17 Jun 2002
Neural networks made from laser diodes could speed up telecommunications networks by giving them all-optical data switching.
An optical neural network with up to 32 neurons and 12 inputs has been built by Dutch researchers at the Eindhoven University of Technology and Philips Research Laboratories. All the key neural operations are performed optically rather than electronically.
The system relies on optical feedback to a multimode laser diode. The longitudinal modes of the laser represent neurons. The modes are separated by a grating and fed through a fast liquid crystal display which adjusts the power of selected wavelengths. Another grating recombines the wavelengths and the beam is fed back to the laser by means of a loop mirror.
The researchers believe that optical neural networks could form the basis of high-speed optical telecommunications systems. Possible applications include the routing, detecting and switching of data. Before these applications can be realized the bit error rate, currently 10-5, must be reduced further.
This research appeared in the 10 July issue of the journal Applied Optics.
SH
© 2024 SPIE Europe |
|