An adaptive bio-inspired analog silicon retina

An analog vision chip has been designed on the basis of retina models. Biological retinas perform massively parallel computation well suited to low level visual processing applications such as edge extraction, motion detection and light adaptation. An analog electrical model has been developed that is suitable for designing analog neural network circuits with low power consumption. This 4000 pixel neuromorphic circuit extracts edges or detects motion over a large dynamic range of luminosity and velocity. Experimental results are presented.