混合制程扰动状态估计与双层结构批间控制器设计

IC manufacturing is the cornerstone of the informational industry, which reflects the national high-tech manufacturing capability comprehensively. It is important to investigate the optimal control theories and methods of the semiconductor manufacturing process, which help the industrial upgrading and structural adjustment. Recently, a high-mix semiconductor manufacturing process (HMSMP) becomes a governing-mode in the automated assembly line, in which there are many products with different recipes, small batch and value-added. The modeling and control of the high-mix production mode impose a great challenge for process engineers. Taking numerous uncertainties into consideration, a collaborative Run-to-Run (R2R) control is developed in this work. Firstly, a minimum-variance unbiased (MVU) filter is designed for the HMSMP under the unknown disturbances, and the performance of MVU is also analyzed. Then, a novel R2R controller is presented to overcome measurement delay in the HMSMP, and the corresponding optimal parameters are also obtained by minimizing the TMSE (Total Mean Square Error). Thirdly, a collaborative R2R controller with double-layer structure is proposed to keep the quality consistency of products in HMSMP. Finally, a platform simulated to the HMSMP with multi-stages is developed. The potential results will provide theoretical and technical support for the HMSMP. Furthermore, some new R2R theories and methods can be expected to promote the development of Chinese IC manufacturing industry.

芯片制造是信息产业的基石，是国家高端制造能力的综合体现，深入研究半导体制造优化控制的理论与方法，对信息产业升级与结构调整均具有重要的意义。近年来，多品种、小批量、高附加值混合产品制程已经成为芯片制造的主要生产模式，对半导体制程控制提出了新的挑战。本项研究针对混合制程的特点与其不确定性，研究扰动状态估计与双层结构批间控制器协同设计方法。研究内容包括：未知扰动的最小方差无偏滤波器设计与性能分析、带测量时延的批间控制器构造与参数优化、双层结构优化的批间协同控制，以及多阶段混合制程协同仿真实验平台的研制。项目成果将为半导体混合加工过程提供理论与技术支撑，从而为我国集成电路制造产业的发展，提供批间控制新理论和新方法。

批间控制技术利用数据的正向补给和反馈来有效地抑制半导体制程的各种干扰，从而提高晶圆的良率，已成为当今半导体制造不可或缺的基石。本项目针对带有“随机干扰、模型失配、多产品、多阶段”特征的半导体制程，研究制程的扰动估计与其控制器设计，取得主要成果为：（1）基于子空间、高斯混合模型，以及贝叶斯模型等，提出新的虚拟测量模型辨识算法，并分析各辨识算法的收敛性、计算复杂度等相关性能；（2）提出基于扰动观测器的批间控制设计方法，并证明与文献中各种设计方案的等价关系，给出控制器的稳定域；（3）引入深度强化学习理论，将传统批间控制器的参数调整转化成马尔科夫决策过程，给出双层结构的批间控制器设计方案；（4）针对半导体混合制程的特点与不确定性，系统研究扰动状态估计与批间控制器协同设计策略，为半导体晶圆混合制程的控制提供理论与技术支撑。在本项目资助下，发表学术论文24篇（其中：SCI收录14篇，EI收录10篇；期刊论文15篇，会议论文9篇），出版学术专著1部，授权中国发明专利3件，受理中国发明专利2件，培养博士研究生2名，硕士研究生3名。本项目成果为半导体制程的扰动抑制提供了解决方案，具有较高的理论意义与工程应用价值，丰富了半导体制造系统的控制理论和方法体系。