Side-channel attacks on convolutional neural networks based on the hybrid attention mechanism

In the field of security assessment of password chips, side-channel attacks are an important and effective means of extracting sensitive information by analysing the physical characteristics of the chip during operation, providing an important basis for security assessment. In recent years, deep learning technology has been widely used in the field of side-channel attacks, which can automatically learn and identify the physical leakage characteristics of the chip and improve the efficiency and accuracy of the attack. However, deep learning-based side-channel attacks may be disturbed by environmental noise during the training process, and there are also problems of model overfitting and slow convergence. In order to more effectively extract feature information in the trajectory to implement a side-channel attack, this paper proposes a new attention mechanism convolutional neural network model architecture. The model combines a convolutional neural network with an attention mechanism. It optimises the traditional CNN model by improving the convolutional layer and introducing a fused hybrid attention mechanism, enhancing the model's ability to capture global information to effectively extract relevant leaked information. Experimental results show that the model has good attack results on the ASCAD public dataset. Compared with other models, it requires 74.87% less power consumption for side-channel analysis, and the model accuracy is significantly improved. It solves the problems of overfitting and slow convergence speed, and can meet the requirements of side-channel modeling and analysis.