基于朴素贝叶斯分类器的硬件木马检测方法

摘要

在侧信道分析的基础上,针对芯片中存在的硬件木马,提出一种基于朴素贝叶斯分类器的硬件木马检测.该方法能够利用训练样本集构建分类器,分类器形成后便可将采集到的待测芯片功耗信息准确分类,从而实现硬件木马检测.实验结果表明,对于占电路资源1.49％和2.39％的两种木马,贝叶斯分类器的误判率仅为2.17％,验证了该方法的有效性和适用性.此外,在与欧氏距离判别法比较时,基于朴素贝叶斯分类器的方法表现出了更高的判别准确率,同时也具有从混杂芯片中识别出木马芯片与标准芯片的能力,这又是马氏距离判别法所不具备的.%This paper proposed a naive Bayesian classifier to detect the hardware Trojan based on the side-channel analysis,and used the power consumption characteristic to train the classifier.After the completion of the classifier construction,it was able to identify the integrated circuits precisely and classify the circuits under test into different categories.So it could be used to detect the hardware Trojan.Experimental results show that the naive classifier can find out a hardware Trojan which occupies only 1.49％ of the AES module with 2.17％ misjudgment rate.Besides,results of the comparison with the Euclidean distance method show ahigher accuracy rate of discrimination,and the advantage of the naive Bayesian classifier which is able to identify the genuine circuits and the Trojan circuits from the hybrid circuits,is not available from the Mabalanobis distance discrimination method.