基于SincNet的侧信道攻击

SincNet-based Side Channel Attack

侧信道攻击利用密码算法在物联网设备上执行时产生的时间、功耗、电磁辐射和故障输出等泄露来恢复密钥或者其他敏感信息,它已经成为了加密安全设备的重要威胁之一.近年来,建模类侧信道攻击在加密算法安全性评估中发挥着重要的作用,它被认为是现阶段最强大的攻击方法.随后,深度学习技术应用于建模类侧信道攻击,并且在公开数据集上取得了良好的效果.在本文中,我们提出了一种优化的卷积神经网络侧信道攻击方法,该方法将一种新的网络结构SincNet应用于侧信道攻击,SincNet卷积层只需要学习滤波器的高和低两个截止频率,相比于传统的卷积层,学习的参数量更少.为了检验该攻击方法的有效性,我们使用公开的ASCAD数据集和DPA contest v4.1数据集对其进行评估.实验结果表明,我们在ASCAD.h5上仅需要170条能量轨迹就能恢复出正确的子密钥.另外,我们也在ASCAD_desync50.h5和ASCAD_desync100.h5这两个轨迹非对齐的数据集上进行评估,该方法有效地缓解了轨迹非对齐造成的影响,得到了优于Prouff等人在2018年的实验结果.对于DPA contest v4.1数据集,我们使用了CNN网络和SincNet网络对其进行训练和测试,均可以达到很好的攻击效果,仅需要一条能量轨迹就可以恢复出子密钥,为了证明SincNet网络的有效性,我们减少训练轨迹的条数,发现SincNet网络能够使用更少的训练轨迹条数恢复出子密钥,然后我们对经过SincNet层处理之后的能量轨迹作了相关性分析,发现相关性得到了一定的提升.

Side channel attacks exploit side-channel leakage,such as time,power consumption,electromagnetic radiation,and fault output to recover the key or other sensitive information of the cryptographic algorithm implemented in IoT devices.They have become one of the most important threats to cryptographic security devices.In recent years,profiling attacks play an important role in the security evaluation of encryption algorithms,and they are considered to be one of the most powerful attacks.Subsequently,the deep learning techniques have been applied to profiling attacks and achieved good results on public datasets.This paper proposes an optimized convolutional neural network method for side channel attack,which applies a new network structure called SincNet.The SincNet layer only needs to learn the high and low cutoff frequency of the filter.Compared to traditional convolutional layer,there are fewer parameters to learn.The proposed method is evaluated with the public ASCAD database and the DPA contest v4.1 dataset.The experimental results show that 170 electromagnetic power traces are sufficient to recover the correct subkey.In addition,the proposed method is extended to two non-aligned datasets,ASCAD_desync50.h5 and ASCAD_desync100.h5,and the effectiveness of the proposed method is shown to outperform those proposed by Prouff et al.in 2018.For the DPA contest v4.1 dataset,the CNN network and SincNet network are used to train and test,which achieves a good attack effect.Based on the proposed method,only one power trace is sufficient to recover the subkey.In order to prove the effectiveness of the SincNet network,the number of training traces are reduced and the SincNet network is able to recover the subkey with fewer training traces.The correlation of the power traces processed by the SincNet layer is analyzed,and the results show that the correlation is improved.