Avoiding Artificial Bottlenecks in Semiconductor Wafer Fabrication Facilities

小行星9813345 该基金通过两个相关的模型资助半导体晶圆制造设施的能力研究：流体模型，它捕获系统动态，但忽略随机变化和多类封装网络，它描述了晶圆制造厂相当详细。 李雅普诺夫函数的分析应用于流体模型的底层封装网络的目的是确定人为瓶颈的来源，降低了晶圆厂的能力低于限制所施加的个别工具。 这种类型的分析还可以识别给定的调度策略或调度策略族是否容易受到人为瓶颈的影响。更详细的底层决策网络的布朗模型旨在帮助区分基于次要性能标准的替代政策。 最后，赠款资金的努力，与模拟公司，以实现家庭的实际和有效的政策，在其商业软件中使用的晶圆制造厂。 本研究的目的是通过以下三个方面提高现代晶圆厂中昂贵的瓶颈工具的利用率：（1）通过更好地理解调度策略如何导致不必要地降低晶圆厂产能的人为瓶颈;（2）通过确定新的有效调度策略，考虑到晶圆厂的实际情况，如设置和调试;以及（3）通过提供分析和计算工具来评估和优化性能标准，如跨有效调度策略族的平均调度时间和周期时间。

9813345Vande Vate This grant funds the study of capacity in semiconductor wafer fabrication facilities through two related models: fluid models, which capture system dynamics, but ignore stochastic variability and multi-class queueing networks, which describe wafer fabs in considerable detail. Analysis of Lyapunov functions applied to fluid models of the underlying queueing networks is intended to identify the sources of artificial bottlenecks that reduce the capacity of the fab below the limits imposed by individual tools. This type of analysis can also identify whether or not a given dispatch policy or family of dispatch policies is susceptible to artificial bottlenecks. More detailed brownian models of the underlying queueing networks are intended to help distinguish among alternative policies on the basis of secondary performance criteria. Finally, the grant funds efforts to work with simulation companies to implement families of practical and effective policies in their commercial software for use in wafer fabs. This research is intended to lead to improved utilization of expensive bottleneck tools in modern wafer fabs in three ways: (1) by providing better understanding of how dispatch policies lead to artificial bottlenecks that unnecessarily reduce fab capacity; (2) by identifying new efficient dispatch policies that consider the practical realities of wafer fabrication like setups and batching; and (3) by providing analytical and computational tools for for evaluating and optimizing performance criteria like averageWIP and cycle time across families of efficient dispatch policies.