Semiconductor production planning using robust optimization

We consider a semiconductor chip production planning problem, where chips with different performance characteristics are produced from the same wafer supply simultaneously. Due to long production cycle times, decisions on the wafer production need to be executed prior to knowing the demands and binning information exactly. Once this information is realized, assignment decisions are then executed to allocate the available production to satisfy the demands. Furthermore, product substitution is allowed in the allocation. To address the issue of data uncertainty in the planning process, in this work we propose to use the robust optimization approach to develop a new planning model for the problem. Our model is based on a two-stage robust network flow problem, and we demonstrate that by using our proposed model, we are able to achieve production plans that can hedge against the random variations in the data without over-sacrificing the solution quality. Furthermore, the robust optimization models require limited distributional assumptions and result in linear programming counterpart problems, which can be solved efficiently using commercial solvers.