A large number of VLSI implementations of neural network models have been reported. The diversity of these implementations is noteworthy. This paper attempts to put a group of representative VLSI implementations in perspective by comparing and contrast(cid:173) ing them. Design trade-offs are discussed and some suggestions forthe direction of future implementation efforts are made.
IMPLEMENTATION Changing the way information is represented can be beneficial. For example a change of representation can make information more compact for storage and transmission. Implementation of neural computational models is just the process of changing the representation of a neural model from mathmatical symbolism to a physical embodi(cid:173) ement for the purpose of shortening the time it takes to process information according to the neural model.
FLEXIBIliTY VS. PERFORMANCE
Today most neural models are already implemented in silicon VLSI, in the form of pro(cid:173) grams running on general purpose digital von Neumann computers. These machines are available at low cost and are highly flexible. Their flexibility results from the ease with which their programs can be changed. Maximizing flexibility, however, usually results in reduced performance. A program will often have to specify several simple op-