厚膜集成电路的计算机辅助检验
|
摘要
随着信息技术的飞速发展，集成电路的发展已成为我国十五规划中，主要发展的
一个产业。而厚膜集成电路作为集成电路产业的一个分枝，已广泛应用于通信、航空、
航天、计算机、工业控制等许多领域，每年以30％左右的速度增长，现已开发许多用
于单片集成电路和印刷电路板电路设计与检测的软件及系统。但应用于厚膜集成电路
检验的系统为数很少，国内厚膜集成电路生产厂过程检验大部分以人工检验为主。因
此，研究厚膜集成电路过程检验的自动检测具有非常重要的实用意义。
随着图像处理技术的不断完善和创新，计算机技术的迅猛发展。图像处理广泛应
用于军事、生物、医学工程、工业等许多方面。由于应用的范围不同，图像处理的方
式也不尽相同。
图像处理的前题，首先要有图像采集系统，图像获取有多种方式。如光学成像、
射线成像、超声成像、红外成像、微波成像等。本论文选用的对象为SH723E产品，
用ParagonⅡ SP Series Scanners，扫描仪系统为Ulead Iphoto Plus VⅡ Release Notes
995 Ulead System. Inc.的MFC-600S平板式扫描仪，对SH723E产品进行扫描成像，
选该产品作为本文的处理对象在于该产品是四川仪表六厂为深圳华为技术有限公司生
产的，用于程控交换机的用户线接口部分。属四川仪表六厂的主导产品，该产品月产
60多万块，年产值达三千多万元，因此对该产品实现自动检测有其重要的实用意义。
获取的图像因其背景和目标不同而需作相应处理，图像一般都含有噪声，要获得
理想的处理结果，对图像一般需做平滑、增强等运算。平滑既可在空域中进行，也可
在频率域中进行，可根据图像本身的特性而选定。在空域中可用空域平均法。在频率
域中介绍了几种常用的理想低通滤器如ILPF、BLPF、ELPF、TLPF、同时介绍了中值
滤波器＋图像增强是不考虑图像降质，只将图像中感兴趣的特征有选择性地突出，而
衰减其不需要的特征。图像增强技术也可分为频域法，空域法以及兼有空域法与频域
法的伪彩色处理，空域法增强包括灰度变换，直方图变换。
图像分割是图像处理的重要组成部分，因为一幅图像中背景和目标上的特性不同，
将图像中的各个不同区域的边缘检测出来，可达到图像分割的目的，边缘能够勾画区
域形状－并能传递大部分图像信息等优点，因此边缘检测可以看作是处理许多复杂问
题的关键，边缘并无统一的定义，可以认为是图像上的点即图像点，在该点图像强度
函数的梯度达到它的局部最大值。梯度的计算可以通过微分得到，可以用微分算子，
如Laplacian．模板卷积等。边缘检测也可分为空间域、频率域。本文介绍了几个比较重
要而且最具代表性的边缘检测器，如C侧算法、Mars边缘检测方法、Facet边缘检测
法。由于检测离不开定位，定位方法中介绍了H。U吵变换，该变换用于检测中定位，
在厚膜集成电路自动检测中，定位是通过定位标识并利用H。U吵变换对定位标识的定
位，最终实现电路的定位。
图像分割中，由于我们所处理的图像不仅仅是灰度图像，大部分要处理的图像是
彩色图像，因此在图像分割时，要根据具体应用，充分利用彩色分割，可以得到令人
满意的处理结果，本文充分利用了RoB三色分割。
图像分割需要选用一定的分割算法和阈值，本论文以SH723E产品为处理对象，
详细论述了WalliS阈值分割法、类间最大嫡阈值分割法、均值分割法、CWy分割法，
并对这几种分割法所取得的结果进行了分析，论述了这几种方法在厚膜集成电路自动
检测中的应用范围及用途。
文章结构如下：
第一章 厚膜集成电路的工艺流程和中间检验规范
第二章 厚膜集成电路产品图像的获取
第三章 厚膜集成电路产品图像的处理
第四章SH723E产品图像的处理
第五章 厚膜集戊电路的缺陷检验
第六章 结果分析及总结
With the rapid development of Information technology, integrated circuit industry has
been designated as a pillar industry in our country’s the tenth five-year economic plan. As a
branch of the integrated circuit industries, Hybrid integrated circuit has been widely used in
many areas, such as communication, aviation, aerospace, computer and industry processing
control. The product volume of hybrid integrated circuit increases 30% each year. Some
software and systems had been developed for the design and measurement of Monolithic
integrated circuit and PCB. However, few software systems developed for Hybrid integrated
circuit’s detection. Visual inspection of hybrid integrated circuit still processed by human
eyes. Therefore, studying automatic hybrid integrated circuit inspection process is of great
practical significance.
