International Association
for Cryptologic Research

Deep learning-based side-channel analysis (SCA) became the de facto standard in the profiling SCA. Still, this does not mean it is trivial to find neural networks that perform well for any setting. Based on the developed neural network architectures, we can distinguish between small neural networks that are easier to tune and less prone to overfitting but can have insufficient capacity to model the data. On the other hand, large neural networks would have sufficient capacity but can overfit and are more difficult to tune. This always brings an interesting trade-off between simplicity and performance.

This paper proposes using a pruning strategy and recently proposed Lottery Ticket Hypothesis to improve the deep learning-based SCA. We demonstrate that we can find smaller neural networks that perform on the level of larger networks, where we manage to reduce the number of weights by more than 90% on average. Additionally, we show that pruning can help prevent overfitting and the effects of imbalanced data, reaching top attack performance for small networks when larger networks do not manage to break the target at all.