项目名称: 基于2-D空间离散数据的质量与产出的预测方法研究
项目编号: No.71471096
项目类型: 面上项目
立项/批准年度: 2015
项目学科: 管理科学
项目作者: 王凯波
作者单位: 清华大学
项目金额: 60万元
中文摘要: 准确的质量和产出的预测对制定生产计划、优化配置生产资源、节约生产成本和提高管理效率具有重要的现实意义。在半导体、太阳能和碳纳米管等高技术生产领域,产品质量是由具有2-D结构的空间离散数据所刻画。现有的预测方法和模型通常针对不具有空间结构的连续数据所设计,无法直接或准确的应用到具有2-D空间离散数据的生产场景。 本项目研究基于2-D空间离散数据进行产品质量和过程产出的预测问题。考虑到2-D空间离散数据随机分布但具有空间聚集性的特点,本研究提出建立基于高斯过程的双层模型来刻画2-D缺陷的波动规律和空间相关性,修改广义线性模型将2-D空间离散输出与过程参数和2-D空间连续输入相关联,利用部分采样信息提高2-D空间离散质量输出的预测精度。本研究对基于生产系统丰富数据进行科学定量管理具有重要意义,将为基于2-D空间离散数据进行质量分析和预测提供完整的理论模型,并为实际生产计划提供有效的预测方法。
中文关键词: 统计质量控制;产出预测;空间相关性;半导体制造
英文摘要: In real production management practice, precise prediction of product quality and process yield is key to efficient production planning, resource optimization, cost reduction and efficiency improvement. In high-tech manufacturing systems for semiconductor, solar cell and carbon- nano-tube production, product quality is characterized by two-dimensional (2-D) spatial discrete data. Current prediction methods designed for non-spatial continuous variables cannot be directly or accurately implemented in such systems. This proposal focuses on the prediction of product quality and process yield using 2-D spatial discrete datasets. Considering the randomness and clustering effect of 2-D spatial data, we propose to build a two-layer Gaussian Process model for characterizing the variation patterns and spatial correlations of 2-D spatial defects, modify the generalized linear model to incorporate process variables and 2-D spatial continuous inputs for prediction, and use partially measured 2-D spatial discrete data to improve prediction accuracy. This research demonstrates the successful use of rich production data for scientific and quantitative management. The output of this research will contribute to the academia with systematic models for quality analysis and yield prediction using 2-D spatial data, and to the practitioners with effective prediction methods for production planning.
英文关键词: Statistical Quality Control;Yield Prediction;Spatial Correlation;Semiconductor Manufacturing