A generic architecture for intelligent system hardware

This paper brings out the similarities between the functional models of various intelligent systems and proposes a generic architecture for their implementation in hardware. Hardware implementation of intelligent models has proved to be efficient over software implementations and various architectures have been proposed for their implementation [1]-[3]. However, designing such a system is a difficult task and requires a custom implementation for the specific application. Our attempt is to provide a common hardware platform that can be programmed and used for three different intelligent approaches: fuzzy systems, neural networks and neuro-fuzzy systems. The functional similarity between all the three approaches is exploited and a generic architecture is proposed. After ascertaining the best approach for the problem at hand, the system can be programmed accordingly. Through computer simulation of the proposed architecture, targeting a Xilinx CoolRunner device, we show that the system performs well in all modes of operation.