Non‐profiled Deep‐Learning‐Based Power Analysis of the SM4 and DES Algorithms

Power analysis methods are commonly used for evaluating the security of cryptographic devices. They characteristically low-cost and display a high success rate ability to obtain important device information, e.g., keys. Given current wide application deep-learning technology, there is growing tendency incorporate power-analysis technology in development. This study investigates non-profiled deep-learning-based power analysis. The labels this attack uncertain, conditions required greatly reduced. We choose Recurrent neural network (RNN), multilayer perceptron, convolutional algorithms, which use same structure, recover keys SM4 software DES hardware implementations. propose combining RNN algorithm with analysis, validate benefits experimentally. experimental results show that they all successfully correct key implementation, although by itself achieves better effect. conclusion also applies attacks on implementation but limited based bit model.