旁路密码攻击建模分析方法研究进展与分析

Research Progress and Analysis on the Profiling Analysis Methods for Side-channel Attacks

嵌入加密算法的密码芯片是信息安全的有效保障,而旁路建模分析方法可以有效攻击密码实现。详述了建模类旁路分析原理,包含攻击原理以及攻击方法破密性能的评估手段。介绍了不同时期表现突出的基于旁路泄漏信号的3种建模分析方法:经典建模(Template Attack,TA)、多层感知器建模(Multi-Layer Perceptron Side Channel Analysis,MLPSCA)和卷积神经网络建模(Convolutional Neural Network Side Channel Analysis,CNNSCA)。通过模型对比以及实验结果论证了在攻击理想低噪的旁路信号时,新型建模方法MLPSCA、CNNSCA破密性能明显优于传统模板攻击TA。进一步比较了MLPSCA与CNNSCA的破密性能,用这2种方法分别攻击带高斯噪声的功耗数据以及带防护的功耗数据,这2种功耗数据分别表示有环境噪声干扰或采取防护策略的目标信号。验证结果表明,MLP模型仅关注输入数量级不考虑数据拓扑结构,CNN具有更为鲁棒的结构特性,对失真功耗数据依旧具有良好的识别性能,CNNSCA攻击因噪声和防护对策而产生畸变的旁路信号更具优势。

The crypto chip embedded with the encryption algorithm is an effective guarantee for information security,and the side-channel profiling analysis method can effectively attack the cryptographic implementation.The principle of profiling side-channel analysis is described in detail,including the attack principle and the evaluation method of attack method's declassification performance.Three profiling analysis methods based on side-channel leakage signals with outstanding performance in different periods are introduced:classical Template Attack(TA),Multi-Layer Perceptron Side Channel Analysis(MLPSCA),and Convolutional Neural Network Side Channel Analysis(CNNSCA).It is demonstrated through model comparison and experimental results:when attacking ideal low-noise side-channel signals,the new profiling methods MLPSCA and CNNSCA have significantly better performance than traditional template attack TA.The decryption performance of MLPSCA and CNNSCA is further compared.These two methods are used to attack the power consumption data with Gaussian noise and the power consumption data with protection respectively.These two kinds of power consumption data represent the target signal with environmental noise interference and with protection strategy respectively.The verification results show that the MLP model only pays attention to the magnitude of the input and does not consider the data topology,while CNN has more robust structural characteristics and good recognition performance for distorted power consumption data.CNNSCA has more advantages in attacking side-channel signals that are distorted due to noise and protective countermeasures.