A cellular neural network modeling the behavior of reconfigurable cellular automata

Abstract The design and implementation of a cellular neural network (CNN) architecture capable of modeling the behavior of reconfigurable cellular automata (CA) is presented in this paper. Despite the simplicity of their structure CA are capable of exhibiting extremely complex behaviors. This motivates the development of a reconfigurable CA architecture, capable of exhibiting all types of complex behaviors found in the different CA classes. However, the hardware complexity for developing such an architecture is very high and comes in direct contrast with the inherent modularity, regularity, locality and homogeneity properties of CA architectures. The CNN architecture presented in this paper is capable of learning all the qualitative different CA behaviors, and it matches all the inherent advantages of CA architectures. Additionally, any hardware reconfiguration costs are totally avoided.