VGG-Based Side Channel Attack on RSA Implementation

Profiling attack based on smaller and deeper neural network VGGNet is performed on a smart-card CRT-RSA implementation. CRT-RSA implementation uses security countermeasures including masking and time jittering. An ad-hoc method is applied to extract points from traces to perform an effective deep learning profiling attack. State-of-the-art convolutional networks are trained and compared on our data set. Experiment results show that the VGG13 network achieves the best performance with less training time, which can be a start point balancing performance and computational resources.