Static power analysis of cryptographic devices

Xu, Jiming (2019) Static power analysis of cryptographic devices. Doctoral (PhD) thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (4MB)

Abstract

Side-channel attacks are proven to be efficient tools in attacking cryptographic devices. Dynamic power leakage has been used as a source for many well-known side-channel attack algorithms. As process technology size shrinks, the relative amount of static power consumption increases accordingly, and reaches a significant level in sub-100- nm chips, potentially changing the nature of side-channel analysis based on power consumption. In this thesis, we demonstrate our work in side-channel attacks exploiting static power leakage. Our research interest is particularly focused on profiled attacks. Firstly, we present recent developments of static power analysis and provide our results to further support some of the conclusions in existing publications. We also give a description of the template attack we developed for static power analysis of block ciphers. This template attack uses new distinguishers which are previously applied to other data analysis fields. The results of our study are achieved using simulations in a 45-nm and 65-nm CMOS environment, and demonstrate the viability of static-power-based template attacks. Secondly, we bring kernel density estimation into the scenario of static power analysis. We compare the performance of the kernel method and conventional Gaussian distinguisher. It is demonstrated in our experiments that the static power leakage may not satisfy multivariate Gaussian distribution, in which case the kernel method results in better attack outcomes. Thirdly, we perform template attacks on a masked S-box circuit using static and dynamic power leakage. We are the first to compare static power and dynamic power in the scenario of profiled attacks against masked devices. The attacks are shown to be successful, and by performing multiple attacks and adding Gaussian noise, we conclude that in the 45-nm environment, dynamic power analysis requires a high sampling rate for the oscilloscopes, while the results of static-power-based attacks are more sensitive to additive noise. Lastly, we attempt to combine static and dynamic power leakage in order to take the advantage of both leakage sources. With the help of deep learning technology, we are able to propose more complex schemes to combine different leakage sources. Three combining schemes are proposed and evaluated using a masked S-box circuit simulated with 45-nm library. The experiment results show that the hierarchical LSTM proposal performs the best or close to the best in all test cases.

Actions (login required)

View Item