Parallel interconnection and neural network systems employing semiconductor array technology and computer-generated holography |
| |
Authors: | N. Barnes P. Healey P. McKee A. W. O'Neill M. A. Z. Rejman-Greene E. G. Scott R. P. Webb B. R. White D. Wood |
| |
Affiliation: | (1) British Telecom Research Laboratories, Martlesham Heath, Ipswich, Suffolk, UK |
| |
Abstract: | ![]() An experimental 16-channel parallel interconnection system able to support 100 Mbits–1 per channel and an opto-electronic neural network operating at up to 50 Mbits–1 have been constructed to demonstrate the potential of optics in processing systems. Both experiments operate at a wavelength of 1.5 m. Components developed for these systems include arrays of InGaAs/InP MQW surface modulators for low-power electrical-to-optical conversion; InGaAs/InP for detector arrays, which are hybrid integrated with GaAs amplifier arrays; and computer-generated holograms for efficient beam splitting and to encode weights in the optical beam intensities. Each of these demonstrations has considerable scope for increasing the degree of parallelism and the operating speed. Recent modifications to the neural network enable the weights to be varied and training has been demonstrated with a novel algorithm that uses the high operating speed to advantage. |
| |
Keywords: | |
本文献已被 SpringerLink 等数据库收录! |