Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (6): 268-276.DOI: 10.3778/j.issn.1002-8331.2111-0511

• Network, Communication and Security • Previous Articles     Next Articles

Side Channel Attack Fused with CNN_LSTM

PENG Pei, ZHANG Meiling, ZHENG Dong   

  1. 1.School of Communications & Information Engineering, Xi’an University of Posts and Telecommunications, Xi’an 710121, China
    2.Shaanxi Province Wireless Network Security Technology National Engineering Laboratory, Xi’an University of Posts and Telecommunications, Xi’an 710121, China
  • Online:2023-03-15 Published:2023-03-15



  1. 1.西安邮电大学 网络空间安全学院,西安 710121
    2.西安邮电大学 陕西省无线网络安全技术国家工程实验室,西安 710121

Abstract: Based on the application of deep learning in side-channel attacks, the AES algorithm is first implemented in the Chipwhisperer platform, the corresponding energy trace is measured during its encryption process, and then the CPA technology is used to analyze the location of the point of interest, and model training is made for the point of interest. On the three network models convolutional neural network(CNN), long-short-term memory network(LSTM) and CNN_LSTM hybrid model, combined with data preprocessing technology to train synchronous and asynchronous energy traces. The experimental results show that the accuracy of the three models in the synchronous state is equivalent. In addition, when the asynchronous data is gradually increased while ensuring the model training parameters remain unchanged, the accuracy of the training set and test set of the three models is decreasing, but the declining speed of the new hybrid model is the slowest. When the experimental asynchronous number is increased to 50, the accuracy can still be guaranteed to be above 90%, that is, the correct key can be recovered with almost one energy trace. Therefore, the CNN_LSTM model can better adapt to the situation where the energy trace occurs asynchronously.

Key words: advanced encryption standard(AES), side channel attack, convolutional neural network, long short-term memory network, deep learning

摘要: 基于深度学习在侧信道攻击中的应用,在Chipwhisperer平台中实现AES算法,在其加密过程中测量相应能量迹,再利用CPA技术分析得出兴趣点位置,并针对兴趣点做出模型训练。在卷积神经网络(CNN),长短时记忆网络(LSTM)和CNN_LSTM混合模型三种网络模型上,结合数据预处理技术训练同步和异步能量迹。实验结果表明三种模型同步状态下的准确率相当,另外在保证模型训练参数不变的情况下逐渐增大异步数据时,三个模型训练集和测试集的准确率都在减少,但新提出的混合模型下降速度变化是最慢的,在实验异步数加大到50时,仍可以保证准确率在90%之上,即几乎一条能量迹就可恢复出正确密钥。所以,CNN_LSTM模型可以更好地适应能量迹发生异步的情况。

关键词: 高级加密标准(AES), 侧信道攻击, 卷积神经网络, 长短时网络, 深度学习