An analog continuous-time programmable neural network

Abstract

The authors introduce economical, simple, and versatile MOS cells that achieve vector scalar product and are applicable in the MOS implementation of feedback/feedforward neural networks. The vector scalar product of 2 n-tuple vector inputs is achieved by using 2(n+1) MOS transistors, thus offering an economical alternative to VLSI analog neural networks. The analog neural network is realized by interconnecting double inverters to the new vector scalar product circuits. The output voltage is tunable programmable DC control voltages. Experimental results are presented. >