A Novel Golden Models-Free Hardware Trojan Detection Technique Using Unsupervised Clustering Analysis

Abstract

Recently, hardware Trojan has become a major threat for integrated circuits. Most of the existing hardware Trojan detection works require golden chips or golden models for reference. However, a golden chip is extremely difficult to obtain or even does not exist. In this paper, we propose a novel hardware Trojan detection technique using unsupervised clustering techniques. The unsupervised clustering technique can obtain the structure information of the set of unlabeled ICs, and then distinguishes the suspicious ICs from the ICs under test. We formulate the unsupervised hardware Trojan detection problem into two types of detection models: partitioning-based and density-based detection model. We also propose a novel metric to determine the labels of the clusters. Compared with the state-of-the-art detection methods, the proposed technique can work in an unsupervised scenario with no need of ICs’ prior information. It does not require fabricated golden chips or golden models. We perform simulation evaluation on ISCAS89 benchmarks and FPGA evaluation on Trust-HUB benchmarks. Both evaluation results show that the proposed technique can detect infected ICs in the unsupervised scenario with a good accuracy.