密码学报（中英文）2025，Vol.12Issue(2)：414-428,15.DOI:10.13868/j.cnki.jcr.000771
基于CLRM模型的侧信道攻击
Side Channel Attacks Based on CLRM Model
摘要
Abstract
Although deep learning-based side-channel modeling attacks have achieved certain results,their deep learning models still face some difficulties and challenges,such as the training performance problem of low accuracy in the side-channel modeling phase and the attack efficiency problem of requiring a large amount of energy traces to achieve correct key acquisition in the side-channel attack phase.Therefore,this study proposes a new network structure model CLRM(named after the initials of CNN model,LSTM model,ResNet model),which contains a convolutional neural network module,a long and short-term memory network module,and a residual network module,and introduces a soft threshold in the residual network,so that the CLRM model is able to automatically extract the leakage information in the energy traces,and reduce the number of leakage information in the model.By introducing soft thresholding into the residual network,the CLRM model can automatically extract the leakage information from the energy traces,reduce the number of parameters required for model training,and adaptively soft-threshold the extracted feature information,which improves the performance of the CLRM model.In order to evaluate the performance of the CLRM model,tests were conducted on two publicly available datasets,ASCAD and DPA contest v4.The experimental results show that on the ASCAD dataset,the accuracy of the CLRM model can reach 92.81％,and only 30 energy traces are needed to obtain the correct subkey.Compare to the CNN_best and MLP_best models proposed by Benadjila et al.,the Zaid model proposed by Zaid et al.,as well as the CBAPD model proposed by Zheng Dong et al.,the accuracy of CLRM model is improved by 23.98％ and the efficiency of attack is improved by 40％ .On the DPA contest v4 dataset,the accuracy of CLRM model can reach 90.33％,and only 2 energy traces are needed to obtain the correct subkey,and the accuracy of its model is improved by 0.41％,and the efficiency of attack is increased by 33.33％ .In other words,the performance of the CLRM model has been improved,and the feasibility and effectiveness of the deep learning model performance optimization method proposed in this study are also verified.关键词
侧信道攻击/卷积神经网络/长短期记忆网络/残差网络/软阈值Key words
side channel attack/convolutional neural network/long short-term memory network/residual network/soft thresholding分类
信息技术与安全科学引用本文复制引用
黄海,唐新琳,吴金明,刘志伟,于斌,赵石磊..基于CLRM模型的侧信道攻击[J].密码学报（中英文）,2025,12(2):414-428,15.基金项目
国家重点研发计划(2018YFB2202101) （2018YFB2202101）
中央引导地方科技发展专项(ZY20B11) （ZY20B11）
哈尔滨市制造业科技创新人才项目(CXRC20221104236)National Key Research and Development Program of China(2018YFB2202101) （CXRC20221104236）
Special Projects for the Central Government to Guide the Development of Local Science and Technology(ZY20B11) （ZY20B11）
Harbin Manufacturing Science and Technology Innovation Talent Project(CXRC20221104236) （CXRC20221104236）
