CMOS图像传感器凸点良率随机模型

Statistic Model for Bump Yield of CMOS Image Sensor

倒装技术目前被应用在CMOS图像传感器当中,用以提高传感器感光层的填充系数。使用成熟的制备工艺,CMOS图像传感器中仍然有少量凸点失效。在高像素,小间距的CMOS图像传感器当中,凸点数目较大,无法逐一检测坏点,因此依靠菊链结构检测点阵。然而,菊链的检测结果与点阵良率之间的数量关系罕有涉及。文章根据菊链结构与特性,从0-1分布、二项分布等统计分布出发,建立了四种基于菊链结构的凸点良率随机模型,用以分析高密度、小间距的凸点阵列的凸点良率,并讨论了四种模型的关系以及适用范围。讨论表明,连续二项分布模型的约束条件最少;在凸点数较大的情况下,可采用0-1分布模型简化计算,并能得到与连续二项分布模型极为近似的结果。

To improve the fill factor of detective die,flip chip technology has been used in producing COMS image sensors.The ineffective bumps exist even in the mature process technology.In those CMOS image sensors with small pitch and numerous pixels,the ineffective bumps cannot be detected one by one as the number of the bumps is quite large.Therefore daisy chain by connecting the bumps is used to detect the bump array.However,few models were built to analyze the relation between daisy chain and bump yield.In this article,4 statistic models are built based on 0-1 and binomial distributions for those bump arrays with small pitch and large number of pixels.The relation among the models and the availability are discussed,suggesting that the binomial model was restrained least.For simplicity,0-1 distribution model is available and similar to binomial model when pixel number is quite large.