The Power Analysis Based on Mean–Ridge Regression

Abstract

Comparing the common methods in side channel analysis such as correlation power analysis (CPA), the non-profiled power analysis based on machine learning achieves better attack effect, which increases the analysis accuracy rates by constructing and training a power leakage model. In order to construct an assumption of power leakage model which fits better with the measured power traces, a non-profiled power analysis method - based on mean-ridge regression - was proposed. It constructed the initial dynamic power leakage model by using the measured power consumption for the purpose of calculating the optimal ridge parameters and optimal weight coefficients of ridge regression. Then, the initial dynamic power leakage model was averaged to construct a power leakage model according to mean-ridge regression, which reduced the model overhead and the noise impaction. Finally, the key recovery process for the PRSENT algorithm was achieved by calculating the relationship between the model and the measured power consumption. The experimental results show that: compared with correlation power analysis and the existing method based on ridge regression, the power analysis based on mean-ridge regression expands the correlation coefficients between the model and the measured power traces, increases the attack accuracy, and reduces the number of power traces used for key recovery.