Abstract: To solve the dispatching problem in semiconductor wafer manufacturing system with large amount of uncertain manufacturing environment,an objective-driven dynamic dispatching rule based on Extreme Learning Machine (ELM) was proposed.Through simulation system of semiconductor product line,the needed sample was obtained,and a performance predictive model was built with ELM.By using the predicted performance index and combining with the real statement production line information,the best coefficient needed by Dynamic Dispatching Rule (DDR) algorithm was calculated to help the manufacturing system dispatching lots,which could improve the overall performance of product line.The proposed dispatching rule was validated by simulation platform.

Key words: semiconductor manufacturing, objective-driven, dispatching rule, working condition, extreme learning machine

摘要： 为了求解不确定生产环境下的半导体生产线调度问题,提出一种基于极限学习机的性能指标驱动的动态派工方法。首先通过半导体生产线仿真系统仿真,得到所需样本,进而通过极限学习机建立半导体生产线性能指标预测模型;然后通过预测出的性能指标,结合生产线的实时状态信息,学习产生调度过程中的动态派工规则算法所需的最佳系数,驱动生产线派工决策,使其性能指标趋向预测值,最终达到提高生产线整体性能的目的。通过仿真平台验证了该方法能够有效提高生产线的性能指标。

关键词: 半导体制造, 性能指标驱动, 调度方法, 工况, 极限学习机