Using Path Features for Hardware Trojan Detection and Localization Based on Machine Learning Techniques

• Main Authors: Chen, Jian-You, 陳建佑
• Other Authors: Wu, Kai-Chiang
• Format: Others
• Language: en_US
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/ep24r5

碩士 === 國立交通大學 === 資訊科學與工程研究所 === 107 === In this work, we proposed a hardware trojan detection method by using path features and machine learning techniques to localize the trojan gates of the untrusted circuits. First of all, we extract the \textit{path features} which are highly relative to the malicious trojan circuit. In order to reduce the complexity of path counts in the training set of data and the classification/validation set of data, we use \textit{path filter} to filter out the paths which are very unlikely to be the trojan paths. For our machine learning models, we use the random forest (RF) and support vector machine (SVM) as our path classifiers. Due to the promising result of trojan path classification, we score the \textit{suspected trojan gates} which are on the \textit{suspected trojan paths} identified by the path classifier. Finally, we rank the suspected trojan gates in decreasing order of score, and show how suspicious a gate is in terms of being a trojan gate. On average, we obtain 94.57\% true positive rate (TPR) and 98.54\% true negative rate (TNR) of the \textit{trojan gate localization} of all trojan circuits.