Temperature Dependence of Fixed Pattern Noise in Logarithmic CMOS Image Sensors

This paper presents a model that explains the contribution of temperature to the fixed pattern noise (FPN) in a logarithmic CMOS image sensor. Based on this model, a simpler model is proposed to facilitate the calibration and correction of FPN. To avoid nonlinear optimization, the variation of photodiode leakage current from one pixel to another is neglected. The simplified model uses the dark response of pixels, which depends only on temperature, to help predict FPN in the light response, which depends on temperature and illuminance. Calibration requires images of a uniform stimulus taken at different temperatures and illuminances, which need not be measured. After calibration, FPN is corrected in an arbitrary image using a dark image at the same temperature, which is taken infrequently. Through simulation, using mismatch data from a real CMOS process, an improvement is shown in the residual error per image after calibration, when the proposed method is compared to an established method that does not account for temperature dependence.