Research On Intelligent Evaluation Method Of Side-channel Leakage Of Cryptographic Algorithm Based On Machine Learning

The security of a cryptographic system depends not only on the mathematical security of the cryptographic algorithm,but also on the physical security of the cryptographic implementation.The side-channel attacks are aimed at the vulnerabilities of the cryptographic implementations,utilizing the power consumption,electromagnetic radiation,running time,and other side-channel information generated by cryptographic devices to recover the key.Sidechannel attacks have the characteristics of low cost and significant effect,which pose a fatal threat to the cryptographic products.The powerful ability of side-channel attacks has attracted great attention from cryptographic manufacturers and evaluation agencies.Therefore,researchers use various methods to evaluate the resistance to side-channel attacks of cryptographic products,and the most widely used evaluation methods include CC standards and evaluation methods based on hypothesis testing.However,the CC standard needs to master the implementation details of the cryptographic algorithm and carry out multiple side-channel attacks,which is expensive and not universal.The evaluation methods based on hypothesis testing are easy-to-apply and universal,but they have low evaluation accuracy and evaluation efficiency.Moreover,it is difficult for the evaluation methods based on hypothesis testing to evaluate protection schemes ant they are not able to provide key leakage locations,which is not conducive to manufacturers' improvement of cryptographic products.This dissertation is dedicated to solving the problems such as the inability of existing evaluation methods to obtain key leakage locations,low accuracy and low efficiency,and researching an easy-to-apply,accurate and efficient black-box intelligent evaluation method for side-channel leakage.The main work and contributions are as follows:(1)Firstly,a new evaluation hypothesis is proposed based on machine learning: if there is side-channel leakage in cryptographic device,the power traces generated when encrypting different plaintexts should be sortable,and those trace points that contribute the most to the classification are the key leakage points.According to the above hypothesis,a framework for side-channel leakage assessment based on machine learning is proposed,which regards trace points as original feature sets for supervised learning and classification.The leakage probability is calculated based on classification accuracy to determine whether the cryptographic device leaks.And the optimal subset is selected by feature selection as the key leakage set to accurately identify key leakage locations.(2)Then,the MSR evaluation algorithm and the RGS evaluation algorithm are proposed based on the MLLA framework.The evaluation index of machine learning is introduced for the first time as the performance evaluation criteria of side-channel leakage assessment,which can evaluate the evaluation results more scientifically and accurately.According to the evaluation experiments of AES on the side-channel standard evaluation board SAKURA-G,the performance of the two methods under the MLLA framework is much better than the evaluation methods based on hypothesis testing.In particular,compared with the evaluation method based on the hypothesis test,the MSR evaluation algorithm improves the evaluation accuracy by 76%and the comprehensive evaluation index by 66%,while the false positive rate has been reduced by 69%.The method only needs 4000 power traces to achieve convergence,which is far superior to other methods.(3)Finally,the collection scheme and algorithm parameters of the evaluation framework are discussed in detail and optimized,and the optimal subset selection algorithm is studied to propose the adaptive optimal subset intelligent selection algorithm.Experimental results show that better evaluation performance can be achieved and the optimal number of subsets can be selected automatically according to the encryption algorithm.Finally,this dissertation successfully evaluates the AES-256 RSM masked protection scheme of the DPA Contest international side-channel competition and the candidate algorithm SIKE in the third round of the NIST post-quantum cryptography competition and analyzes their leakage situation.