Performance Analysis of Non-Profiled Side Channel Attacks Based on Convolutional Neural Networks
Ngoc‐Tuan Do, Van‐Phuc Hoang, Van‐Sang Doan · 2020
There are emerging issues about side channel at-tacks (SCAs) on the cryptographic devices which are widely used today for securing secret information. Recently, the neural networks have been introduced as a new promising approach to perform SCA for hardware security evaluation of cryptographic algorithms. In this work, we present a non-profiled SCA using convolutional neural networks (CNNs) on an 8-bit AVR micro-controller device running the AES-128 cryptographic algorithm. We aim to point out the practical issues that occurs in CNN based SCA methods using the aligned power traces with a large number of samples. Furthermore, a method to build a suitable dataset for CNN training is introduced. Especially, practical experiment results of the CNN based SCA methods and a comprehensive investigation on the effect of noise are also presented. These experiments are performed with the original power traces and additive Gaussian noise. The results show that the CNN based SCA with our constructed dataset provides reliable results for non-profiled attacks. However, it is also shown that the Gaussian noise added on power traces becomes a serious problem.