CIS像素RTS噪声的时变物理统计建模与多维非均匀抑制方法研究

The random telegraph signal noises, which are caused by the impurities and defects within the source follower and the photodiode, are the main sources of the dark noises of CMOS image sensor, and restrict severely improving the image quality and extending the dynamic range. In the existing research on analysis and suppression of the random telegraph signal noises, the precise model of the physical statistics to real-time reflect noise random change and the multidimensional nonuniform suppression method are scarce. Aiming at the above problems, this project looks the random telegraph signal noises in the pixels of CMOS image sensor as a random process. According to the physical mechanism of the random telegraph signal noises, stochastic differential equation is established and solved to derive in theory the time-varying statistical model of the pixel random telegraph signal noises. The influence of changing the device size and other factors on the distribution law of noise is analysed, according to the noise model. Using the analysis results,the multidimensional nonuniform suppression methods,which are based on spectral sensitivity function optimization, and based on the correlated double sampling mechanism, are researched to reduce the random telegraph signal noises. These methods not only suppress the random telegraph signal noises, which vary randomly in pixel and distribute non-uniformly between pixels, but also suppress the nonuniform quantum efficiency of the photodiodes between pixels, caused by random telegraph signal noises in photodiode. The research results can provide theoretical guidance and new idea for improving the image quality and the dynamic range of CMOS image sensor.

由像素源跟随管和光电二极管内部的缺陷与杂质引起的像素随机电报信号噪声，是CMOS图像传感器暗光下噪声的主要来源，严重制约了图像质量提高和动态范围扩展。在现有随机电报信号噪声的分析与抑制研究中，缺乏能够实时反映噪声随机变化的精确物理统计模型以及多维非均匀抑制方法。针对上述问题，本项目将CMOS图像传感器像素随机电报信号噪声视为随机过程，建立并求解与其物理产生机理相联系的随机微分方程，从理论上推导出像素随机电报信号噪声的时变统计模型。根据模型分析器件尺寸等因素的变化对噪声分布规律的影响。再利用分析结果，研究基于光谱灵敏度函数优化和基于相关双采样机制的多维非均匀抑制随机电报信号噪声的理论方法，不仅能消除像素内随机变化、像素间非均匀分布的随机电报信号噪声，而且能消除随机电报信号噪声引起的像素间光电二极管量子效率非均匀性。研究成果能为提高CMOS图像传感器的图像质量和动态范围提供理论指导和新思路。

本报告是对国家自然科学基金项目“CIS像素RTS噪声的时变物理统计建模与多维非均匀抑制方法研究”的总结。本项目按照研究计划，完成预期研究内容。.在噪声建模方面：建立了基于二项分布的4T像素源跟随管RTS噪声统计模型，并得出相关双采样电路输出噪声的标准差。实验结果表明，在氧化物缺陷浓度较低时，短沟道器件的RTS噪声拖尾比长沟道器件长，揭示了CIS像素RTS噪声物理统计分布规律。.在传感器光谱灵敏度函数方面：设计了具有新型保护环和基于SOI技术的堆叠结光电二极管结构，分析了这两种结构的光谱灵敏度。实验结果表明，所设计的两种结构均能有效防止边沿过早击穿，提高器件的光子探测效率，为提高传感器图像质量提供理论基础。.在随机噪声抑制方面：设计了针对simplified signal-dependent噪声模型的自适应随机信号依赖噪声消除方法，和针对Poisson-Gaussian噪声模型的随机信号依赖噪声参数的估计方法。实验结果表明，所设计的噪声去除方法能获得较高PSNR值和较低S-CIELAB ΔE*值，所设计的噪声参数估计方法能获得较低绝对平均误差，为研究CIS噪声抑制提供理论依据。