Paper 2018/1213

Deep Learning vs Template Attacks in front of fundamental targets: experimental study

Yevhenii ZOTKIN, Francis OLIVIER, and Eric BOURBAO

Abstract

This study compares the experimental results of Template Attacks (TA) and Deep Learning (DL) techniques called Multi Layer Perceptron (MLP) and Convolutional Neural Network (CNN), concurrently in front of classical use cases often encountered in the side-channel analysis of cryptographic devices (restricted to SK). The starting point regards their comparative effectiveness against masked encryption which appears as intrinsically vulnerable. Surprisingly TA improved with Principal Components Analysis (PCA) and normalization, honorably makes the grade versus the latest DL methods which demand more calculation power. Another result is that both approaches face high difficulties against static targets such as secret data transfers or key schedule. The explanation of these observations resides in cross-matching. Beyond masking, the effects of other protections like jittering, shuffling and coding size are also tested. At the end of the day the benefit of DL techniques, stands in the better resistance of CNN to misalignment.

Metadata
Available format(s)
PDF
Publication info
Preprint. MINOR revision.
Keywords
Side channeldeep dearningtemplate attack
Contact author(s)
francis olivier @ gemalto com
History
2018-12-23: received
Short URL
https://ia.cr/2018/1213
License
Creative Commons Attribution
CC BY

BibTeX

@misc{cryptoeprint:2018/1213,
      author = {Yevhenii ZOTKIN and Francis OLIVIER and Eric BOURBAO},
      title = {Deep Learning vs Template Attacks in front of fundamental targets: experimental study},
      howpublished = {Cryptology ePrint Archive, Paper 2018/1213},
      year = {2018},
      note = {\url{https://eprint.iacr.org/2018/1213}},
      url = {https://eprint.iacr.org/2018/1213}
}
Note: In order to protect the privacy of readers, eprint.iacr.org does not use cookies or embedded third party content.