Digitally-programmable analog cells for artificial neural networks

Author

Almeida, A. Passos ; Franca, J.E.

Author_Institution

Dept. of Electr. & Comput. Eng., Inst. Superior Tecnico, Lisbon, Portugal

fYear

1994

fDate

1-4 May 1994

Firstpage

566

Lastpage

569

Abstract

This paper describes the design and experimental evaluation of integrated mixed analog-digital cells for the implementation of artificial neural networks (ANNs) suitable for applications in nonlinear control and audio signal processing. A novel synapse architecture is proposed using a quasi-passive D/A converter (DAC) followed by a 4 quadrant analog-digital multiplier. The neurons are built with MOS transistors operating in the saturation region and exploiting the inherent quadratic characteristics. Experimental results obtained from a demonstration 1.2 μm CMOS prototype chip are in good agreement with the envisaged design specifications

Keywords

CMOS integrated circuits; digital-analogue conversion; mixed analogue-digital integrated circuits; multiplying circuits; neural chips; 1.2 micron; CMOS prototype chip; MOS transistors; analog-digital multiplier; artificial neural networks; audio signal processing; digitally-programmable analog cells; mixed analog-digital cells; nonlinear control; quadratic characteristics; quasi-passive D/A converter; saturation region; synapse architecture; Analog-digital conversion; Application specific integrated circuits; Artificial neural networks; Flip-flops; MOSFETs; Neurons; Prototypes; Shift registers; Switches; Voltage;

fLanguage

English

Publisher

ieee

Conference_Titel

Custom Integrated Circuits Conference, 1994., Proceedings of the IEEE 1994

Conference_Location

San Diego, CA

Print_ISBN

0-7803-1886-2

Type

conf

DOI

10.1109/CICC.1994.379658

Filename

379658