Deep Learning-Based Power Analysis Attack for Extracting AES Keys on ATmega328P Microcontroller

In recent years, several deep learning (DL) approaches, such as convolutional neural networks (CNN) and recurrent neural networks (RNN), have been utilized to carry out side-channel attacks. These attacks exploit vulnerabilities in cryptographic systems, especially in resource-constrained devices like microcontrollers. By analyzing side-channel measures using DL models, attackers can extract information leaked through physical means, including power consumption, time, heat, sound, and electromagnetic radiation, to uncover the secret keys of cryptographic algorithms. Most of these attacks rely on DL models to extract secrets from cryptographic algorithm implementations. In this article, we present a DL-based power analysis attack (PAA) to extract the advanced encryption standard (AES) keys from the ATmega328P microcontroller. Our approach employs a CNN trained on power consumption traces collected during AES-128 execution on the microcontroller. We explain how DL-based PAA exploit the leakage of information from cryptographic devices to recover the secret keys, and we provide a detailed performance comparison of our method to other methods that target cryptographic devices. Our approach stands out not only in its methodology, but also in its robustness and efficiency in revealing secret keys. In comparison with other state-of-the-art methods, our DL-based PAA, implemented on an ASIC with a 130 nm technology, requires only about 1200 traces to successfully extract all 16 bytes of the AES-128 key. This performance outperforms the next-best method evaluated in this paper by a margin of 100 power consumption traces.