Design of an optical fixed-weight learning neural network

This paper deals with the design, analysis, and simulation of a prototype optical fixed-weight learning neural network. This type of network could have learning rates five orders of magnitude faster than networks based on Von-Neumann platforms. This network has an embedded learning algorithm and dynamically learns new mappings by changing recurrent neural signal strengths. This greatly speed up optical neural network learning since the medium containing the synaptic weights does not change during learning. Software simulations suggest that this design is sound. The physical implementation and evaluation of the prototype were reported elsewhere.