Detection of Hardware Trojan Horse using Unsupervised Learning Approach

Machine Learning helps in detecting hardware trojan and are successful to a large extent to analyze. The supervised learning technique cannot be applied to huge netlist to categorize the samples. This is with respect to two categories namely infected and Trojan free. Such method is complex. So, the usage of unsupervised approach gives good prediction performance without any labeled data. The unsupervised approach works well with the large number of netlists. This work helps in the detection of trojan using a technique called PCA (Principal Component Analysis) which deals with dimensionality reduction and helps in ease of computation. This is followed by unsupervised clustering algorithm from which the detection of hardware trojans is performed. A comparative analysis of k-means clustering and hierarchical clustering is also carried out. The method is evaluated on Trust-Hub circuits and the results highlight the improvement obtained.