Since image processing technology and computer technology are continual
development and innovation, image processing was extensively used in military, biological,
medicine, induitry, etc. Different application fields use different image processing methods.
There are many ways for image acquisition: Optical, X-Ray, Ultrasonic, Infrared,
Microwave, eta. In order to processing image, an image acquisition system is needed. In this
research, a Paragon lISP Series Scanner with Ulead Iphoto Plus V II Release Notes System
were used as image acquisition equipment. Sichuan Instrument No.6 Factory’s SH723E
product is the test sample of this research. . SH723E is interface circuit for a progranixnable
digital switching system of Huawei Company in Shenzhen. The product has been
manufactured more than 600 thousand pieces each month; annual gross monetary value
reached 3 millions Yuan. Therefore, if the product automatic inspection can be realized,
potential economic benefit can be very promising.
On account of the difference between the image background and object, image mixed
with noise frequency needs to be smoothed or processed by other enhancement means.
Smoothing not only can be implemented in spatial domain, but also can be done in
frequency domain. According to different image characteristics, either space domain
processing or frequency domain processing can be chosen. In spatial domain, space average
method is commonly used. In this thesis, several commonly used frequency domain method
are discussed in depth, including ILPF, BLPF, ELPF, TLPF, and Median filters. The purpose
of image enhancement is to emphasis the interested feature of certain image and suppresses
other uninterested ones, so the original image restoration is not the first priority. Image
enhancement technology can be divided according to processing domains: Frequ~ncy
domain method, Pseudo color method and Space domain method. Space domain
enham~ement method includes gray scae transform and histogram modification etc.
Image segmentation is an important image-processing step for inspection. Detecting
different regions?edge can be used to segment an image. Edge detection approach is key for
dealing with complex problems. Since edge can represent a region’s shape, the edge map
delivers much information of an image. Edge has no unified: definition, so edge point ca~ be
defined as an image point, where gradient of image inteUsity function reaches its lpcal
maximum. Gradient value can be computed by approxh iation of derivatives of image
intensity function. Also, differential operators, such as Laplaeian mask, etc. can be u~ed.
Edge detection can be divided into two domains: space domain and frequency domain.
