Side-channel attacks process large amounts of collected data, tens of thousands of traces that may have millions of samples each. There is thus a need for high-performance implementations of algorithms used in side-channel attacks. Due to the nature of the processed data, these algorithms can efficiently implemented on parallel architectures, such as GPUs.

The student should research algorithms used in side-channel attacks and their parallel implementations. The student should then implement a selected subset of these algorithms including: trace set average, trace set variance, and trace set correlation coefficient, using a GPU computation platform such as CUDA or OpenCL. The implementation should support processing larger-than-GPU memory datasets. Finally, the performance of GPU implementations should be compared to a CPU implementation. The implementation of algorithms should take place in the pyecsca toolkit: