Deep-Learning-Based Cryptanalysis of Lightweight Block Ciphers Revisited.

Area of Science:

• Cryptography
• Artificial Intelligence
• Machine Learning

Background:

• Deep learning is increasingly applied to cryptanalysis, particularly for key recovery using known plaintext.
• Lightweight block ciphers like S-DES, S-AES, and S-SPECK are common targets for cryptanalysis.

Purpose of the Study:

• To propose a novel cryptanalysis method using state-of-the-art deep learning techniques.
• To evaluate the method's effectiveness on lightweight block ciphers S-DES, S-AES, and S-SPECK.

Main Methods:

• Utilized deep learning architectures incorporating residual connections and gated linear units.
• Applied known-plaintext cryptanalysis to recover secret keys for selected lightweight block ciphers.

Main Results:

• Achieved a 93.16% reduction in training parameters compared to prior state-of-the-art methods.
• Increased average bit accuracy probability by approximately 5.3%.
• Successfully performed cryptanalysis on S-AES and S-SPECK with 12-bit and 6-bit keys, respectively.

Conclusions:

• The proposed deep learning method offers significant improvements in efficiency and accuracy for lightweight cipher cryptanalysis.
• Deep learning-based key recovery for full-round, full-key modern cryptography remains practically infeasible with current techniques.