An adaptive neuro/fuzzy CMOS chip

Summary form only given. A modular chip, Tinychip in 2 micron CMOS process, contains a fuzzy-neuro system that is comprised of the usual fuzzification, inference, and defuzzification blocks. The circuit building blocks consist of single-transistors, differential pairs, transconductance multipliers, and bump circuits. The fuzzification and the defuzzification blocks each uses single-layer neural mapping with nonlinear synapses composed of multipliers and differential pairs. The inference block is realized by a two-layer feedforward network using nonlinear synapses and learning circuitries based on the modified, continuous-time back-propagation and optimal control techniques. The fuzzy-neuro chip can be embedded within a control processing loop and work to preserve the control system´s overall stability and performance