Cryptology ePrint Archive: Report 2019/803

Methodology for Efficient CNN Architectures in Profiling Attacks

Gabriel Zaid and Lilian Bossuet and Amaury Habrard and Alexandre Venelli

Abstract: The side-channel community has recently investigated a new approach, based on deep learning, to significantly improve profiled attacks against embedded systems. Previous works have shown the benefit of using Convolutional Neural Networks (CNN) to limit the effect of some countermeasures such as desynchronization. In comparison to Template Attacks, deep learning techniques can deal with traces misalignment and the high dimensionality of the data. The pre-processing phases are no longer mandatory. However, the performance of attacks highly depend on the choice of each hyperparameters that compose a CNN architecture. Hence, we cannot perfectly harness the potential of deep neural networks without a clear comprehension of the networks inner-workings. In order to reduce this gap, we propose to clearly explain the role of each hyperparameters during the feature selection phase by using some specific visualization techniques such as Weight Visualization, Gradient Visualization and Heatmap. By highlighting which features are retained by filters, Heatmaps come in handy when a security evaluator tries to interpret and understand the efficiency of CNN. We propose a methodology for building efficient CNN architectures in terms of attack efficiency and network complexity, even in the presence of desynchronization. We evaluate our methodology on public datasets with and without desynchronization. In each case, we outperform the previous state-of-the-art CNN models while significantly reducing the network complexity. Our networks are up to 25 times more efficient than previous state-of-the-art while their complexity is up to 31810 times smaller. Our results show that CNN networks do not need to be too complex for getting good performance in the side-channel context.

Category / Keywords: secret-key cryptography / Side-Channel Attacks, Deep Learning, Architecture, Weight Visualization, Heatmap, Feature selection, Desynchronization

Date: received 11 Jul 2019, last revised 12 Jul 2019

Contact author: gabriel zaid at univ-st-etienne fr

Available format(s): PDF | BibTeX Citation

Version: 20190714:155137 (All versions of this report)

Short URL: ia.cr/2019/803


[ Cryptology ePrint archive ]