MO-DLSCA: Deep Learning Based Non-profiled Side Channel Analysis Using Multi-output Neural Networks

Abstract

Differential Deep Learning Analysis (DDLA) is the first side-channel analysis (SCA) attack using deep learning (DL) in non-profiled scenarios. However, DDLA requires many training processes to distinguish the correct key. In this paper, we propose a novel SCA technique using multi-output multi-loss neural networks, which can predict all possible hypothesis keys simultaneously in a short time of the training process. Specifically, a multi-output classification (MOC) model and a multi-output regression (MOR) model are introduced. Especially, we first suggest using identity labeling for MOR model to determine the trend of the training metric for each hypothesis key in the non-profiled SCA scenario. As a result, the correct key can be distinguished easily. The efficiency of proposed model is clarified on different SCA-protected schemes, such as masking and combined hiding-masking countermeasure methods. Significantly, our approach remarkably outperforms the DDLA model and parallel network in terms of the execution time and the success rate. In addition, by using shared layers, the proposed model achieves a higher success rate of at least 25 % in the case of combined hiding-masking countermeasure.