Dynamic response of a CNN hardware system by Lyapunov analysis and Wolf algorithm

This paper presents an analysis of the dynamic response of a CNN hardware system with programmable synaptic weights. The performed measurements on an actual CNN board have been mathematically analized by means of the Lyapunov method which is suitable to detect the stability of dynamic systems. In particular, the Wolf algorithms BASGEN and FET for the Lyapunov exponent evaluation have been used. In this way it has been possible to identify the ranges of stable, unstable and chaotic responses of the CNN in function of the different imposed CNN synaptic weights. Finally, the description of the CNN Board and the implemented dedicated software for the evaluation of the Lyapunov coefficients (LeChaDe: Lyapunov Exponent Chaos Detection) are described