Technical Paper: Journal of Manufacturing Systems 2022

Authors:

Ruaridh Williamson, Semya Elaoud, Ioannis Konstantelos, Jay Maguire, and Dennis Xenos.

Abstract:

This work examines the scheduling of a Renesas semiconductor wafer fabrication plant modelled as a complex constrained flexible job-shop problem. We model the clean and diffusion furnace area with batch scheduling where different batches of jobs have different operating costs and consecutive steps of a job are constrained with timelinks (Timelinks mainly define the maximum amount of time that cane lapse between two or more consecutive process steps of a lot to prevent oxidation and contamination). The model also involves other process aspects such as flexible tool downtime (i.e. a maintenance or a downtime period can start within a window rather than having a fixed start time), incompatible job families, different job sizes, and parallel machines. We present a hybrid two-stage solution strategy, combining an advanced decomposition strategy for Mixed Integer Linear Programming (MILP) models and heuristics. This solution strategy is then benchmarked by minimizing the total weighted batching cost, queuing time, and the number of violations of timelink constraints. Results are compared against those that the wafer fab actually incurred over a historical period of running its dispatching system built on rule-based heuristics. We also detail the simulation methodology used to evaluate the quality of the solution strategy under uncertainty. The comparison of this work’s schedules evaluated under uncertainty against factory schedules when solving large industrial instances shows the significant improvements that our solution can bring.

Keywords:

semiconductor manufacturing, job shop scheduling, MILP, timelink constraints, decomposition techniques, simulation-based performance evaluation