Design of a digital neural chip: application to optical character recognition by neural network

Abstract

Presents the design of a dedicated chip achieving the recognition phase of layered neural networks. General back-propagation (GBP) and learning vector quantization (LVQ) neurons can be emulated on this chip (called the OCR-chip). It consists of five processors: four neuron processors interconnected in a ring, each are computing several states of different GBP neurons, and a LVQ processor used to compute the states of the LVQ neurons. Connections between GBP neurons folded on the same processor are implemented in each processor by using an address generator based on modulus m counters. An optical character recognition (OCR) neural network (840 neurons in 4 layers and 800 LVQ neurons) is used as demonstrator.<>