## Cryptology ePrint Archive: Report 2019/1068

Not a Free Lunch but a Cheap Lunch: Experimental Results for Training Many Neural Nets

Joey Green and Tilo Burghardt and Elisabeth Oswald

Abstract: Neural Networks have become a much studied approach in the recent literature on profiled side channel attacks: many articles examine their use and performance in profiled single-target DPA style attacks. This paper contributes to this ongoing discourse by taking a slightly different angle: we train networks for many intermediates of a typical AES implementation on an ARM Cortex-M0 processor, and compare their performance with classical profiling methods. Because the cost of finding good hyperparameters for networks is high, we demonstrate how to configure a network with a set of hyperparameters for a specific intermediate (SubBytes) that can also be used for learning the leakage of other intermediates. This is interesting because although we can't beat the no free lunch theorem (i.e. we find that different profiling methods excel on different intermediates), we can still get good value for money'' (i.e. reasonable classification results across many intermediates with reasonable profiling effort). To put the trained classifiers into side channel practice we integrate them not only into a standard profiled single-target attack on SubBytes, but we use them as part of a (multi-target) belief-propagation attack.

Category / Keywords: secret-key cryptography / AES, Inference Based Attacks, Side Channel Attacks, Template Attacks, Deep Learning, Neural Network