The Implementation of Neural Network Encryption Algorithms for Side-Channel Attack Protection

With the rapid development of deep learning in the field of side-channel analysis, neural network-trained encryption algorithms have demonstrated numerous advantages, providing novel ideas for resisting side-channel attacks. We implemented a bit neural network-based block encryption scheme resistant to side-channel attacks. Experimental verification shows that the complete algorithm combined with this scheme exhibits correctness, reliability, efficiency, and resistance to side-channel attacks. This scheme has two significant advantages: First, bit networks can achieve functions that multilayer perceptrons (MLPs) cannot perform. For example, using the AES encryption algorithm, we successfully reduced the column mixing network loss during MLP training from 0.25 to 0. Second, bit networks can integrate with MLPs without intermediate value leakage issues. Once combined with MLPs, the generalization capability of the AES round operation model is significantly enhanced, while ensuring that the number