Short-Interval Detailed Production Scheduling in 300mm Semiconductor Manufacturing using Mixed Integer and Constraint Programming

Fully automated 300mm manufacturing requires the adoption of a real-time lot dispatching paradigm. Automated dispatching has provided significant improvements over manual dispatching by removing variability from the thousands of dispatching decisions made every day in a fab. Real-time resolution of tool queues, with consideration of changing equipment states, process restrictions, physical and logical location of WIP, supply chain objectives and a myriad of other parameters, is required to ensure successful dispatching in the dynamic fab environment. However, the real-time dispatching decision in semiconductor manufacturing generally remains a reactive, heuristic response in existing applications, limited to the current queue of each tool. The shortcomings of this method of assigning WIP to tools, aptly named "opportunistic scavenging" as stated in G. Sullivan (1987), have become more apparent in lean manufacturing environments where lower WIP levels present fewer obvious opportunities for beneficial lot sequencing or batching. Recent advancements in mixed integer programming (MIP) and constraint programming (CP) have raised the possibility of integrating optimization software, commonly used outside of the fab environment to compute optimal solutions for scheduling scenarios ranging from order fulfillment systems to crew-shift-equipment assignments, with a real-time dispatcher to create a short-interval scheduler. The goal of such a scheduler is to optimize WIP flow through various sectors of the fab by expanding the analysis beyond the current WIP queue to consider upstream and downstream flow across the entire tool group or sector. This article describes the production implementation of a short-interval local area scheduler in IBM's leading-edge 300mm fab located in East Fishkill, New York, including motivation, approach, and initial results