Multi-byte side-channel attacks on chaotic block cipher: Autoencoder-based feature selection and CNN-Transformer collaborative attack framework

In recent years, chaotic block ciphers have been the focus of significant research and deployment, owing to the robust pseudo-random characteristics and unpredictability of chaotic systems. However, chaotic block ciphers have been shown to leak physical information during hardware encryption, and there is a risk of side-channel attacks. Furthermore, there are relatively few studies related to the security analysis of chaotic block ciphers. Concomitant with the continuous development in the field of artificial intelligence, deep learning has been extensively utilized in side-channel attacks, data processing and other domains. In order to enhance the efficacy of side-channel attacks, this paper proposes a novel approach to energy trace preprocessing. This approach is based on wavelet denoising and autoencoder feature selection, with the objective of extracting leakage information from the attack location with efficiency and reducing the number of sampling points by a significant margin. The method of correlation power analysis is utilized in the attack. It can be observed that the preprocessed energy trace accurately predicts the key through approximately 20 energy traces. A comparison of the results before and after preprocessing indicates a substantial reduction in attack time, which is decreased by approximately a factor of 10, thereby facilitating the efficient training process of the side-channel attack for deep learning. Meanwhile, this paper also proposes a method for conducting a deep learning side-channel attack. This method combines convolutional neural networks and Transformer parts of the architecture. The architecture contains two parts: single-byte attack and multi-byte attack, and attacks on chaotic block cipher. As demonstrated by the experimental results, the duration of the model’s training for each epoch is approximately 3–4s, and the accuracy exhibits a substantial increase, indicating a relatively high level of accuracy. The correct round key is guessed through less than 100 energy traces, and in the multi-byte case, the entire round key of 16 bytes can be guessed through about 200 energy traces. The proposed methodology in this study has the potential to enhance the capacity for analyzing side-channel attacks on chaotic block ciphers, thereby contributing to the advancement of chaotic block ciphers and the enhancement of their security.