3110. Improvements of a Column Generation Approach for Production Planning in Semiconductor Manufacturing
Invited abstract in session WA-21: Metaheuristics for Scheduling and Production, stream Metaheuristics .
Wednesday, 8:30-10:00Room: Esther Simpson 2.12
Authors (first author is the speaker)
| 1. | Camil ZARROUK
|
| 2. | Nabil Absi
|
| Ecole des Mines de Saint-Etienne - LIMOS | |
| 3. | Stéphane DAUZERE-PERES
|
| Manufacturing Sciences and Logistics, Ecole des Mines de Saint-Etienne - LIMOS |
Abstract
Semiconductor manufacturing includes the most complex production processes, involving several hundred steps for each product. Additionally, re-entrant flows and congestion must be considered, resulting in a production timeline of approximately 8 to 20 weeks per wafer of chips. This complexity is further exacerbated by the large volume of wafers processed in a single factory with hundreds of heterogeneous machines, as well as the large number of different products in high-mix factories.
In this work, we consider an operational production planning problem in semiconductor manufacturing. The column generation approach of Béraudy et al. (2022), that relies on the notion of timed routes where a specific time period is assigned to each production step for a product, has been extended by Anthouard et al. (2023) to consider various industrial constraints. However, to improve the convergence of the column generation approach for industrial instances, it is crucial to start with an appropriate initial set of timed routes. Given that the production process spans several months and that production plans are executed in a rolling horizon, we propose in this work to integrate and adapt previously generated timed routes to enhance the stability of the production plan and to significantly reduce the computational times. Computational results on industrial instances will be presented.
Keywords
- Column Generation
- Production and Inventory Systems
Status: accepted
Back to the list of papers