Paper 2023/1615

Order vs. Chaos: A Language Model Approach for Side-channel Attacks

Abstract

We introduce the Order vs. Chaos (OvC) classifier, a novel language-model approach for side-channel attacks combining the strengths of multitask learning (via the use of a language model), multimodal learning, and deep metric learning. Our methodology offers a viable substitute for the multitask classifiers used for learning multiple targets, as put forward by Masure et al. We highlight some well-known issues with multitask classifiers, like scalability, balancing multiple tasks, slow learning, large model sizes, and the need for complex hyperparameter tuning. Thus, we advocate language models in side-channel attacks. We demonstrate improvements in results on different variants of ASCAD-V1 and ASCAD-V2 datasets compared to the existing state-of-the-art results. Additionally, we delve deeper with experiments on protected simulated datasets, allowing us to control noise levels and simulate specific leakage models. This exploration facilitates an understanding of the ramifications when the protective scheme's masks do not leak and allows us to further compare our approach with other approaches. Furthermore, with the help of unprotected simulated datasets, we demonstrate that the OvC classifier, uninformed of the leakage model, can parallelize the proficiency of a conventional multi-class classifier that is leakage model-aware. This finding implies that our methodology sidesteps the need for a predetermined leakage model in side-channel attacks.