Linear Regression Analysis for Leakage Model of Differential Side Channel Cryptanalysis

Abstract:

An advanced statistical method, linear regression model, is proposed to construct the power leakage model for the differential side channel analysis (DSCA) attacks on cryptographic devices. Even with only a limited knowledge on how the device leaks information, the linear regression leakage model can be constructed, which overcomes the limitations of the traditional leakage models. First, the stochastic approach for analysis of power leakage is investigated and the linear regression model is built. Then the coefficients of the linear regression model are estimated with two methods: least square estimator(LSE) and least absolute estimator(LAE). Finally the mathematical model and methods are realized by an experimental analysis of an advanced encryption standard(AES) implementation on an 8 bit microcontroller based PayTV smartcard platform. A comparative analysis of both estimators shows that LSE is more suitable than LAE concerning the linear regression analysis of leakage model. In addition, investigation on the curves of the estimated model coefficients shows that linear regression analysis can be applied to preprocessing the measurement traces and the preprocessing helps to increase the efficiency of leakage modeling.