An integrated device capable of performing edge and motion detection in real time for image processing system is to be built. The device is based on the "silicon retina" of Carver Mead, but adds optical outputs, allowing the preprocessed images to be fed directly to further optical parallel processing stages. These "retinas" are VLSI (very large scale integration) implementations of neural networks which duplicate many of the functions of the vertebrate retina. The technology used here, the addition of ferroelectric liquid crystal light modulators to standard CMOS (complementary metal oxide semiconductor) VLSI, can utilize the full repertoire of VLSI electronic functionality including photodetection, and can be applied to individual die, making for rapid research iterations and the economical application to custom products. its 104Hz readout rate is accomplished with minimal power dissipation, and is more than fast enough for neural image processing. The work focuses on developing techniques for isolating read light from write light in these completely optically interconnected devices, and on developing retinal binocular images. The device should have many commercial applications for example, in the fields of optical inspection and robotics.
