【作者】 伍冯洁；

【导师】 吴黎明；

【作者基本信息】 广东工业大学 ， 测试计量技术及仪器， 2006， 硕士

【摘要】 随着半导体芯片工艺的极小化,已经达到深亚微米级水平,特征图形尺寸不断减小,晶圆面积持续增加,导致缺陷密度越来越低,真假缺陷更难分辨。进行晶片图像采集时,从理论上说,获取单张芯片图像不是问题,但深亚微米级IC晶片是超高精密产品且芯片集成度更高,为了获得高分辨率的图像,图像采集系统需要工作在高倍放大倍数下,视野相对较小,采集的图像张数以平方速度增加,显然,海量图像数据的存在制约着晶片质量检测的效率。 本文在广东省2004年科技计划项目“基于数字图像处理的IC晶片显微自动检测系统”阶段性成果的基础上,针对深亚微米级IC晶片质量检测中存在的巨大数据量与检测实时性的矛盾,综合应用数字图像处理技术、自动光学检测技术、数据库技术、模式匹配及模式识别理论等,主要从图像检测策略、自动分析算法、特征参数管理等几个关键问题展开研究。 本文分别从算法理论和实际应用的角度,对深亚微米级IC晶片图像检测及自动分析的相关技术难点进行深入分析,主要研究内容如下: 图像检测策略的研究与实验验证,根据IC晶片的先验知识,采用关键区域检测法及先粗后精检测法,实现非遍历检测;同时针对不同的检测目标,采用相应的图像检测算法,实现不同缺陷的识别,对图形规整或有规律可寻的图像采用模式匹配的检测方法,反之,则引入人机交互的检测方法。 采用投影定理与基于像素特征的检测方法得到投影变换的像素分布,通过检测像素点的分布特性实现冗余物缺陷的检测;同时采用Hough变换及基于骨架特征的检测方法,在骨架提取与优化的基础上,通过骨架跟踪及Hough直线检测的方法实现丢失物缺陷的识别,并确定芯片的失效形式。 提出了缺陷自动识别的方法,通过基于区域特征的图像搜索方法实现目标区域的定位,根据区域特征选取合适的参考模板,采用SSDA加速匹配算法快速实现区域的搜索,然后对目标区域进行图像处理与分析、识别;并以压焊点质量检测为例验证了该方法的可行性。 提出三层结构的数据库构建模型,提高了图像处理与分析、检测结果入库及数更多还原

【Abstract】 As semiconductor technics is smaller and smaller, having arrived to the deep-submicrometer level. Wafer’s CD is decreasing while its area is increasing that leads to defect density is lower and lower as well as hard to distinguish true or false defect. From the theoretical angle, it is not difficult to catch a single-chip image when wafer image is gathered. Because deep-submicrometer wafer is the more highly precise product and chip density is higher, the image gathering system must work in the highly amplified multiples in order to catch the high-resolution image, but field of vision is relatively small while images gathered increase with square speed. Obviously, the large image data restrict the efficiency of wafer quality detection.The paper is on the basis of the science and technology item of Guangdong province in 2004: Automatic inspection system of IC wafer micrograph based on digital image process. It aims at the contradiction between the large image data and the real-time requirement of detection. And it synthetically uses technology of digital image processing, AOI technology, database technology, theory of pattern match and pattern recognition etc., mainly studying in some key problems such as strategy of image detection, automatically analytical algorithm and management of characteristic parameter etc.The paper deep analyzes the relative technology difficulties of image detection and automatic analysis for deep-submicrometer wafer from algorithm theory and practical application, its main contents as follows:Studying and experimenting the strategies for image detection. It uses the key region and the coarse to precise detection method to detect wafer not all over according to the experience and knowledge. And it recognizes different defects by relevant algorithm of image detection aiming to different detection objects. As for the regular images, it uses the detection method of pattern match while using the detection method of human and PC for the non-regular ones.Using the projection theorem and detection method based on pixel character to get the pixel distribution of projection transfer, and successfully detect the redundancymaterial defect by detecting the distributed character of pixel. And it uses the detection method based on Hough and skeleton to recognize the dropped material defect and ascertain the invalidate forms of chip after skeleton extraction.Coming up the automatic recognition method for defects.lt orients the object region by method of image search based on the region characteristic and uses SSDA to successfully search the object region by choosing the suitable matching template, on the basis of these, processing and analyzing the object image and recognizing it.Finally, taking detection of welding quality as an example to verify the feasibility of the automatic recognition method.Using the model of three-layer configuration to construct system’s database.And it improves the compatibility of image processing and analysis, restore of detection result, data output etc. Finding out the interesting and useful information by technology of data extraction based on description, as well as outputting analysis reports of detection results.更多还原