Title :
Analog programmable chips for implementing ANNs using capacitive weight storage
Author :
Kub, F.J. ; Moon, K.K. ; Modolo, J.A.
Author_Institution :
US Naval Res. Lab., Washington, DC, USA
Abstract :
The design approach for fully multiplexed 32×32 and 128×64 programmable analog vector-matrix multipliers utilizing capacitive weight storage is described. A fully differential design has been used throughout the signal path for cancellation of common-mode noise feedthroughs and reduction of multiplier and multiplexer offsets. The 32×32 vector-matrix multiplier circuit has been fabricated using the MOSIS 2-μm P-well CMOS technology. A method to increase dynamic range and a method to reduce power dissipation by an order of magnitude are described. Results showing operation of the 32×32 vector-matrix multiplier circuit are reported
Keywords :
CMOS integrated circuits; linear integrated circuits; matrix algebra; multiplying circuits; neural nets; MOSIS 2-μm P-well CMOS technology; capacitive weight storage; common-mode noise feedthroughs; differential design; dynamic range; multiplexer offsets; multiplier offsets reduction; noise cancellation; power dissipation; programmable analog vector-matrix multipliers; vector-matrix multiplier circuit; Artificial neural networks; CMOS technology; Capacitance; Decoding; Dynamic range; MOSFET circuits; Moon; Multiplexing; Signal processing algorithms; Voltage;
Conference_Titel :
Neural Networks, 1991., IJCNN-91-Seattle International Joint Conference on
Conference_Location :
Seattle, WA
Print_ISBN :
0-7803-0164-1
DOI :
10.1109/IJCNN.1991.155227