| Three different approaches for scheduling semiconductor manufacturing operations are proposed.;The first approach proposes a fab-wide dispatching scheme called Balanced Machine Workload (BMW) which considers K-machine look ahead and J-machine look back. This research is motivated by the Minimum Inventory Variability Scheduling (MIVS) policy which considers the inventories at the next K-steps and the previous J-steps. Instead of looking at the fab flow from the step viewpoint, the BMW dispatching scheme looks at it from the machine viewpoint. The proposed dispatching scheme checks the workload of K-downstream machines and J-upstream machines for multiple products. Simulation results demonstrate superior performance of the BMW dispatching scheme over MIVS.;With the second approach, a control heuristic is proposed to deal with pre-cleaning (wet etch) and furnace. In semiconductor manufacturing, it is critical to schedule batching operations efficiently. Previous studies in batch scheduling have focused on efficiently creating the batches with the given jobs and deciding whether to start a partial batch or to wait for future arrivals to make a full batch. The new proposed scheduling system, called NACH+, tries to integrate the incoming inventory into the batch operation. Simulation results demonstrate superior performance of the NACH+ scheduling heuristic over Minimum Batching Size (MBS).;The third approach proposes real-time scheduling for semiconductor manufacturing based on integer programming. The disjunctive programming IP formulation is extended to the following models: (1) a full-enumeration scheduling problem which minimizes Cmax, (2) a real-time scheduling problem which simply maximizes the number of job assignments at the current time, and (3) a real-time scheduling problem which considers a line balance aspect while maximizing job assignments at the current time. The effectiveness of the real-time scheduler in terms of solution quality and run time is evaluated through computer experiments. The results obtained from the real-time scheduler are compared with Cplex results. Obviously, the real-time scheduling model cannot guarantee global optimality since it seeks optimality only for a short time frame. However, the experimental study shows the well-defined optimization-based real-time scheduling heuristic can generate close-to-optimal solutions. |
