Abstract: Aiming at the existing problems such assignal redundancy and high dimension problems in Hardware Trojan detection based on the side-channel analysis, this paper explores the feasibility of feature selection method in removing redundancy and reducing the dimension of side-channel signal, then puts forward a feature selection method using in-class distance and between-class distance as class divisible criterion of feature selection method and processing the side-channel in advance. First, this paper analyzes the feature selection problem in side-channel, then elaborates the class separability criterion based on in-class distance and between-class distance and feature selection algorithms. In the end, implants the Trojans into the FPGA crypto chip, and test it using the K-L method, through the comparison before and after the detection effectinside-channel signal feature selection, this methods can help to distinguish side-channel signal statistical characteristic difference between“gold chip”（without Trojans implanted） and Trojan chip（contain Trojan）, and better implementation the Hardware Trojan Detection.

Key words: hardware Trojan detection, side-channel analysis, feature selection, in-class distance and between-class distance

摘要： 针对基于旁路分析的硬件木马检测中存在的旁路信号冗余以及高维问题，探究特征选择方法在去除冗余、降低旁路信号维数方面的可行性，提出了一种以类内类间距离作为可分性判据的特征选择方法对旁路信号进行预先处理。首先分析了IC芯片旁路信号的特征选择问题，然后阐述了基于类内类间距离的可分性判据以及特征选择搜索算法，最后在FPGA密码芯片中植入硬件木马，并基于K-L方法进行检测实验，通过对旁路信号进行特征选择前后的木马检测效果对比发现，该特征选择方法能有助于分辨出无木马的“金片”与含木马芯片之间旁路信号的统计特征差异，更好地实现硬件木马的检测。

关键词: 硬件木马检测, 旁路分析, 特征选择, 类内类间距离