Mismatch reduction with relative reset in integrate-and-fire photo-pixel array

We have reduced the mismatch error in a bio-inspired vision sensor, of the kind that is popularly known as dasiaoctopus retinapsila. Mismatch and noise reduction techniques developed for traditional imagers, like correlated double sampling (CDS), cannot be applied to this frame-free asynchronous vision sensor. In an dasiaoctopus retinapsila, each self timed pixel integrates photo current until a threshold is reached, whereupon it creates a pixel-event and is reset. To reduce threshold mismatch, 1/f and reset noise in this implementation, pixels are not reset to a common baseline but to a level that is relative to the individual firing threshold. This is achieved by subtracting a fixed charge from the integrating node. The matching of that charge is dependent on the matching of a carefully designed capacitor. Monte Carlo simulations of the pixel with manufacturer mismatch statistic parameters predicted a total pixel mismatch of only 0.26%. The physical implementation on a AMS 0.35 micro-meter ASIC with a resolution of 74 times 74 pixels achieved a mismatch of below 1%.