Paper 2019/1474

Remove Some Noise: On Pre-processing of Side-channel Measurements with Autoencoders

Lichao Wu and Stjepan Picek

Abstract

In the profiled side-channel analysis, deep learning-based techniques proved to be very successful even when attacking targets protected with countermeasures. Still, this does not mean that countermeasures do not make the attacks more difficult or that deep learning attacks will always succeed. As such, to improve the performance of attacks, an intuitive solution is to remove the effect of countermeasures. In this paper, we investigate whether we can consider certain types of countermeasures as noise and then use deep learning to remove that noise. We conduct a detailed analysis of four different types of noise and countermeasures either separately or combined and show that in all scenarios, denoising autoencoder improves the attack performance significantly.