High density VLSI implementation of neural networks and fuzzy systems

Author

Murgan, A.T. ; Dogaru, Radu

Author_Institution

Dept. of Electron. & Commun., Politehnica Univ. of Bucharest, Romania

fYear

1995

fDate

11-14 Oct 1995

Firstpage

367

Lastpage

370

Abstract

A new design flow for high density neuro-fuzzy systems is presented, based on the basic building blocks offered by the current-mode design style rather than on imitating one or other neuro-fuzzy mathematical model. A comparison between digital and analog approaches is carried out, concluding that the current-mode design technique is the most suitable for high density VLSI implementations of neural and fuzzy systems. Original implementation solutions for a nonmonotonic activation function and for a weighted summation module followed by a hard nonlinearity (CMOS perceptron) are presented

Keywords

CMOS analogue integrated circuits; VLSI; analogue processing circuits; circuit CAD; fuzzy neural nets; fuzzy systems; integrated circuit design; neural chips; perceptrons; switched current circuits; CMOS perceptron; current-mode design style; fuzzy systems; hard nonlinearity; high density VLSI implementation; neural networks; neuro-fuzzy systems; nonmonotonic activation function; weighted summation module; Computer architecture; Delay effects; Fuzzy neural networks; Fuzzy systems; Mathematical model; Neural networks; Neurons; Silicon; System testing; Very large scale integration;

fLanguage

English

Publisher

ieee

Conference_Titel

Semiconductor Conference, 1995. CAS'95 Proceedings., 1995 International

Conference_Location

Sinaia

Print_ISBN

0-7803-2647-4

Type

conf

DOI

10.1109/SMICND.1995.495038

Filename

495038