In recent years, the deep learning method has emerged as a mainstream approach to non-profiled side-channel attacks. However, most existing methods of deep learning-based non-profiled side-channel attack rely on traditional metrics such as loss and accuracy, which often suffer from unclear results in practical scenarios. Furthermore, most previous studies have not fully considered the properties of power traces as long time-series data. In this paper, a novel non-profiled side-channel attack architecture is proposed, which incorporates the attention mechanism and derives a corresponding attention metric. By attaching the attention mechanism after the network layers, the attention mechanism provides a quantitative prediction of correct key. Moreover, this architecture can effectively extract and analyze the features from long power traces. The success rate on different datasets is at least 86%, which demonstrates the superior reliability of this architecture compared to other works when facing various countermeasures and noise. Notably, even in scenarios where traditional loss and accuracy metrics fail to provide reliable results, the proposed attention metric remains capable of accurately distinguishing the correct key.

With the rapid advancement of deep learning, the neural network has become the primary approach for non-profiled side-channel attacks. Nevertheless, challenges arise in practical applications due to noise in collected power traces and the substantial amount of data required for training deep learning neural networks. Additionally, acquiring measuring equipment with exceptionally high sampling rates is difficult for average researchers, further obstructing the analysis process. To address these challenges, in this paper, we propose a novel architecture for non-profiled differential deep learning analysis, employing a hybrid-supervised neural network. The architecture incorporates a self-supervised autoencoder to enhance the features of power traces before they are utilized as training data for the supervised neural network. Experimental results demonstrate that the proposed architecture not only outperforms traditional differential deep learning networks by providing a more obvious distinction, but it also achieves key discrimination with reduced computational costs. Furthermore, the architecture is evaluated using small-scale and downsampled datasets, confirming its ability recover correct keys under such conditions. Moreover, the altered architecture designed for data resynchronization was proved to have the ability to distinguish the correct key from small-scale and desynchronized datasets.