Programmable fuzzifier circuits with high precision for analog Neuro-fuzzy system

In this paper, we propose a Membership function Generator (MFG) circuit in the form of Gaussian and trapezoidal for Neuro-fuzzy applications which is programmed by four voltage signals. Two signals define the knees where output signals begin falling or rising, while other ones change the rising or falling slopes of Gaussian and trapezoidal functions, independently. So there is no need to change the sizes of transistors or to switch parallel transistors. This is also caused the circuit flexibility increases and the chip area decreases. Using two stages, the accuracy of the circuit to generate Gaussian function is improved. Since three generated functions (small, medium, and large) are produced by a circuit simultaneously, low power consumption with small occupied area is obtained. Finally, simulation results which were done by HSPICE (level49) in 0.35μm CMOS process are presented. The Layout of the circuit realized less than 1300μm×μm area.