Print Email Facebook Twitter NASCTY Title NASCTY: Neuroevolution to Attack Side-Channel Leakages Yielding Convolutional Neural Networks Author Schijlen, Fiske (Student TU Delft) Wu, Lichao (Student TU Delft) Mariot, L. (TU Delft Cyber Security; University of Twente) Date 2023 Abstract Side-channel analysis (SCA) is a class of attacks on the physical implementation of a cipher, which enables the extraction of confidential key information by exploiting unintended leaks generated by a device. In recent years, researchers have observed that neural networks (NNs) can be utilized to perform highly effective SCA profiling, even against countermeasure-hardened targets. This study investigates a new approach to designing NNs for SCA, called neuroevolution to attack side-channel traces yielding convolutional neural networks (NASCTY-CNNs). This method is based on a genetic algorithm (GA) that evolves the architectural hyperparameters to automatically create CNNs for side-channel analysis. The findings of this research demonstrate that we can achieve performance results comparable to state-of-the-art methods when dealing with desynchronized leakages protected by masking techniques. This indicates that employing similar neuroevolutionary techniques could serve as a promising avenue for further exploration. Moreover, the similarities observed among the constructed neural networks shed light on how NASCTY effectively constructs architectures and addresses the implemented countermeasures. Subject genetic algorithm (GA)neural architecture search (NAS)neural network (NN)side-channel analysis (SCA) To reference this document use: http://resolver.tudelft.nl/uuid:5fe00d29-0427-454d-8ac0-1fc22ecf8b73 DOI https://doi.org/10.3390/math11122616 ISSN 2227-7390 Source Mathematics, 11 (12) Part of collection Institutional Repository Document type journal article Rights © 2023 Fiske Schijlen, Lichao Wu, L. Mariot Files PDF mathematics_11_02616.pdf 657.11 KB Close viewer /islandora/object/uuid:5fe00d29-0427-454d-8ac0-1fc22ecf8b73/datastream/OBJ/view