Multiplexed Circuit for Star-CNN Architecture

Star-CNN is a particular architecture of cellular neural networks that has been recently proposed. This dynamic nonlinear system is defined by connecting N identical dynamical system called local cell with a central system in the shape of a star. Each of the local cells communicates with others through the central system. Because of the hardware requirements of such a system, its implementation comes out extremely expansive from the silicon area occupation point of view. This paper presents a hardware implementation of this new CNN architecture based on a time division approach that allows to significantly reduce the silicon area occupation by minimizing the number of the analogue multipliers