Lowering the Bar: Deep Learning for Side Channel Analysis

Conference:  BlackHat USA 2018



The presentation discusses the use of deep learning in site analysis for extracting cryptographic keys from power consumption traces.
  • The first stage of site analysis involves acquiring power consumption traces using oscilloscopes and devices.
  • The traces are manually analyzed to extract the cryptographic key.
  • Filtering and alignment techniques are used to improve the accuracy of the analysis.
  • Template analysis is used to build models that relate intermediate values to power usage and extract keys from power traces.
  • Deep learning is introduced as a potential solution to improve the accuracy and efficiency of site analysis.
  • Backpropagation algorithm is used to train the machine to classify power traces and extract keys.
  • The ongoing process of site analysis involves using tools and getting feedback to improve the accuracy of the analysis.
The speaker demonstrates the process of filtering and alignment using a power trace with misalignment. By adjusting the parameters, the speaker is able to align the traces and improve the accuracy of the analysis. The speaker also highlights the importance of selecting relevant parts of the trace for template analysis to avoid adding noise to the training data.


Deep learning can help automate the signal analysis process in power side channel analysis. So far, power side channel analysis relies on the combination of cryptanalytic science, and the art of signal processing. Deep learning is essentially a classification algorithm, but instead of training it on cats, we train it to recognize different leakages in a chip. Even more so, we do this such that typical signal processing problems such as noise reduction and re-alignment are automatically solved by the deep learning network. We show we can break a lightly protected AES, an AES implementation with masking countermeasures and a protected ECC implementation and show a live demo of the attack in action. These experiments show that where previously side channel analysis had a large dependency on the skills of the human, first steps are being developed that bring down the attacker skill required for such attacks. This talk is targeted at a technical audience that is interested in latest developments on the intersection of deep learning, side channel analysis and security.