Wavelet Support Vector Machine Algorithm in Power Analysis Attacks

Template attacks and machine learning are two powerful methods in the field of side channel attacks. In this paper, we aimed to contribute to the novel application of support vector machine (SVM) algorithm in power analysis attacks. Especially, wavelet SVM can approximate arbitrary nonlinear functions due to the multidimensional analysis of wavelet functions and the generalization of SVM. Three independent datasets were selected to compare the performance of template attacks and SVM based on various kernels. The results indicated that wavelet SVM successfully recovered the offset value of the masked AES implementation for each trace, which was obviously 5 to 8 percentage points higher than SVM-RBF. And also, the time required was almost reduced by 40% when using the optimal parameters of wavelet SVM. Moreover, wavelet SVM only required an average of 5.4 traces to break the secret key for the unmasked AES implementation and less than 7 traces for the masked AES implementation.