Deep-Learning Side-Channel Attacks (DLSCAs) have become a realistic threat to cryptographic algorithms, such as Advanced Encryption Standard (AES). Since the encryption has to run in hardware at some point to actually do things, there might be some unintentional physical leakage, such as the different amount of power consumed by the victim device. By using deep-learning models to analyze the power traces, the attacker is able to derive the secret key. In this project, we implement a real deep-learning based attack to against a STM32 implementation of AES. We apply four different types of neural networks, MLP, CNN, LSTM and RNN, to classify traces. Afterwards, we evaluate to which extent different types of models could make the attack more efficient.