面向半导体生产线的工件聚类方法研究

发布时间：2018-05-21 01:43

本文选题：半导体制造 + 数据挖掘FCM算法　；参考：《北京化工大学》2015年硕士论文

【摘要】：21世纪以来国家大力扶持半导体制造业,在科学技术不断发展的潮流中,半导体制造设备不断更新换代,产品的需求量不断增加,产品质量要求不断上升。半导体制造系统成为世界各国学者、科学家研究的热点。半导体制造系统具有多重入、工序工艺复杂、约束条件繁多、不确定性和多目标等特点。在半导体生产线中,特征指标有着重要的作用,他们表示工件的属性,以及半导体生产线性能好坏。如何在这些特征指标中挖掘出有效的信息,对半导体生产线进行改进优化是越来越多学者,科学家研究的课题。在众多挖掘算法中,聚类算法是较普遍的一种。这种算法基于工件的特征指标,针对工件有效的分类成调度实例。本文以半导体生产线为背景,重点研究了模糊聚类算法理论以及其面向半导体制造过程数据上的应用。本文主要研究内容如下：(1)面向半导体制造过程数据背景下,对模糊C均值(FCM)算法进行了理论研究。FCM算法是本篇论文的理论基础,通过仿真实验,发现该方法有着较好的聚类效果,并将FCM聚类算法用于半导体生产线的工件聚类,利用工件的特征指标,进行聚类分析。(2)通过研究FCM算法,发现其初始聚类中心是随机确定的,论文将减法模糊聚类(SUB-FCM)算法应用到半导体生产线背景下。通过仿真实验,证明SUB-FCM方法准确度和速度都优于FCM算法,在半导体生产线工件聚类上得到很好应用。(3)通过研究FCM算法以及半导体生产线的特点,发现半导体生产线中,动态,不确定性情况较多,从而导致半导体生产线工件的特征指标存在一些结构不一致的异常点,这种异常点对普通FCM聚类有干扰。在半导体制造过程数据背景下将二型模糊C均值(Type-2FCM)算法应用在半导体工件聚类分析中,这种算法对结构不一致的异常点有着较强的抗干扰能力。通过仿真实验以及工件特征指标聚类分析得到较好的效果。
[Abstract]:Since the 21st century, the state has vigorously supported the semiconductor manufacturing industry. In the trend of the continuous development of science and technology, the semiconductor manufacturing equipment has been continuously updated, the demand for products has been increasing, and the requirements of product quality have been rising. Semiconductor manufacturing system has become a hot spot for scholars and scientists all over the world. Semiconductor manufacturing system has the characteristics of multiple re-entry, complex process, various constraints, uncertainty and multi-objective. In semiconductor production line, characteristic index plays an important role, they express the properties of the workpiece and the performance of the semiconductor production line. How to find out the effective information in these characteristic indexes and how to improve and optimize the semiconductor production line are more and more scholars and scientists studying the subject. Among many mining algorithms, clustering algorithm is a common one. This algorithm is based on the characteristic index of the job and classifies the job into scheduling instance effectively. In this paper, the theory of fuzzy clustering algorithm and its application to semiconductor manufacturing process data are studied in the background of semiconductor production line. The main contents of this paper are as follows: (1) in the background of semiconductor manufacturing process data, the fuzzy C-means FCM algorithm is studied theoretically. FCM algorithm is the theoretical basis of this paper. It is found that this method has a good clustering effect, and the FCM clustering algorithm is applied to the job clustering of semiconductor production line. By using the characteristic index of the workpiece, the clustering analysis is carried out. (2) by studying the FCM algorithm, it is found that the initial clustering center is randomly determined. In this paper, subtraction fuzzy clustering algorithm is applied to semiconductor production line. The simulation results show that the accuracy and speed of SUB-FCM method is better than that of FCM algorithm, and it is well applied in the clustering of semiconductor production line. By studying the FCM algorithm and the characteristics of semiconductor production line, we find out the dynamic state in semiconductor production line. There are many uncertainties, which leads to the existence of some abnormal points in the characteristic index of the semiconductor production line, which interfere with the ordinary FCM clustering. In the background of semiconductor manufacturing process data, the type 2 fuzzy C-means Type-2FCMalgorithm is applied to the clustering analysis of semiconductor workpieces. This algorithm has strong anti-interference ability to the abnormal points with inconsistent structure. A good result is obtained by simulation experiment and clustering analysis of feature index of workpiece.
【学位授予单位】：北京化工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN305

【参考文献】