面向半导体制造过程动态调度的关键参数预测模型研究
发布时间：2018-07-31 19:21
【摘要】：在半导体市场需求快速变化和行业竞争白热化的背景之下,缩短产品的加工周期、提高设备利用率和产品质量,是企业保持盈利和较强市场竞争力的重要手段。而要改善半导体生产线的性能,提高生产效益,需对半导体制造过程动态调度的关键性能指标进行预测研究,通过分析性能指标的影响因子,找到可调控的影响参数来实现优化目标。本课题主要针对加工周期、设备利用率以及成品率这三个关键性能指标进行研究,通过建立多性能指标同步预测模型来实现对加工周期和设备利用率的同步预测及分析,利用半导体生产线获取的晶圆缺陷数据来完成晶圆缺陷聚集特性分析和对成品率的预测研究。本文围绕半导体制造过程动态调度的关键参数预测模型开展了如下的研究：1、针对半导体生产线动态环境下的加工周期和设备利用率的预测问题,研究一种基于贝叶斯神经网络的多性能指标同步预测模型,并构建了一种闭环修正模型结构。此外,利用贝叶斯神经网络可根据输入对输出的重要性来调整网络权值的特性,设计一种权值分析法,来识别加工周期和设备利用率的关键影响因子。2、针对晶圆缺陷问题,研究一种缺陷数据驱动的半导体成品率预测方法,利用基于密度的聚类方法对晶圆缺陷聚集特性进行分析,并给出一种模糊支持向量机方法来构建成品率的预测模型。将本文提出的两种预测模型应用到半导体制造过程关键性能指标的预测中,通过仿真结果和分析表明,这两种预测模型能够很好的实现对加工周期、设备利用率和成品率的预测和分析,为解决半导体制造过程动态调度问题提供了指导,为实现半导体生产线多目标优化奠定了基础。
[Abstract]:Under the background of rapid change of semiconductor market demand and intense competition in the industry, shortening the processing cycle of products, improving the utilization rate of equipment and product quality is an important means for enterprises to maintain profitability and strong market competitiveness. In order to improve the performance and benefit of semiconductor production line, it is necessary to predict and study the key performance indexes of dynamic scheduling of semiconductor manufacturing process. Find adjustable impact parameters to achieve optimization goals. This paper mainly focuses on three key performance indexes, such as processing cycle, equipment utilization ratio and finished product rate, and realizes synchronous prediction and analysis of machining cycle and equipment utilization rate by establishing synchronous prediction model of multi-performance index. The wafer defect data obtained from semiconductor production line are used to analyze the aggregation characteristics of wafer defects and to predict the yield of wafer defects. In this paper, the key parameter prediction model of dynamic scheduling of semiconductor manufacturing process is studied as follows: 1. Aiming at the prediction of processing cycle and equipment utilization under the dynamic environment of semiconductor production line, A synchronization prediction model with multiple performance indexes based on Bayesian neural network is studied, and a closed loop modified model structure is constructed. In addition, according to the importance of input to output, Bayesian neural network is used to adjust the characteristics of network weights, and a weight analysis method is designed to identify the key influencing factors of processing cycle and equipment utilization. In this paper, a defect data-driven method for predicting the yield of semiconductor products is studied. The density based clustering method is used to analyze the aggregation characteristics of wafer defects, and a fuzzy support vector machine (FSVM) method is proposed to build a prediction model of the yield. The two prediction models proposed in this paper are applied to the prediction of the key performance indexes in semiconductor manufacturing process. The simulation results and analysis show that the two prediction models can realize the processing cycle well. The prediction and analysis of equipment utilization ratio and finished product rate provide guidance for solving the dynamic scheduling problem of semiconductor manufacturing process and lay a foundation for multi-objective optimization of semiconductor production line.
【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
本文编号：2156582
[Abstract]:Under the background of rapid change of semiconductor market demand and intense competition in the industry, shortening the processing cycle of products, improving the utilization rate of equipment and product quality is an important means for enterprises to maintain profitability and strong market competitiveness. In order to improve the performance and benefit of semiconductor production line, it is necessary to predict and study the key performance indexes of dynamic scheduling of semiconductor manufacturing process. Find adjustable impact parameters to achieve optimization goals. This paper mainly focuses on three key performance indexes, such as processing cycle, equipment utilization ratio and finished product rate, and realizes synchronous prediction and analysis of machining cycle and equipment utilization rate by establishing synchronous prediction model of multi-performance index. The wafer defect data obtained from semiconductor production line are used to analyze the aggregation characteristics of wafer defects and to predict the yield of wafer defects. In this paper, the key parameter prediction model of dynamic scheduling of semiconductor manufacturing process is studied as follows: 1. Aiming at the prediction of processing cycle and equipment utilization under the dynamic environment of semiconductor production line, A synchronization prediction model with multiple performance indexes based on Bayesian neural network is studied, and a closed loop modified model structure is constructed. In addition, according to the importance of input to output, Bayesian neural network is used to adjust the characteristics of network weights, and a weight analysis method is designed to identify the key influencing factors of processing cycle and equipment utilization. In this paper, a defect data-driven method for predicting the yield of semiconductor products is studied. The density based clustering method is used to analyze the aggregation characteristics of wafer defects, and a fuzzy support vector machine (FSVM) method is proposed to build a prediction model of the yield. The two prediction models proposed in this paper are applied to the prediction of the key performance indexes in semiconductor manufacturing process. The simulation results and analysis show that the two prediction models can realize the processing cycle well. The prediction and analysis of equipment utilization ratio and finished product rate provide guidance for solving the dynamic scheduling problem of semiconductor manufacturing process and lay a foundation for multi-objective optimization of semiconductor production line.
【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】
相关期刊论文 前1条
1 曹政才;赵会丹;吴启迪;;基于自适应神经模糊推理系统的半导体生产线故障预测及维护调度[J];计算机集成制造系统;2010年10期
相关硕士学位论文 前1条
1 朱丹丹;集成电路设计中针对随机缺陷的成品率研究[D];大连理工大学;2011年
，本文编号：2156582
本文链接：https://www.wllwen.com/kejilunwen/dianzigongchenglunwen/2156582.html
