高度混合制程的批间控制与性能评估

With the development of semiconductor manufacturing industry, production resembles a high-mix automated assembly line in which many similar products with slightly different specifications are manufactured step-by-step, and a complicated physiochemical batch process is performed by a number of tools in each step. There are many challenges in the modeling and control of the high-mix production systems. In this project, several novel identification and control algorithms are presented under the run-to-run control framework. Firstly, based on identification of tool-effect, product-effect and their interactive effect, a novel quality model and non-thread state estimation method are developed. Then, a relationship between the discount factor and the manufacturing frequency of wafer is built to improve the quality of low frequency products. Thirdly, the stability and robustness of EWMA controller are derived to overcome the lag between the processing and measurement of wafer at a manufacturing step. Finally, a performance assessment and fault detection & classification are developed using the advance statistical theories. Several numeral simulations and industrial applications are provided to illustrate the importance of considering methods in the high-mix production system. In the end, a series of practiced and systematical algorithms of modeling, control and assessment are presented. Due to directly face the industrial demands, the proposed approaches demonstrate the strong innovation character and have a high potential in industrial applications.

随着半导体产业的发展，晶圆生产已从最初的单一模式发展到现在的多产品共线的高度混合生产形态，这对控制系统的设计与评估都提出了新的挑战。本项研究在批间控制的框架下，深入研究高度混合制程的机台效应、产品效应、及其交互效应，提出晶圆品质模型辨识与非线程的状态估计算法；在此基础上，给出EWMA控制器调节参数与产品数量之间的关系矩阵，以改善量少产品的质量；并研究在信息传输的随机时延条件下，批间控制器稳定性条件与鲁棒性设计方法；协同先进统计理论，发展混合制程性能评估与故障诊断方法；结合典型半导体制程，开展实验研究，形成一种面向实际需要的、系统化的半导体混合制程的模型辨识与批间控制器设计方法。由于直接面向实际工业生产需要解决的难点问题，本项研究具有极大的应用价值，且研究内容属于过程控制与半导体制造的交叉领域，亦具有很强的创新性。

随着半导体产业的发展，晶圆生产已从最初的单一模式发展到现在的多产品共线的高度混合生产形态，这对控制器的设计与评估都提出了新的挑战。而控制器设计的首要步骤即为系统建模。然而，半导体机台FDC(Fault Detection and Classification)的海量、共线性、不同采样速率数据，难以直接用于系统建模。为此，本项研究在批间控制的框架下，深入研究基于多率系统、关键变量筛选、多元统计分析、贝叶斯算法等的多种模型辨识算法；在此基础上，给出多线程状态估计、自适应控制器/带时延EMWA控制器设计、以及控制器性能评估等若干方法，有效的改善控制系统的性能；此外，根据半导体生产过程的控制要求，本项目还讨论了相关半导体的制造工艺；并结合典型半导体制程，开展了实验研究，形成一种面向实际需要的、系统化的半导体混合制程的模型辨识与批间控制器设计方法。.本项目已发表学术论文33篇（其中：SCI收录8篇，EI收录17篇，ISTP收录1篇；期刊论文19篇，会议论文14篇），中国发明专利12件（其中：授权中国发明专利9件，受理中国发明专利3件），培养博士研究生2名，硕士研究生6名。