Implementation of multiplicative lateral inhibition in a GaAs sensory neural-network photodetector array

Abstract

An integrated circuit (IC) implementation of a multiplicative lateral inhibition (MLI) sensory neural network applied to a linear, 30-element photodetector array is described. The neural cells and the photodetector elements were fabricated using a five-mask GaAs depletion-mode metal-semiconductor field-effect transistor (MESFET) process. Unilateral interconnections between neighboring neural cells were achieved using a single MESFET, and each neural cell thus consists of only five transistors and three level shifting Schottky diodes. This compact analog implementation is shown to exhibit image processing features commonly associated with biological vision systems, including adaptation to mean luminance levels and enhanced contrast around spatial edges. >