基于Alexnet卷积神经网络的加密芯片模板攻击新方法

A New Template Attack Method for Encryption Chip Based on Alexnet Convolutional Neural Network

摘要:

针对经典高斯模板攻击存在的问题，在分析了卷积神经网络方法具有的优势的基础上，提出了一种基于卷积神经网络的加密芯片旁路模板攻击新方法。该方法可以有效地处理高维数据，且可以通过不断地调整网络权值与偏置实现对数据无限逼近，明确分类的精确关系，提高模板刻画精度。最后选取AT89C52微控制器（单片机）运行的AES加密算法第一轮异或操作为攻击点，与传统的模板攻击进行了对比实验，实验结果表明：虽然在匹配成功率方面稍低于传统的模板攻击，模型结构和超参数仍需要进一步优化，但新方法在处理高维特征点方面较传统的模板攻击具有较大优势。

Abstract:

Aiming at the problems of classical gauss template attacks, based on the analysis of the advantages of convolutional neural network, a new method based on convolution neural network for encrypting chip side-channel template attack is proposed. This method can deal with high-dimensional data effectively, and it can achieve infinite approxima -tion of data by adjusting weights and biases constantly, and clarify the precise relationship between classifications, and improve the accuracy of template characterization.Selects the AT89C52 microcontroller (MCU) running AES encryption algorithm first round XOR operation is the point of attack, compared with traditional template attacks. Experimental results show that although the success rate of matching is slightly lower than that of traditional template attacks, the model structure and hyperparameter still need to be further optimized. However, the new method has greater advantages in dealing with high-dimensional feature points than traditional template attacks.