Profiling Attacks against ECC: Side Channel Analysis Based on Deep Learning for Curve-25519
Jiajun Xu, Meng Li, Lixin Liang, Yiwei Zhang, Shaohua Xiang, Zhe Ma · 2021 IEEE 21st International Conference on Communication Technology (ICCT) · 2021
This paper combines the deep learning technology and the side-channel analysis method to achieve effective cross-technology analysis for the ECC Curve-25519 algorithm on an STM32F407 (ARM-CortexM4) chip. The implementation of the target algorithm, which is protected with point randomization and constant-time operations (Montgomery powering ladder), uses security countermeasures. Therefore, the analysis explores the leakages from memory-related operations (conditional swap), which happen before each point addition or point doubling operation. The side-channel analysis method is used to determine and label the swap bits values of the TOE (target of evaluation), and then the electromagnetic radiation energy traces of the ECC Curve-25519 algorithm are fed into the convolutional neural network (CNN) for iterative training. The experiment shows that both the AlexNet and the VGG-13 network can effectively analyze the ECC Curve-25519 algorithm with security countermeasures, and the analysis accuracy rate reaches 96.11% and 98.61%, respectively.