Files
Download Full Text (560 KB)
Description
Deep learning-based side-channel attacks have shown to outperform non-deep learning-based approaches in certain categories within the side-channel analysis field. One of these categories is cryptographic cross-device side-channel attacks. In these attacks, one or more devices are used to construct a model from side channel data, and that model is used to recover a cryptographic key on a physically different device. Heterogeneous cross-device attacks are a type of cross-device attack where the side channel data used to construct a model is from a device that contains a different processor from a different manufacturer compared to the attack device. In this work, we train a convolutional neural network (CNN) and a multilayer perceptron (MLP) with power consumption data from a Riscure Pinata development board and attack a heterogeneous field programmable gate array (FPGA) device and a similar STM32F device that contains countermeasures. All devices were performing Advanced Encryption Standard 128 bit key (AES-128) with fixed key implementation. Unfortunately, we were unable to successfully recover the key from either attack devices using the MLP or CNN models. However, the models did reduce the key search space and the MLP model reduced the key search space more than the CNN model.
Publication Date
3-2025
Department
Computer Science
City
Mobile
Disciplines
Cybersecurity | Information Security | Other Computer Sciences
Recommended Citation
Kivilcim, Berk, "Heterogenous Gross Device Deep Learning Power Analysis Attack" (2025). Shelby Hall Graduate Research Forum Posters. 9.
https://jagworks.southalabama.edu/southalabama-shgrf-posters/9
