With the deployment of infrastructure projects such as the Internet of Things and artificial intelligence, security chips known as “data safes” are increasingly being challenged. The way of side-channel attack is particularly valued. Attack and defense synergistic artificial intelligence technology for side-channel analysis is rapidly evolving, and neural networks can perform fast and efficient analytical calculations. In the work of Prouff2016 [5] can directly omit the process of leak point modeling. In this study, the side channel attack based on the LSTM model is implemented and compared with the CNN model. The machine learning base on ASCAD and the traces with Gaussian noise were simulated on the database to analyze the attack results of the two models. Experimental data show that CNN learning extraction for inconspicuous features is better, and LSTM learning speed and accuracy for obvious features is faster and more accurate. We find that LSTM model can save 40% of the training time, making the loss and accuracy converge rapidly.