SWITCHED-CAPACITOR NEURAL NETWORKS FOR DIFFERENTIAL OPTIMIZATION

The primary goal of this paper is to solve some types of optimization problems whose objective functions and constraints are smooth and continuously differentiable and which are not suited for digital signal processing. The optimization problems are mapped into systems of first- and second-order non-linear ordinary differential equations and/or systems of difference equations. These systems of equations are simulated by appropriate switched-capacitor (SC) circuits employing some neural network (neurobiological) principles. New switched-capacitor architectures for on-line solving of non-linear optimization problems are proposed and their properties are investigated. Various circuit structures are investigated to find which are best suited for SC CMOS implementation. The structures developed exhibit a high degree of modularity, and a relatively small number of basic building blocks (computing cells) are required to implement effective and powerful optimization algorithms. Basic mathematical operations, e.g. multiplication, addition and non-linear scaling transformation, are accomplished employing advanced SC techniques. The validity and performance of the circuit structures developed are illustrated by intensive computer simulations employing TUTSIM and NAP programmes.