Hardware implementation of neuro-fuzzy system with the analogue-digital hybrid neural chip

The paper describes a novel method for VLSI implementation of a neuro fuzzy hybrid system for high speed and flexible operation. Proposed circuits are based on mixed analogue and digital operations and include a linear voltage controlled MOSFET resistance. Membership functions, MIN functions and other functions are available by using either linearly controlled resistance or pulse operation, which are already used in the neural chip of URAN. Any numbers of membership functions are evaluated in parallel and are incorporated in a fuzzy neural network, because the same principle and the same basic circuit are employed for the fuzzy system and the neural network. The nominal unit operation time is evaluated as 500,000 operations per second and the arbitrary complex fuzzy membership function is evaluated at one time. Concerning the neural network, one programmable basic cell corresponds to one synaptic connection. The same types of voltages or voltage pulses are used to signal inputs and outputs through the neural network and the fuzzy system