基于ResNet的KLEIN算法改进模板攻击

Improved Template Attack Based on ResNet Against KLEIN Algorithm

针对传统模板攻击存在的多元高斯正态分布假设受限、预处理复杂度高且不适用于带掩码防护的应用场景等问题,研究基于深度学习的模板攻击的改进方法.利用深度学习模型ResNet,对轻量级分组密码算法KLEIN实施改进模板攻击,根据数据的标签对数据进行分类.在密钥恢复阶段利用密钥优势叠加的方法,平均需要15条相同密钥加密所产生的能量迹即可有效区分正确密钥.相较于传统的模板攻击,本文的攻击方法成功恢复密钥所需攻击能量迹减少了83.7%,降低了模板攻击的难度,有效提高了模板攻击的成功率和效率.

In order to overcome the limitation of multivariate Gaussian normal distribution assumption,high preprocessing complexity and inappropriateness for masked application scenarios in traditional template attacks,an improved template attacks based on deep learning is proposed in this paper.The improved template attack is implemented based on a lightweight block cipher algorithm KLEIN,and uses the deep learning model ResNet.The data is classified according to its tags.Using the method of key dominance overlay in key recovery phase,an average of 15 power traces generated by the same key encryption are needed to distinguish the correct key effectively.Compared with traditional template attacks,the improved template attack based on ResNet presented in this paper reduces the power traces required to successfully recover the key by 83.7%,which reduces the difficulty of template attacks.It shows that the success rate and the efficiency of template attack have been significantly improved.