EfficientNet-based electromagnetic attack on AES cipher chips
Wenxu Ning, Hongxin Zhang, Danzhi Wang, Fan Fan, Lei Shu · 2023
This paper presents a bypass attack on the Field Programmable gate array (FPGA) cryptographic chip Advanced encryption standard (AES) encryption algorithm. Since the accuracy of half-byte classification is low and the model is slow to converge and easy to overfit when using deep learning, this experiment proposes for the first time to introduce the deep learning network EfficientNet model into the field, which only targets the electromagnetic leakage information of the hardware during encryption for any half-byte, when the plaintext and ciphertext are unknown. This is followed by a "divide and conquer" approach to the entire key, where each half-byte is attacked to obtain the entire key. The accuracy of the model used in this study was found to be about 17% higher than that of common networks such as ResNet and DenseNet, and the model converged better.