Paper 2021/401

Output Prediction Attacks on SPN Block Ciphers using Deep Learning

Hayato Kimura and Keita Emura and Takanori Isobe and Ryoma Ito and Kazuto Ogawa and Toshihiro Ohigashi

Abstract

Cryptanalysis of symmetric-key ciphers, e.g., linear/differential cryptanalysis, requires an adversary to know the internal structures of the targeted ciphers. On the other hand, deep learning-based cryptanalysis has attracted significant attention because the adversary is not assumed to have knowledge of the targeted ciphers except the interfaces of algorithms. Such a blackbox attack is extremely strong; thus we must design symmetric-key ciphers that are secure against deep learning-based cryptanalysis. However, previous attacks do not clarify what features or internal structures affect success probabilities; therefore it is difficult to employ the results of such attacks to design deep learning-resistant symmetric-key ciphers. In this paper, we focus on toy SPN block ciphers (small PRESENT and small AES) and propose deep learning-based output prediction attacks. Due to its small internal structures, we can build learning models by employing the maximum number of plaintext/ciphertext pairs, and we can precisely calculate the rounds in which full diffusion occurs. We demonstrate the following: (1) our attacks work against a similar number of rounds attacked by linear/differential cryptanalysis, (2) our attacks realize output predictions (precisely plaintext recovery and ciphertext prediction) that are much stronger than distinguishing attacks, and (3) swapping the order of components or replacement components affects the success probabilities of the attacks. It is particularly worth noting that swapping/replacement does not affect the success probabilities of linear/differential cryptanalysis. We expect that our results will be an important stepping stone in the design of deep learning-resistant symmetric key ciphers.