VLSI implementation of a digital neural network with on-chip learning

Abstract

This thesis explores the VLSI implementation of a neural network with on-chip learning capabilities. This digital ASIC is intended to be used as the basis for a high-speed neural accelerator board. A unique digital neural network architecture is presented and analyzed. A novel VLSI design style is introduced and developed. Implementation details and design issues are also presented. Fundamental issues of neural networks are introduced and major learning paradigms are discussed. A thorough examination of digital ANNs is presented.