Accession Number : ADA311781
Title : Adaptable Locally-Interconnected Architectures.
Descriptive Note : Final rept. Nov 93-May 96,
Corporate Author : ARIZONA STATE UNIV TEMPE DEPT OF ELECTRICAL AND COMPUTER ENGINEERING
Personal Author(s) : Akers, Lex
PDF Url : ADA311781
Report Date : AUG 1996
Pagination or Media Count : 36
Abstract : This project has built on our previous work in developing a theory of the processing power of locally interconnected architectures and the studies of the implementation of such architectures. Three locally interconnect architectures, inspired by biology, were developed and fabricated in VLSI. These systems are an analog Gaussian basis circuit integrated with a nonvolatile storage memory cell, a localed spatial frequency filter, and a habituation system. Analog Gaussian basis circuit integrated with a nonvolatile storage memory cell was developed. Hardware implementations of the Gaussian basis circuit with on chip learning is needed for real time and portable applications. Each Gaussian basis cell is symbolically interlinked with its own longterm storage memory cell forming a highly localized architecture. Experimental results were obtained. Receptive field structures found in the visual cortex of the mammalian brain act as oriented, localized spatial frequency filters. These receptive field structures resemble Gabor filters. We have implemented analog VLSI cells whose outputs resemble the receptive field profiles. And lastly, rehabituation is a feature of habituation which allows a system to more rapidly disregard inputs that are not novel. We have implemented habituation, rehabituation and recovery in hardware. Our experimental results demonstrate these responses.
Descriptors : *NEURAL NETS, *COMPUTER ARCHITECTURE, REAL TIME, VERY LARGE SCALE INTEGRATION, CHIPS(ELECTRONICS), MEMORY DEVICES, SPATIAL FILTERING, LEARNING, ANALOG SYSTEMS, NONVOLATILE MEMORIES, VISUAL CORTEX, BIONICS, HABITUATION LEARNING.
Subject Categories : Bionics
Distribution Statement : APPROVED FOR PUBLIC RELEASE