Analog VLSI implementation of adaptive neuro-fuzzy inference systems

This paper presents an analog VLSI implementation of adaptive neuro-fuzzy inference systems (ANFIS). Stochastic perturbative techniques, which are more VLSI friendly than standard learning techniques such as back-propagation, are used for on-chip learning. The system is tested by the task of predicting the Mackey-Glass chaotic time series. The system is built and simulated with SPICE using CMOS 1.2 /spl mu/m N-well technological parameters with 5 V supply. The obtained results have shown how on-chip learning is very fast compared to software implemented learning algorithms.