LCAF: A Novel Side-Channel Analysis Method Based on Self-Attention Mechanism

Cryptography technology is widely used in various fields to protect information security, and the security evaluation of cryptographic modules or chips themselves is also receiving increasing attention. Side-channel analysis is a method to extract sensitive information, such as secret keys, by analyzing side-information leaked during the operation of encryption devices. It largely avoids the complexity of traditional cryptanalysis, and greatly improves the efficiency of analysis attacks. In recent years, researchers have applied deep learning to side-channel analysis attacks, greatly improving the performance of analysis attacks. However, most deep learning methods still suffer from fast overfitting, slow convergence, and shallow feature extraction, and are only effective for specific scenarios. We propose a novel method LCAF based on Long-short term Memory (LSTM), Convolutional neural network (CNN), and self-attention mechanism. Experimental results on four different datasets show that our method outperforms the existing techniques, and can successfully recover the key with only few traces.