ABSTRACT
This chapter reviews some of the pioneering and representative work in solid state implementation of neural networks. Graf et al. fabricated an associative memory neural network chip which, given an incomplete vector finds the closest, in the Hamming distance sense, memory state. Furthermore even though the association is a distributed operation and summation is performed via analog current summing, the chip has the peripheral circuitry of a standard digital static Random Access Memory. The associative memory and learning circuits suffer from lack of purpose. The sensory neural networks do try to mimic functional capabilities of biological systems by replicating some aspects of their neuroanatomy and electrophysiology. A research group at Bell Communications Research has been among the first to concentrate on implementing learning networks. The chapter describes framework for electronic realization of the networks whose biological feasibility and relevance and self-sufficient processing capabilities. It discusses the different components of the design framework.
