Smart Neural Tools to Outsmart SIMON and SIMECK Ciphers

In the realm of cryptography, deep learning has been making waves. At a conference called CRYPTO in 2019, a researcher named Gohr showed how deep learning could be used to crack the NSA's Speck32/64 cipher. Later, a team led by Lu improved on this by targeting SIMON and SIMECK ciphers with related-key differential neural distinguishers. Building on their work, a new method has been developed to enhance these distinguishers even further. The key here is to pick the right input differences. To do this, scientists have come up with a smart way using weighted bias scores. This helps figure out which input differences are most suitable for the job. Using two input differences instead of one, the new method shows better results than the previous ones by Lu et al. It's like having a smarter tool for the job.

For SIMON32/64, the improvement is clear. The accuracy for 12-round and 13-round basic related-key differential neural distinguishers jumped by 3% and 1. 9%, respectively. Even more impressive, the enhanced method beats the basic related-key differential neural distinguishers. For 13-round SIMON32/64, 13-round SIMON48/96, and 14-round SIMON64/128, the accuracy increased to 0. 567, 0. 696, and 0. 618 from 0. 545, 0. 650, and 0. 580. SIMECK ciphers also saw improvements. For 15-round SIMECK32/64, 19-round SIMECK48/96, and 22-round SIMECK64/128, the accuracy went up to 0. 568, 0. 523, and 0. 526 from 0. 547, 0. 516, and 0. 519.

Actions