Side Channel Analysis Based On Attention Mechanism And Multi-label Learning

Side-channel analysis is a powerful technique used to attack cryptographic algorithms by analyzing the information leaked during the operation of a device in order to recover the encryption key.Over time,advancements in cryptographic devices and the implementation of various protective measures have significantly reduced the efficiency and success rate of traditional side-channel analysis.To address these challenges and enhance analysis efficiency,deep learning techniques have been applied to side-channel analysis.However,deep learning-based side-channel analysis encounters its own set of problems and challenges,such as model optimization issues and the need for repetitive modeling and training to recover all byte keys.This article focuses on deep learning approaches for side-channel analysis,utilizing attention mechanisms and multi-label learning methods to construct and train deep learning models.These models are experimentally validated using the publicly available dataset of the AES algorithm.The key highlights of this article are as follows:1.To address the optimization issues in model construction and feature extraction in deep learning side-channel attacks,a model construction method combining attention mechanism and dense connection convolutional network is proposed,followed by conducting side-channel attacks.The attention mechanism effectively improves the model’s feature extraction ability,and the dense connection convolutional network makes the enhanced features more prominent,and the gradient propagation is more effective,improving the training effect of the model.The test results show that the attention mechanism can effectively improve the model’s feature extraction ability compared to existing deep learning side-channel attack models.2.Currently,side-channel analysis in deep learning primarily employs a divide-and-conquer strategy,which involves repeatedly building and training models to recover all byte keys,resulting in low efficiency.To address this issue,a multi-label convolutional neural network(CNN)model is proposed for multi-label learning to recover multi-byte keys.Test results indicate that the multi-label model can effectively establish models and conduct attacks on multiple byte keys.It significantly reduces the overhead of repetitive model building and training while efficiently accomplishing the recovery of multiple key bytes.3.To efficiently conduct side-channel attacks on the same cryptographic algorithm running on heterogeneous devices,the study explores a multi-label convolutional neural network model.Through multi-label learning,the energy traces corresponding to single-byte key labels under heterogeneous devices are constructed to form a multi-label dataset of heterogeneous devices.A convolutional neural network model is established and trained,and finally,the key recovery of the same cryptographic algorithm running on heterogeneous devices is achieved.Test results demonstrate that the multi-label model can effectively accomplish key recovery in such scenarios,enhancing the efficiency and success rate of side-channel analysis attacks.