Reconfigurable array interconnection by photorefractive correlation

Electronic parallel processors might communicate more effectively by photons sent through glass or air than by electrons sent through wires, but quickly routing thousands of optical signals remains a problem. Previous photorefractive interconnection networks have dedicated one hologram to each input...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Applied Optics 1994-08, Vol.33 (23), p.5363-5377
Hauptverfasser: Ford, J E, Fainman, Y, Lee, S H
Format: Artikel
Sprache:eng
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Electronic parallel processors might communicate more effectively by photons sent through glass or air than by electrons sent through wires, but quickly routing thousands of optical signals remains a problem. Previous photorefractive interconnection networks have dedicated one hologram to each input channel. Instead, we compute a control image from the entire network configuration and store it as a single color-keyedvolume hologram. This lets us use hologram superposition for fast switchingbetween multiple prestored patterns. During operation, data signals from the input modulator array, powered by a shared wavelength-tunable laser, are correlated optically with one color-matched connection hologram to produce the output array. This decouples both data rate and interconnect switching speeds from the slow photorefractive response. We can display arbitrary connection weights using simple binary-phase spatial light modulators and gracefully accommodate modulator limitations by trading off control-image bandwidth for output signal-to-noise ratio. Experimental results with color-multiplexed reflection holograms in z-cut LiNbO(3) confirmed our theoretical predictions that this approach works best for densely connected networks with high fan-in to each output. We obtained an average aggregate signal-to-noise ratio of more than 200:1 for 1024 inputs and outputs.
ISSN:1559-128X
0003-6935
1539-4522
DOI:10.1364/AO.33.005363