Mixed-signal VLSI design of adaptive fuzzy systems

Hardware realization of adaptive fuzzy systems is discussed. A highly-parallel architecture based on an active-rule driven scheme is considered so that only the parameters associated with the active rules take part in the inference and learning phases. Mixed-signal circuits are used to implement a gradient-descent algorithm for tuning consequent values and a weight perturbation method to adjust suitably chosen antecedent parameters. Several applications of adaptive fuzzy chips are discussed showing that low-cost realizations are feasible