Since 2017 several national and private schemes mandate the application of deep learning in side-channel analysis (SCA) evaluations of state-of-the-art secure crypto implementations. Deep learning is an advanced machine learning technique that uses a trained deep neural network to classify datasets. As such, deep learning can be used as an advanced assessment tool to assess whether side-channel leakage is present.
A training procedure over a big dataset is conducted, which sometimes can be a very time-consuming step. The design of a neural network involves the understanding of hyper-parameter effects on specific datasets. This training is also intended to be a guideline for the hyper-parameter definition of a deep neural network.
In the field of security evaluations, classical side-channel attacks like template attacks or correlation power analysis are nowadays considered as main attack methods. These attacks require an end-to-end procedure starting in a leakage assessment strategy, followed by human action like optional dimensionality reduction method, alignment, filtering and the application of the attack method itself. Deep learning provides an alternative framework scenario where the leakage assessment and the dimensionality reduction are inherently performed by a neural network. Furthermore, some neural network topologies can overcome misalignment issues in side-channel traces. This way, deep learning replaces human efforts with machine effort.
You selected to get the course Deep SCA (DLSCA). Please Login or Signup to get it.