Optimization Models for the downbinning problem in Semiconductor Manufacturing

From Iskomunidad
Revision as of 20:36, 10 February 2012 by Mqmallonga (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Darlene Monoy Fenix

Thesis (MS Industrial Engineering) - University of the Philippines Diliman-2007

Abstract

The semiconductor industry is a complex global industry that presents many opportunities for optimization. Within semiconductor manufacturing is a process called downbinning, which is configuring higher grade units to support lower grade demand. Although downbinning reduces revenue potential since the units, as these are sold at the price of the lower grade, it is still practiced to support customer demand. There are opportunity costs associated to downbinning and incurring backlog. Thus, it is important to determine optimal quantities to downbin that minimizes total cost. There are existing supply chain models that tackle downbinning but these are insufficient to address the specific requirements in semiconductor manufacturing. Therefore, the need generate alternative models is warranted.

The present study explored the use of dynamic programming (DP) and linear programming (LP) to model the downbinning problem in semiconductor manufacturing. DP and LP are widely-used and commonly-accepted models in the industry. The two models tested on five historical datasets and performance assessments of the two models were based on availability, stockout fractions and downbinning quantities.

The results showed that the LP showed better performance in terms of availability and stockout fractions, as it considered all future demand and was able to reserve inventory to support higher priority customer requirements. For downbinning, LP used more higher grade supply to support lower grade demand during the earlier part of the solving horizon to avoid incurring backlog penalties week to week.

The study was able to successfully generate the DP and LP models to address the downbinning problem in semiconductor manufacturing. Some scenarios were also specified where the use of these models will be appropriate.


Subject Index : Semiconductor industry