摘要倒装芯片技术已经在我们的社会生活中起着不可缺少的作用，它是利用焊球 来将芯片与焊球相互连接，它的处理效率较高，也很便捷。但是，因为芯片和基 底的热膨胀系数差异很大，在加载温度下，受热或冷却时的变形也不一样，导致 焊点的应力不匹配，会出现变形、弯曲等封装缺陷。而且，倒装芯片的封装密度 增加了，从而使得焊球间的尺寸也变得越来越小，因为焊球的缺陷隐藏在芯片与 基底之间，检测出它们变得越来越困难，这说明改进缺陷的检测方法是非常有必 要的。72686

超声波检无损测技术已经普遍涉及我们的社会生活，它是当今微电子封装技 术中经常使用的检测方法。本文需要对倒装芯片的扫描图像处理，主要有四个工 作：倒装芯片的检测，将焊球分离出来，分析焊球的特点，并进行分类。但是由 于检测条件的限制，必须要在低成本的情况下，获得高效的处理结果。因此，我 们想出将图像进行超分辨率重构，本文运用了一种改进过的贝叶斯超分辨率方 法，即在贝叶斯框架下，基于水平和垂直方向上的图像像素一阶差分的 L1 范数 超分辨率重建图像及其参数预估，会得到一个高分辨率图像，再根据焊球的特征， 利用支持向量机将它们进行分类。

最终结果表明，我们提出的基于 L1 范数的贝叶斯超分辨率重构技术，与原 始的图像检测结果比较，具有一定的优势。

毕业论文关键词：倒装芯片 缺陷检测 图像超分辨率重建 支持向量机

Reconstruction and application of flip chip defect detection signal

Abstract Flip chip technology has in our social life plays a indispensable role, it is the use of solder balls to the chip and solder balls are connected to each other, However, because of the chip and the substrate thermal expansion coefficient difference, under the temperature load, heating or cooling the deformation is not the same, causing the solder joint stress mismatch will appear such as deformation and bending packaging defects。 And flip chip packaging density increased, so that the solder ball size has become more and more small, because of the defects of solder ball hidden between the chip and the substrate to detect them becomes more and more difficult, which indicates that the improved methods for defect detection is very necessary。

In this paper, there are four steps: the detection of welding ball, the processing and segmentation of the ball, the extraction of the characteristics of the ball and the

classification of the ball。 However, due to the limitations of the test conditions。

Therefore, we propose an image super-resolution reconstruction method, used in this paper is the L1 norm of the Bayesian super-resolution method based on, is under the Bayesian framework, based on image pixels in the vertical and horizontal first order variance of L1 prior resolution reconstruction of image and its parameter estimation will get a high resolution image, again according to the characteristics of the solder ball, using support vector machine (SVM) and sorting them out。

The final results show that our proposed Bayesian image super resolution technique based on L1 norm is effective in comparison with the original low resolution image detection results。

Key  Words:  flip chip defect detection image  super  resolution reconstruction support vector machine

目 录

摘要-Ⅰ

Abstract--Ⅱ

图清单-Ⅳ

第 1 章  绪论 1

1。1 研究背景及意义