A continuous time pattern recognition retina

We present in this paper a pattern recognition retina operating in continuous time. During a programming phase, the reference image is acquired by the sensor. After thresholding, the resulting binary image is stored in the sensor´s pixels. During the recognition phase, the retina calculates, in an analog way and in continuous time, the zero displacement intercorrelation of the current image with the binary reference image and its complement. These two intercorrelation values are output as currents. By comparing these currents to expected values, determined during the programming phase, a shift of the pattern or a difference between the observed and programmed pattern can be detected. A 100×100 pixels validation chip was fabricated using a standard CMOS triple level metal 0.6µm process. Characterization results are presented.