• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 图形与图像 • 上一篇    下一篇

芯片表面印刷符号结构缺陷的质量评估方法研究

饶永明1,吴帅1,2,闫峰2,罗月童1   

  1. (1.合肥工业大学计算机与信息学院,安徽 合肥 230601;2.合肥图迅电子科技有限公司,安徽 合肥 230000)
  • 收稿日期:2018-11-14 修回日期:2019-01-25 出版日期:2019-09-25 发布日期:2019-09-25
  • 基金资助:

    国家重点研发项目(2017YFB1402200);安徽省科技强警计划项目(1604d0802009);浙江大学 CAD&CG 国家重点实验室开放课题(A1814); 中央高校基本科研业务费专项资金(JZ2017HGBH0915);安徽省高等学校省级质量工程项目(2017jyxm0045)

A quality evaluation method for structural

RAO Yong-ming1,WU Shuai1,2,YAN Feng2,LUO Yue-tong1   

  1. (1.School of Computer Science and Information Engineering,Hefei University of Technology,Hefei 230601;
    2.Hefei ImageEx Electronics Technology Company,Hefei 230000,China)
  • Received:2018-11-14 Revised:2019-01-25 Online:2019-09-25 Published:2019-09-25

摘要:

因为芯片表面符号用于标识芯片名称、性能和功能信息,所以芯片表面符号质量检测是芯片生产的重要环节。目前,基于机器视觉的芯片表面符号质量评估被广泛使用,其中多数是基于逐像素对比的评估方法,但它面临2个主要问题:(1)符号的少量形变会导致质量评估结果变差;(2)没有考虑符号的缺陷特征。因此,
提出面向芯片表面符号的结构缺陷的评估方法。首先采用薄板样条插值将待评估符号和参考符号对齐,消除细小形变的影响,同时提出形变量公式,用于判定形变较大的待评估符号;接着通过提出缺陷的检测方法和缺陷簇、关键位置等概念,得到质量评估的2个主要影响因子:缺陷的本身特征和缺缺陷所在位置是否关键;最后基于上述特征给出合理的打分策略。本质量评估方法是一个客观评价方法,它的优势是:(1)在只需要1张参考图像的前提下,对存在形变的评估符号进行质量评估具有较好的鲁棒性;(2)本方法聚焦于图像符号的结构质量评估,而非图像的本身质量评估,使得评估结果在具备客观性的同时,也符合人对符号内容的实际感觉。采用实际生产线数据进行实验,实验结果表明了本文方法的有效性。

 

关键词: 质量评估, 芯片, 印刷符号, 结构缺陷, 薄板样条插值

Abstract:

Since chip surface symbols are used to identify chip names, performance, and functional information, their quality evaluation is an important part of chip production. Currently chip surface symbol quality evaluation methods based on machine vision are widely used, and most of them are  evaluation methods based on pixel-by-pixel comparison, facing two main problems: (1) a small amount of deformation of the symbols leads to poor quality evaluation results; (2) the internal defect characteristics of the symbols are not considered. We propose an evaluation method for structural defects of printed symbols on chip surface. Firstly, the thin-plate spline interpolation is used to align the symbols to be evaluated with the reference symbols to eliminate the influence of small deformation. At the same time, we propose a deformation formula to determine the larger deformation symbols to be evaluated. Secondly, we propose a defect detection method, define the concept of defect cluster and key position, and obtain the two main influencing factors of quality assessment: the inherent characteristics of defects and whether the location of defects is critical. Finally, we define a reasonable scoring strategy based on the above characteristics. This quality evaluation method is an objective evaluation method, and its advantages are: (1) on the premise of only one reference image, it has good robustness when evaluating the quality of images with deformation; (2) The method focuses on the structure quality evaluation of image symbols rather than that of the image itself, so the evaluation results are objective and conform to human's actual feeling for symbol contents. The experimental data of an actual production line demonstrates the validity of the proposed method.
 

Key words: quality evaluation, chip, printed symbol, structural defect, thin plate spline interpolation