Several important and typical edge detectors -Canny algorithm, Marr algori
一个产业。而厚膜集成电路作为集成电路产业的一个分枝，已广泛应用于通信、航空、
航天、计算机、工业控制等许多领域，每年以30％左右的速度增长，现已开发许多用
于单片集成电路和印刷电路板电路设计与检测的软件及系统。但应用于厚膜集成电路
检验的系统为数很少，国内厚膜集成电路生产厂过程检验大部分以人工检验为主。因
此，研究厚膜集成电路过程检验的自动检测具有非常重要的实用意义。
随着图像处理技术的不断完善和创新，计算机技术的迅猛发展。图像处理广泛应
用于军事、生物、医学工程、工业等许多方面。由于应用的范围不同，图像处理的方
式也不尽相同。
图像处理的前题，首先要有图像采集系统，图像获取有多种方式。如光学成像、
射线成像、超声成像、红外成像、微波成像等。本论文选用的对象为SH723E产品，
用ParagonⅡ SP Series Scanners，扫描仪系统为Ulead Iphoto Plus VⅡ Release Notes
995 Ulead System. Inc.的MFC-600S平板式扫描仪，对SH723E产品进行扫描成像，
选该产品作为本文的处理对象在于该产品是四川仪表六厂为深圳华为技术有限公司生
产的，用于程控交换机的用户线接口部分。属四川仪表六厂的主导产品，该产品月产
60多万块，年产值达三千多万元，因此对该产品实现自动检测有其重要的实用意义。
获取的图像因其背景和目标不同而需作相应处理，图像一般都含有噪声，要获得
理想的处理结果，对图像一般需做平滑、增强等运算。平滑既可在空域中进行，也可
在频率域中进行，可根据图像本身的特性而选定。在空域中可用空域平均法。在频率
域中介绍了几种常用的理想低通滤器如ILPF、BLPF、ELPF、TLPF、同时介绍了中值
滤波器＋图像增强是不考虑图像降质，只将图像中感兴趣的特征有选择性地突出，而
衰减其不需要的特征。图像增强技术也可分为频域法，空域法以及兼有空域法与频域
法的伪彩色处理，空域法增强包括灰度变换，直方图变换。
图像分割是图像处理的重要组成部分，因为一幅图像中背景和目标上的特性不同，
将图像中的各个不同区域的边缘检测出来，可达到图像分割的目的，边缘能够勾画区
域形状－并能传递大部分图像信息等优点，因此边缘检测可以看作是处理许多复杂问
题的关键，边缘并无统一的定义，可以认为是图像上的点即图像点，在该点图像强度
函数的梯度达到它的局部最大值。梯度的计算可以通过微分得到，可以用微分算子，
如Laplacian．模板卷积等。边缘检测也可分为空间域、频率域。本文介绍了几个比较重
要而且最具代表性的边缘检测器，如C侧算法、Mars边缘检测方法、Facet边缘检测
法。由于检测离不开定位，定位方法中介绍了H。U吵变换，该变换用于检测中定位，
在厚膜集成电路自动检测中，定位是通过定位标识并利用H。U吵变换对定位标识的定
位，最终实现电路的定位。
图像分割中，由于我们所处理的图像不仅仅是灰度图像，大部分要处理的图像是
彩色图像，因此在图像分割时，要根据具体应用，充分利用彩色分割，可以得到令人
满意的处理结果，本文充分利用了RoB三色分割。
图像分割需要选用一定的分割算法和阈值，本论文以SH723E产品为处理对象，
详细论述了WalliS阈值分割法、类间最大嫡阈值分割法、均值分割法、CWy分割法，
并对这几种分割法所取得的结果进行了分析，论述了这几种方法在厚膜集成电路自动
检测中的应用范围及用途。
文章结构如下：
第一章 厚膜集成电路的工艺流程和中间检验规范
第二章 厚膜集成电路产品图像的获取
第三章 厚膜集成电路产品图像的处理
第四章SH723E产品图像的处理
第五章 厚膜集戊电路的缺陷检验
第六章 结果分析及总结
With the rapid development of Information technology, integrated circuit industry has
been designated as a pillar industry in our country’s the tenth five-year economic plan. As a
branch of the integrated circuit industries, Hybrid integrated circuit has been widely used in
many areas, such as communication, aviation, aerospace, computer and industry processing
control. The product volume of hybrid integrated circuit increases 30% each year. Some
software and systems had been developed for the design and measurement of Monolithic
integrated circuit and PCB. However, few software systems developed for Hybrid integrated
circuit’s detection. Visual inspection of hybrid integrated circuit still processed by human
eyes. Therefore, studying automatic hybrid integrated circuit inspection process is of great
practical significance.
