Improved (Related-Key) Differential-Based Neural Distinguishers for SIMON and SIMECK Block Ciphers.

This article focuses on improving neural distinguishers (NDs) based on (related-key) differential cryptanalysis for Simon and Simeck lightweight block ciphers. By incorporating multiple ciphertext pairs with an enhanced data format and leveraging domain knowledge of differential cryptanalysis, the authors design a SE-ResNet deep learning architecture that achieves higher accuracy in distinguishing ciphertext pairs than previous methods. They systematically investigate the impact of input differences on hybrid distinguishers (HDs) to select optimal parameters, and successfully construct both differential and related-key differential neural distinguishers (RKNDs) for round-reduced versions of Simon 32/64, Simon 64/128, Simeck 32/64, and Simeck 64/128. The study demonstrates improved accuracy over prior work and provides the first related-key neural distinguishers against Simon-like ciphers, highlighting the potential of deep learning combined with cryptanalytic domain knowledge in block cipher analysis.