Incremental production planning for the semiconductor industry using linear programming

Over the past decade, several production planning approaches emerged for the semiconductor industry. These approaches vary in their basic algorithm technology from heuristics, to artificial intelligence, to linear programming. Although most of these approaches were originally conceived to run in a batch or regenerative fashion, recent advances in the ability to capture and move data on a transaction level has provided motivation to develop production planning systems which react in near-real time to disruptive events within the supply chain. Application areas for an incremental type planning system include: (1) providing the ability to react quickly to unpredicted events within the supply chain, (2) facilitating scenario based analysis of alternative decisions or parameter inputs, and (3) enabling quick re-planning exercises as part of a delivery quotation process.; Although linear programming has gained support as a batch planning tool for the semiconductor industry, most recent initiatives for incremental planning employ alternative algorithm technologies. A perception has developed that while its ability to consider the planning problem as a whole and simultaneously evaluate tradeoffs amongst all products is a key strength to using linear programs in a batch fashion, it is not practical to apply linear programming as a tool for quick re-planning. A common conclusion is that a more localized approach is required to achieve acceptable re-planning times.; This work explores the feasibility of a paradigm shift for linear programming from batch based planning to incremental planning within the framework of the semiconductor industry. It develops techniques which quickly generate new production plans in response to various supply chain disruptions and demand opportunities. These techniques preserve the feasibility of the formulations and produce results with no loss in solution quality as compared to plans generated under a batch planning paradigm using linear programming. The techniques developed are tested in several experiments on a reference industrial data set from a commercial semiconductor company. It is demonstrated that single-digit second re-planning times for individual demand changes and single-digit minute re-planning times for individual supply changes are achievable with existing hardware and software technology.