Since image processing technology and computer technology are continual
development and innovation, image processing was extensively used in military, biological,
medicine, induitry, etc. Different application fields use different image processing methods.
There are many ways for image acquisition: Optical, X-Ray, Ultrasonic, Infrared,
Microwave, eta. In order to processing image, an image acquisition system is needed. In this
research, a Paragon lISP Series Scanner with Ulead Iphoto Plus V II Release Notes System
were used as image acquisition equipment. Sichuan Instrument No.6 Factory’s SH723E
product is the test sample of this research. . SH723E is interface circuit for a progranixnable
digital switching system of Huawei Company in Shenzhen. The product has been
manufactured more than 600 thousand pieces each month; annual gross monetary value
reached 3 millions Yuan. Therefore, if the product automatic inspection can be realized,
potential economic benefit can be very promising.
On account of the difference between the image background and object, image mixed
with noise frequency needs to be smoothed or processed by other enhancement means.
Smoothing not only can be implemented in spatial domain, but also can be done in
frequency domain. According to different image characteristics, either space domain
processing or frequency domain processing can be chosen. In spatial domain, space average
method is commonly used. In this thesis, several commonly used frequency domain method
are discussed in depth, including ILPF, BLPF, ELPF, TLPF, and Median filters. The purpose
of image enhancement is to emphasis the interested feature of certain image and suppresses
other uninterested ones, so the original image restoration is not the first priority. Image
enhancement technology can be divided according to processing domains: Frequ~ncy
domain method, Pseudo color method and Space domain method. Space domain
enham~ement method includes gray scae transform and histogram modification etc.
Image segmentation is an important image-processing step for inspection. Detecting
different regions?edge can be used to segment an image. Edge detection approach is key for
dealing with complex problems. Since edge can represent a region’s shape, the edge map
delivers much information of an image. Edge has no unified: definition, so edge point ca~ be
defined as an image point, where gradient of image inteUsity function reaches its lpcal
maximum. Gradient value can be computed by approxh iation of derivatives of image
intensity function. Also, differential operators, such as Laplaeian mask, etc. can be u~ed.
Edge detection can be divided into two domains: space domain and frequency domain.
Several important and typical edge detectors -Canny algorithm, Marr algori
引文
1． Pun T. Entropic thresholding , a new approach , CGIP, 1981, 210-239
2． Brink AP thresholding of digital images using of two deminationel Entropies, Patlem Recognitios. 1992． 25,803-808
3． Pratt W. Digital Image Processing. New York: Wiley. 1978．
4． Weszka J S. A survey of threshold selection techniques. Vision Graphics Image Process, 1978,259～265
5． Bergholm F. Edge focusing. IEEE Trans Pattern Anal Mach Intel,1987,9(9) : 726～741
6． Choi Y and Krishnapuram R. A robust approach to image enhancement based on fuzzy logic. IEEE Trans on Image Processing, 1997,6(6) .
