In this paper, we present an optoelectronic architecture for the implementation of a radial basis function network (RBFN). This architecture, which exploits the optoelectronic hardware of the Hopfield neural network demonstrator, is shown to be capable of operating with positive weighted interconnections. A systematic investigation of the major sources of optical interconnection uniformity reveals the critical stages in the fabrication of these optical elements and allows an estimate of the robustness of the network to these errors to be made. © 2005 IOP Publishing Ltd.
- Diffractive optics
- Neural networks