性能驱动的半导体生产线闭环优化动态调度方法研究

The semiconductor wafer fabrication facility is a typical complex production system. With the rapid development of semiconductor technology, its scheduling faces new challenges, such as the serious uncertainties in its production environment and complexities of its production process and constraints. As a result, efficient dynamic scheduling methods are demanded to enable rapid response to real-time running environments, meet the requirements of complex constraints and achieve anticipated performance issues. This project abstracts the mentioned requirements as a scientific problem of performance-driven closed-loop dynamic scheduling. Taking swarm intelligence, data mining and system simulation as the main tools, it puts forward a research framework of performance-driven closed-loop dynamic scheduling, consisting of generation of samples, prediction models of performance issues and adaptive dynamic scheduling methods. Firstly, break through related key technologies, such as complex data pretreating, intelligent data analysis, data-based modeling and optimization, compound priority-based scheduling, etc.; Secondly, build state-performance predictive models and (performance, state)-parameter relational models, and propose an adaptive performance-driven closed-loop dynamic scheduling method, coupling constraint satisfaction, flexible preventative maintenance with dynamic dispatching. Finally, develop a semiconductor wafer fabrication facility simulation environment integrating simulating, scenario analysis and scheduling decision support to offer samples, validate and verify the proposed methods. These research achievements enrich and advance the scheduling theory of complex production systems, provide new ideas to solve the scheduling problems of semiconductor wafer fabrication facilities, and provide a scientific basis to invent new optimal ways for the scheduling of complex production systems.

随着半导体工艺的快速发展，其生产环境的不确定性与生产流程及工艺约束的复杂度不断提高，调度面临着新的挑战，迫切需要能够实时响应动态运作环境、满足工艺约束、获得期望运作性能的动态调度方法。本项目将此需求凝炼为性能驱动的闭环优化动态调度科学问题，综合运用群体智能、数据挖掘、系统仿真等方法技术，建立学习样本生成、性能建模、自适应动态调度的求解流程，确立性能驱动的闭环优化动态调度的实现框架；研究突破复杂数据预处理、智能数据分析、基于数据的建模与优化、复合优先级调度等关键技术；建立状态-性能预测模型、(性能,状态)-调度算法参数关系模型，提出耦合工艺约束满足、柔性维护与动态派工的自适应闭环优化动态调度方法，构建联结模拟、情景分析以及决策支持等环节的模拟实验环境。本项目研究丰富和发展复杂生产系统调度理论，提出解决半导体生产线调度难题的新思路与新方法，为发现复杂生产系统调度新的优化手段提供科学依据。

随着半导体工艺的快速发展，其生产环境的不确定性与生产流程及工艺约束的复杂度不断提高，调度面临着新的挑战，迫切需要能够实时响应动态运作环境、满足工艺约束、获得期望运作性能的动态调度方法。本项目将此需求凝炼为性能驱动的闭环优化动态调度科学问题。首先，聚焦现代半导体生产线调度特点及半导体制造商实际需求，确立性能驱动的半导体生产线闭环优化动态调度实现框架；其次，提出耦合工艺约束满足、柔性维护与动态派工的动态调度规则；再次，基于多层次信息融合的数据分析框架，建立状态—性能预测模型；然后，综合运用深度学习、集成学习等方法技术，建立(性能,状态)—参数关系模型，形成带有闭环优化特征的半导体生产线自适应动态调度方法；最后，以实际生产线为背景，研发半导体生产线运作模拟实验环境，该环境具有与实际半导体生产线“平行”的能力，用于学习样本的生成与提出方法的验证优化。本项目丰富和发展了复杂生产系统调度理论，提出了解决半导体生产线调度难题的新思路与新方法，为发现复杂生产系统调度新的优化手段提供了科学依据。