7． [Canny 86] J.Canny A computational approach to edge detection. IEEE Trans.Pattern Anal. And Machine Intelligience, 8:679-698,1986
8． [Rice 45] S.O.Rice. Mathematical analysis radom noise. Bell Syst. Tech. J., 24:46～156,1945．
9． [Haralick and Zuniga 87] R.M. Haralick and 0． A. Zuniga. Integrated directional derivative gradient operator. IEEE Trans. Systems, Man and Cybernetics,17:508～517,1987
10． 实现图像边缘检测的改进广义模糊算子法 洪文松 陈武凡 中国图像图形学 报Vol4(A)。No．2 Feb．1999 143～146。
11． 一种基于图像分割的增强算法 孙全 叶秀清 顾伟康 中国图像图形学报 Vol4． No．5 May．1999． 409～413
12． 一种直接直方图形修正方法 余日泰 蔡吸礼 中国图像图形学报Vol．4(A)． No．6 June 1999 467～469
13． 一种基于灰度期望值的图像二值化算法 高永英 张利 吴国威 中国图像 图形学报Vol．4(A)．No．6 June 1999 524～528
14． 一种基于方向曲线的目标检测方法 李华 朱光喜 朱耀庭 中国图像图 形学报Vol．4(A)，No7． July．1999． 590～593
15． 图像轮廓特征提取新方法研究 吕铁英 彭嘉雄 中国图像图形学报Vol．4
(A)．No8． Aug.1999 655～658
16． PCB板检测中的定位研究 张利 高永英 汪浩 吴国威 中国图像图形学报 Vol．4(A)．No．8 Aug．1999 659～661
17,数字图像处理 东南大学出版社 夏正良编
18,一种新的指纹图像细化算法 冯星奎 李林艳 颜祖泉 中国图像图形学报 vol．4(A)．No．10 Oct．1999 834～838
19． 指纹图像合成分割法 冯星奎 颜祖泉 肖兴明 尹洪君 计算机应用研究No．1 2000年76～77
20． 基于图像分割的人眼边缘提取方法研究 林峰 刘政凯 俞能海 计算机应 用研究No．8 2000年100～102
2． Brink AP thresholding of digital images using of two deminationel Entropies, Patlem Recognitios. 1992． 25,803-808
3． Pratt W. Digital Image Processing. New York: Wiley. 1978．
4． Weszka J S. A survey of threshold selection techniques. Vision Graphics Image Process, 1978,259～265
5． Bergholm F. Edge focusing. IEEE Trans Pattern Anal Mach Intel,1987,9(9) : 726～741
6． Choi Y and Krishnapuram R. A robust approach to image enhancement based on fuzzy logic. IEEE Trans on Image Processing, 1997,6(6) .
7． [Canny 86] J.Canny A computational approach to edge detection. IEEE Trans.Pattern Anal. And Machine Intelligience, 8:679-698,1986
8． [Rice 45] S.O.Rice. Mathematical analysis radom noise. Bell Syst. Tech. J., 24:46～156,1945．
9． [Haralick and Zuniga 87] R.M. Haralick and 0． A. Zuniga. Integrated directional derivative gradient operator. IEEE Trans. Systems, Man and Cybernetics,17:508～517,1987
10． 实现图像边缘检测的改进广义模糊算子法 洪文松 陈武凡 中国图像图形学 报Vol4(A)。No．2 Feb．1999 143～146。
11． 一种基于图像分割的增强算法 孙全 叶秀清 顾伟康 中国图像图形学报 Vol4． No．5 May．1999． 409～413
12． 一种直接直方图形修正方法 余日泰 蔡吸礼 中国图像图形学报Vol．4(A)． No．6 June 1999 467～469
13． 一种基于灰度期望值的图像二值化算法 高永英 张利 吴国威 中国图像 图形学报Vol．4(A)．No．6 June 1999 524～528
14． 一种基于方向曲线的目标检测方法 李华 朱光喜 朱耀庭 中国图像图 形学报Vol．4(A)，No7． July．1999． 590～593
15． 图像轮廓特征提取新方法研究 吕铁英 彭嘉雄 中国图像图形学报Vol．4
(A)．No8． Aug.1999 655～658
16． PCB板检测中的定位研究 张利 高永英 汪浩 吴国威 中国图像图形学报 Vol．4(A)．No．8 Aug．1999 659～661
17,数字图像处理 东南大学出版社 夏正良编
18,一种新的指纹图像细化算法 冯星奎 李林艳 颜祖泉 中国图像图形学报 vol．4(A)．No．10 Oct．1999 834～838
19． 指纹图像合成分割法 冯星奎 颜祖泉 肖兴明 尹洪君 计算机应用研究No．1 2000年76～77
20． 基于图像分割的人眼边缘提取方法研究 林峰 刘政凯 俞能海 计算机应 用研究No．8 2000年100～102