VLSI implementation of locally connected neural network for solving partial differential equations

Abstract

This brief presents a locally connected neural network for solving a class of partial differential equations. Each neural cell is designed using active and passive components. An architecture is described to control the weights between the neurons. The major benefit of the architecture is that it does not require additional space outside of the cell for routing the control lines no matter how many cells are used. The CMOS VLSI implementation of a sixteen cell network was fabricated and measured. The results of this network are compared to the numerical solution of the partial differential equations.