基于主成分分析的多元分段模型预测集成电路晶圆良率的应用

发布时间：2018-12-15 06:43

【摘要】：随着中国政府推出《国家集成电路产业发展推进纲要》后,集成电路制造产业在我国蓬勃发展。在我国,集成电路制造多以代工形式为主,因此对于产业发展与生存的关键在于集成电路芯片的良率控制与提高。在集成电路良率预测模型的研究上,从上世纪60年代开始就有学者提出各种良率模型,早期的研究主要集中在寻找良率和线上缺陷的关系。随着工艺提升、设计难度增大,良率预测也希望能更精确。文中,我们提出良率损失的两种模式。第一类良率损失模式,线上缺陷导致的良率损失。第二类良率损失模式,芯片工艺设计缺陷导致的良率损失。我们以电性测试参数为自变量,利用统计模型来精确预测良率损失的情况。通过建立两类良率模式的多元分段函数,尝试用主成分分析法减少重复信息,用决策树方法找到合适分段点,并分别对分段部分建立Logistic回归模型。根据模型分析的结果,发现此良率预测模型对误差有很大的改善。而此多元分段模型,也解释了两种良率损失模型并存的现象。基于此方法,可以将统计模型的构建方式建立一套标准流程,并可以借用计算机程序将此标准自动化。便于一线工程师使用此工具,及时发现生产异常,减少损失。
[Abstract]:With the introduction of the National Integrated Circuit Industry Development Program by the Chinese government, the integrated circuit manufacturing industry has flourished in China. In China, the manufacturing of integrated circuits is mainly in the form of contract manufacturing, so the key to the development and survival of the industry lies in the control and improvement of the yield of integrated circuit chips. In the research of integrated circuit yield prediction model, some scholars have put forward a variety of yield models since the 1960s, and the early research mainly focused on finding the relationship between yield and on-line defects. As the process improves, the design becomes more difficult, and the yield prediction is expected to be more accurate. In this paper, we propose two models of yield loss. The first type of yield loss mode, the loss of yield caused by line defects. The second type of yield loss mode, chip process design defects caused by yield loss. We use the electrical test parameters as independent variables and use the statistical model to accurately predict the loss of yield. By establishing multivariate piecewise functions of two kinds of yield patterns, we try to reduce repeated information by principal component analysis (PCA), find suitable segmental points by using decision tree method, and establish Logistic regression models for segmented parts respectively. According to the results of model analysis, it is found that the yield prediction model improves the error greatly. This multivariate piecewise model also explains the coexistence of two yield loss models. Based on this method, a set of standard flow can be established for the construction of statistical model, and the standard can be automated by computer program. Easy to use this tool, timely detection of production anomalies, reduce losses.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：O212.1

【相似文献】