Digital Neurohardware: Principles and Perspectives
T. Schoenauer, Axel Jahnke, Ulrich Roth, Heinrich Klar · 1998
Over the past years a huge diversity of hardware for artificial neural networks (ANN) has been designed. This paper gives an overview of the different types of digital ANN hardware and discusses a few example architectures. The development of digital neurohardware is driven by the desire to speed-up the simulation of ANN and/or to achieve a better performance-to-cost ratio than general-purpose systems. The most basic approach to speed-up ANN algorithms is to parallelize processing. Therefore, strategies to efficiently map neural algorithms to parallel hardware are outlined. On the other hand, the performance of general-purpose sequential hardware platforms, such as workstations and PCs, has also dramatically improved in the last couple of years. Hence, at the end we raise the question, if there is still a need for neurohardware and discuss future perspectives. Keywords: Neurocomputer, Neuro-Accelerators, Neurochips, Parallel Processing 1