The research of the side-channel analysis method based on deep learning for trusted platform module

Abstract

At present, the trusted platform module (TPM) has been widely used in electricity, finance, transportation, and other industries, and its security has attracted much attention. The research on the side-channel attack (SCA) can be conducive to improving the security design technology of trusted platform modules. The side channel attack on the trusted platform module is studied from the initial differential power analysis (DPA) to the machine learning method. This paper introduces the deep learning techniques in SCA and proposes a new SCA technique based on deep learning. Firstly, on basis of the characteristics of SCA on TPM, we improve the convolutional neural network (CNN) models, including ResNet, VGG, and Google Net. Then, for the SM4 cipher implemented with and without countermeasures, we implement some attack experiments by exploiting the SCA based on various deep learning models and compare the experimental results with the DPA attack. The experiment results show that under the same circumstance, the SCA based on the deep learning method is more effective. We compare the effects of different CNN models and hyper-parameters to provide the basis and data for further research.