Parameters selection and optimization in profiled side channel analysis based on convolutional neural network

Profiled side channel analysis is considered as the most powerful side channel analysis by obtaining a copy of the attacking targe. With the population of deep learning technology in side channel analysis, Benadjila. R et al. proposed a profiled side attacks strategy based on convolutional neural network for AES with first-order mask countermeasure. However, this paper only implements training analysis while keeping the number of convolutional layers in the convolution module consistent and without trying to use common convolution kernel sizes. This paper conducts parameter selection experiments on the number of convolutional layers under the convolution module in the convolutional neural network and the size of the commonly used convolution kernel, and proposes a new optimized network model.As a result, this model can find out the correct key within 54 traces, which is better than the 180 traces record of Benadjila. R et al.