Paper 2023/021

DLPFA: Deep Learning based Persistent Fault Analysis against Block Ciphers

Yukun Cheng, Wuhan University
Changhai Ou, Wuhan University
Fan Zhang, Zhejiang University
Shihui Zheng, Beijing University of Posts and Telecommunications
Abstract

Deep learning techniques have been widely applied to side-channel analysis (SCA) in recent years and shown better performance compared with traditional methods. However, there has been little research dealing with deep learning techniques in fault analysis to date. This article undertakes the first study to introduce deep learning techniques into fault analysis to perform key recovery. We investigate the application of multi-layer perceptron (MLP) and convolutional neural network (CNN) in persistent fault analysis (PFA) and propose deep learning-based persistent fault analysis (DLPFA). DLPFA is first applied to advanced encryption standard (AES) to verify its availability. Then, to push the study further, we extend DLPFA to PRESENT, which is a lightweight substitution–permutation network (SPN)-based block cipher. The experimental results show that DLPFA can handle random faults and provide outstanding performance with a suitable selection of hyper-parameters.

Metadata
Available format(s)
PDF
Category
Attacks and cryptanalysis
Publication info
Preprint.
Keywords
Deep learningFault analysisPFAAESPRESENT
Contact author(s)
kuin @ mail ustc edu cn
ouchanghai @ whu edu cn
fanzhang @ zju edu cn
shihuizh @ bupt edu cn
History
2023-01-15: revised
2023-01-06: received
See all versions
Short URL
https://ia.cr/2023/021
License
Creative Commons Attribution
CC BY

BibTeX

@misc{cryptoeprint:2023/021,
      author = {Yukun Cheng and Changhai Ou and Fan Zhang and Shihui Zheng},
      title = {DLPFA: Deep Learning based Persistent Fault Analysis against Block Ciphers},
      howpublished = {Cryptology ePrint Archive, Paper 2023/021},
      year = {2023},
      note = {\url{https://eprint.iacr.org/2023/021}},
      url = {https://eprint.iacr.org/2023/021}
}
Note: In order to protect the privacy of readers, eprint.iacr.org does not use cookies or embedded third party content.