Performance Analysis of Non-profiled Side Channel Attack Based on Multi-layer Perceptron Using Significant Hamming Weight Labeling

Deep learning (DL) techniques have become popular for side-channel analysis (SCA) in the recent years. This paper proposes and evaluates the applications of multilayer perceptron (MLP) models for non-profiled attacks on the AES-128 encryption implementation in different scenarios, such as high dimensional data, imbalanced classes, and the impact of additive noise. Along with the designed models, a labeling technique called significant Hamming weight (SHW) and dataset reconstruction method are introduced for solving the imbalanced dataset problem. In addition, using SHW in the non-profiled context can reduce the number of measurements needed by approximately 30. The experimental results show that the DL based SCA with our reconstructed dataset for different targets of ASCAD, RISC-V microcontroller has achieved a higher performance of non-profiled attacks. Comparing to the binary labeling technique, SHW labeling provides better results with the presence of the additive noise. © 2022, ICST Institute for Computer Sciences, Social Informatics and Telecommunications Engineering.