Security Analysis Of Lightweight Block Cipher SIMON Based On Deep Learning

In the era of digitization and informatization,the pervasive information collection equipment makes it difficult for everyone to stay alone.Information security has gradually become the focus of social attention.Lightweight block ciphers can provide the security requirements for underlying network transmission equipment and attract the attention of many researchers.Deep learning has overcome lots of obstacles in the field of artificial intelligence.It can automatically obtain the information behind the data after a series of operations including abstraction,extraction and summary.Deep learning is usually suitable for fields with complex data,noisy,and lack of unified theory.Correspondingly,the security analysis of block ciphers is facing with countless complex search and optimization strategy problems,which can be solved using Deep learning.This is the bridges to gap between deep learning and block cipher.In this paper,the deep learning technology is used to the security analysis of SIMON,a lightweight block cipher algorithm released by the National Security Agency(NSA)in2013.After systematically explaining the characteristics of the round function of the SIMON algorithm and related theories of deep learning,the neural network differential distinguishers are built.Furthermore,the performance of the distinguishers is analyzed and improved in some aspects,such as the network structure and the data set.The main research results of this paper are as follows:(1)The feedforward neural network and the convolutional neural network are used to simulate the case of single input differential to multi output differential in multi differential cryptanalysis.Some deep learning distinguishers of 6-round(or even 9-round)reduced SIMON32/64 are designed,and both the advantages and disadvantages of the two neural network structures under different conditions are investigated.A candidate key sieving method for the 9-round reduced SIMON32/64 is also presented by extending the 7-round distinguisher of the feed-forward and the convolution neural networks,where one round forward and one round backward of this 7-round distinguisher are respectively considered.The experimental results show that 65535 candidate keys were dramatically reduced to 675 by only using 128 chosen plaintext pairs.Compared with the traditional differential distinguishers of reduced SIMON 32/64,the new distinguishers combined with deep learning notably reduce both the time complexity and data complexity.(2)Aiming at the problem that the accuracy of neural network distinguisher based on deep learning decreases with the increase of iteration rounds,the influence of input differential data set on the performance of neural network distinguisher is studied.It is found that the input difference data set in the same network structure is the key factor affecting the performance of neural network distinguisher after discussing the difference characteristics of Simon round function,the position of input difference in the difference trail,the difference probability and Hamming weight.A good input difference should have the characteristics of high probability and low Hamming weight.According to this principle,an improved algorithm for neural network discriminator based on STA/SMT is proposed,which uses automated search technology STA/SMT to search for input differences with low Hamming weight and high probability.This algorithm effectively avoids the blind selection of the input difference and improves the accuracy of the neural network distinguisher.