Research On Side-Channel Black-Box Attack Based On Deep Learning User Dynamic Interest Model

Side channel attack is a method to attack cryptographic devices by using physical information such as power consumption generated during the running of cryptographic algorithms,which poses a serious threat to cryptographic products.Therefore,side channel attack and evaluation in the research and development stage of cryptographic products are helpful for cryptographic manufacturers to understand the potential security vulnerabilities of cryptographic products and protect them.With the development of deep learning technology,side channel attacks based on deep learning become more efficient and automated.However,the existing side channel attack based on deep learning is based on the leakage operation of the cryptographic algorithm.However,the leakage operation of the cryptographic algorithm requires experienced attackers to review the implementation code of the cryptographic algorithm.This paper aims to solve the problems of the existing deep learning-based side channel attack,such as cumbersome steps and heavy reliance on leakage operations,and research a more general and efficient side channel black-box attack method.The main work and contributions of this paper are summarized as follows:(1)A global leakage information extraction method based on machine learning is proposed.There may be multiple power leakage points which depend on plaintext and key in the power information of the cryptographic algorithm.These leakage points can help us to carry out the attack without relying on a specific intermediate value and reduce the difficulty of the attack.Therefore,we propose a method of extracting global leaked information based on machine learning,which can make the model better learn tag-related information,reduce the amount of data and achieve black-box attack.(2)A new label calculation method is proposed.In order to abandon the manual search for leakage operation,the data set is no longer labeled by the corresponding intermediate value and leakage model.In this paper,we propose a label method which only uses plaintext and key for calculation,and does not need to calculate the intermediate value corresponding to the leakage operation according to the cryptographic algorithm,so as to improve the universality of side channel attack.(3)A GLA attack architecture suitable for side channel black-box attack is proposed.For the first time,the Conformer model with convolution layer and attention mechanism is applied to the field of side channel attack.Based on the new labeling method,the Conformer model is used to extract the local and global information in the global leakage information,so as to fully learn the features of power consumption data dependent on plaintext and key.At the same time,in order to improve the efficiency of side channel black-box attack,the GLA attack architecture is optimized to realize side channel attack only by relying on single digit power consumption under the premise of unknown cryptographic algorithm leakage operation.