A New Scheme for Implementing S-box Based on Neural Network

Abstract

S-box (Substitution box) is one of the most important components in the block cipher. As the high non-linearity of neural network (or artificial neural network, ANN) is in high accordance with the properties of cipher, the application of neural network in cryptography becomes a significant orientation. In this paper, we present a new scheme for implementing S-box used in ciphers basing on neural network. Differing from the previous network models, the proposed network, which can be used to implement any Boolean function in S-box, consists of multiple neural network perceptrons, and each perceptron only has a low number of input variables (4-bits input). By DNA-like learning algorithm, it is very convenient to train the weight and threshold values of the network.