Hardware Trojan Detection Combines with Machine Learning: an Isolation Forest-based Detection Method

Abstract

Hardware security has recently become a research hotspot due to the separating trend of integrated circuit design and manufacturing in the global industry chain. Among all security threats, the hardware trojan (HT) inserted into the integrated circuit is one of the most concerns. As a flexible and nondestructive approach, the side-channel analysis techniques are widely used for HT detection. However, noise sensitivity and golden chip dependence make side-channel based approaches face greatly challenge in practice. In this paper, we propose a novel HT detection approach based on Isolation Forest, which provides high precision detection capability, while reducing the dependence of the golden chip. First, we perform special preprocessing, named Main Difference Accumulation (MDA), on the side-channel information to magnify the impacts of HT. Then, we train the isolation forest model using the processed information. During the test phase, the obtained model can score the test data. Finally, we complete the detection of HT based on the score and threshold. Moreover, we use the K-means clustering algorithm to achieve the automatic selection of the threshold in the isolation forest. To verify the feasibility of the proposed approach, we select the power signal as the side-channel information and insert four different kinds of HTs into the AES-128 crypto core. Results obtained using a Xilinx Spartan-6 FPGA board show that the accuracy of HT detection can reach 100%, with a 1.13% relative area ratio of the original circuit.