Design of a digital VLSI neuroprocessor for signal and image processing

Abstract

An efficient processing element for data/image processing has been designed. Detailed communication networks, instruction sets and circuit blocks are created for ring-connected and mesh-connected systolic arrays for the retrieving and learning phases of the neural network operations. 800 processing elements can be implemented in 3.75 cm*3.75 cm chip by using the 0.5 mu m CMOS technology from TRW, Inc. This digital neuroprocessor can also be extended to support fuzzy logic inference.<>