earticle

영어

In this paper, we propose deep learning-based cryptanalysis on lightweight block ciphers, SIMON and SPECK. The block-sized bit arrays are encrypted with the block ciphers applying different number of round functions. The deep learning models are trained to generate the ciphertexts from the plaintexts and recover the plaintexts from the ciphertexts, which are attacks called Encryption Emulation (EE) and Plaintext Recovery (PR), respectively. The results are compared by using Bit Accuracy Probability (BAP) for each bit. It is shown that the round-reduced SIMON is higher BAP than round-reduced SPECK32/64. These results indicate that the round-reduced SIMON32/64 is more vulnerable than the round-reduced SPECK32/64.