Large photoconductive arrays for optically programmed neural networks.

We have recently completed a series of experiments which indicates that we can build neural networks many times larger than have been previously attempted. The synaptic elements in these devices are continuously programmable and exhibit full analog depth. Our approach is to use an array of amorphous silicon photoconductive elements to achieve a fully interconnected resistive matrix as is required for neural networks. In this study, we have looked in some detail at the properties of individual photoconductive synapses to determine their electrical and optical properties. Working from these results we have begun preliminary experiments looking at the fabrication and operation of large arrays of these devices. At the same time we are exploring system issues required for using these devices in a number of applications.