关键词: 差分分析, 卷积神经网络, DES加密算法, 旁路攻击

Abstract: Due to the large sample size requirement for traditional Differential Power Analysis（DPA） methods and the problem of high computational resource consumption and long training cycles for the deep learning-based side-channel template attack, the implementation principle of convolutional neural networks is introduced. With the technical characteristics and theoretical analysis of the traditional differential energy analysis method in the actual attack, there is a Ghost peaks phenomenon and the appropriate data classification rules are selected. Based on the convolutional neural network, a new method of differential attack is proposed for encryption chips. By performing differential analysis and comparison experiments on the microcontroller running the DES encryption algorithm（AT89C52）, the experimental results show that the new method has greatly improved on the sample size requirements, and the new method does not need to continuously increase the number of iterations. Meanwhile, the new method has been optimized in terms of computing resource consumption and training cycle.

Key words: differential analysis, convolutional neural network, DES encryption algorithm, side-channel attack