A Method of Neural Network Controller Implementation in VLSI Design

This article presents an approach to neural network implementation in VLSI, which is called as neural network controller based on Petri net. The structure of the neurons in the network is uniform; it is triggered by external events, and output place and transition signals of petri net. Two types of neurons are introduced, one is standing for serial process, and the other is used in synchronization of processes. The dual types of neurons are chained together by stimulate inputs, and compose the fabric. The controller is designed to conquer the side effects of state machine, and improves the performance and reliability. Typically, the controller is a precise description of the circuits. It is optimized to timing closure against constraints much easier than state machine. Reduplication of each node in neural network decrease single event upset (SEU). Finally, the controller is easy to rebuild. The new design flow has applied in practice, and proved effectively.