Efficient Convolutional Neural Network Based Side Channel Attacks Based on AES Cryptography

In recent years, profiled side-channel attacks have emerged as a particularly potent kind of side-channel attack that can circumvent cryptographic equipment’s security. Convolutional neural networks (CNNs) have been widely used as deep learning infrastructure for attacks in recent research that has examined a new kind of profiled attack based on deep learning. The design of CNNs will have a significant impact on attack effectiveness. However, image recognition fields are frequently the foundation of the CNN architecture currently used for profiled attacks. Additionally, it is still challenging to choose the proper parameters and CNN infrastructure concerning adaptation to profiled assault types. In the current study, an effective CNN-based profiled attack was suggested that can be used against masking-protected cryptographic devices. The Grey Wolf Optimization (GWO) approach obtains the CNN architecture parameters proposed in this research. The characteristics of intriguing areas on the power trace determine these parameters. The proposed attacks were tested using experiments over a tracing set gathered from an intelligent card Atmega8515 processing the ASCAD public dataset.