Can KANs Do It? Toward Interpretable Deep Learning-based Side-channel Analysis

Kota Yoshida, Sengim Karayalcin, Stjepan Picek

Abstract: Recently, deep learning-based side-channel analysis (DLSCA) has emerged as a serious threat against cryptographic implementations. These methods can efficiently break implementations protected with various countermeasures while needing limited manual intervention. To effectively protect implementation, it is therefore crucial to be able to interpret \textbf{how} these models are defeating countermeasures. Several works have attempted to gain a better understanding of the mechanics of these models. However, a fine-grained description remains elusive. To help tackle this challenge, we propose using Kolmogorov-Arnold Networks (KANs). These neural networks were recently introduced and showed competitive performance to multilayer perceptrons (MLPs) on small-scale tasks while being easier to interpret. In this work, we show that KANs are well suited to SCA, performing similarly to MLPs across both simulated and real-world traces. Furthermore, we find specific strategies that the trained models learn for combining mask shares and are able to measure what points in the trace are relevant.