Options
1995
Conference Paper
Titel
Survey and current status of neural network hardware
Abstract
During the last ten years neural networks made their way from research toys to well accepted industrial tools. Actual application however requires real-time capable implementations of neural algorithms. The wide range of applications and performance requirements as well as the considerable number of algorithms motivated numerous efforts of neural network hardware implementation that in their numbers and diversity are hard to overlook even for researchers working in the field. Introducing, this paper will give an outline of relevant neural network implementation categories, criteria for performance assessment, and of principle design issues, e. g., accuracy of computation. Ensuing, a survey of state-of-the-art implementations will be given, focusing on recent designs that contribute intriguing and promising features or architectures, and systems that made all the way from concept to actual scientific or commercial application. Concluding, a perspective of anticipated future developments will be given.