Current controlled neuro-fuzzy membership function generation

Neuro-fuzzy systems are encouraging new approaches to mimic human-like decision making in presence of the vagueness and imprecision in the input data. However, majority of current implementations of these systems are limited to small-scale software designs, ignoring the potential advantages of the analogue hardware. This paper presents a completely tunable membership function generator, which is a vital part of designing neuro-fuzzy systems. Designed using a 130nm CMOS process, this circuit provides the ability to modify the shape as well as the mid-point of a widely used bell-shaped membership function. The circuit uses 11 transistors and provides this control of the membership function with an area of 0.9 µm ² and power consumption of 12 µW. More importantly, only three control voltages are used without the need of any external digital control. Extensive simulation results are presented to verify the circuit.