shown a remarkable enhancement in HT detection compared to other reverse engineering detection techniques. Moreover, the proposed approach has the advantages of being automatic, systematic and capable of detecting and diagnosing hardware Trojans accurately with high detection rate, while keeping low false positive rate.

• 出版日期2017-2