摘要
由于深度学习对数据内在特征的敏感性,将深度学习算法应用于硬件加密芯片的侧信道分析,提高了侧信道分析的效率和准确率.但深度神经网络学习算法依旧是非线性结构未知的深层黑盒模型,模型结构和性能不一定是最优.该文提出一种基于树突网络的侧信道分析方法,由于树突网络内部非线性结构的可解释性,其系统辨识能力和运算复杂度均优于深度学习网络.在ChipWhisperer侧信道分析实验平台的CW308T-STM32F3和ATXMEGA128D4目标板上,针对AES-128加密算法进行侧信道分析实验,实验结果表明,基于树突网络的侧信道分析在模型参数规模、攻击精度、训练时间等方面都要优于多层感知机、卷积神经网络、循环神经网络等深度学习模型.
Due to the sensitivity of deep learning to the intrinsic characteristics of data,applying deep learning algorithms to the side channel analysis of hardware encryption chips improves the efficiency and accuracy of side channel analysis.However,the deep neural network learning algorithm is still a deep black box model with unknown nonlinear structure,and the model structure and performance are not necessarily optimal.In this paper,we propose a side channel analysis method based on dendritic networks,which outperforms deep learning networks in terms of system recognition capability and operational complexity due to the interpretability of the nonlinear structure inside the dendritic network.On the CW308T-STM32F3 and ATXMEGA128D4 target board of ChipWhisperer side-channel analysis experimental platform,side-channel analysis experiments are carried out for AES-128 encryption algorithm.The experimental results show that the dendritic network-based side channel analysis outperforms the deep learning networks in terms of model parameter scale,attack accuracy,and training time,such as multilayer perceptron,convolutional neural network,recurrent neural network,and other deep learning models.
作者
王俊年
王皖
于文新
胡钒梁
WANG Jun-nian;WANG Wan;YU Wen-xin;HU Fan-liang(School of Physics and Electronics,Hunan University of Science and Technology,Xiangtan 411201,China;Hunan Provincial Key Laboratory of Intelligent Sensors and Advanced Sensor Materials,Xiangtan 411201,China;Knowledge Processing and Networked Manufacturing Key Laboratory in Universities of Hunan Province,Xiangtan 411201,China)
出处
《湘潭大学学报（自然科学版）》
CAS
2021年第2期16-30,共15页
Journal of Xiangtan University(Natural Science Edition)
基金
国家自然科学基金(61973109)。
