Paper 2024/1389 Original: eprint.iacr.org/2024/1389 ↗

DL-SITM: Deep Learning-Based See-in-the-Middle Attack on AES

Abstract

The see-in-the-middle (SITM) attack combines differential cryptanalysis and the ability to observe differential patterns in the side-channel leakage traces to reveal the secret key of SPN-based ciphers. While SITM presents a fresh perspective to side-channel analysis and allows attacks on deeper cipher rounds, there are practical difficulties that come with this method. First, one must realize a visual inspection of millions of power traces. Second, there is a strong requirement to reduce noise to a minimum, achieved by averaging over 1000 traces in the original work, to see the patterns. Third, the presence of a jitter-based countermeasure greatly affects pattern identification, making the visual inspection infeasible. In this paper we aim to tackle these difficulties by using a machine learning approach denoted as DL-SITM (deep learning SITM). The fundamental idea of our approach is that, while a collision obscured by noise is imperceptible in a manual inspection, a powerful deep learning model can identify it, even when a jitter-based countermeasure is in place. As we show with a practical experiment, the proposed DL-SITM approach can distinguish the two valid differentials from over 4M differential traces with only six false positives. Extrapolating from the parameters of this experiment, we get a rough estimate of $2^{43}$ key candidates for the post-processing step of our attack, which places it easily in the practical range. At the same time, we show that even with a jitter countermeasure shifting the execution by $\pm15$ samples, the testing f1-score stays at a relatively high (0.974).