Performance of a Serial-Batch Processor System with Incompatible Job Families under Simple Control Policies

Abstract

A typical example of a batch processor is the diffusion furnace used in wafer fabrication facilities (otherwise known as wafer fabs). In diffusion, silicon wafers are placed inside the furnace, and dopant is flown through the wafers via nitrogen gas. The higher the temperature, the faster the dopant penetrates the wafer surface. Then, a thin layer of silicon dioxide is grown, to help the dopant diffuse into the silicon. This operation can take 10 hours or more to finish processing, as compared to one or two hours for other wafer fab operations, according to Uzsoy [8]. Diffusion furnaces typically can process six to eight lots concurrently; we call the lots processed concurrently a batch. The quantity of lots loaded into the furnace does not affect the processing time. Only lots that require the same chemical recipe and temperature may be batched together at the diffusion furnace.

We wish to control the production of a manufacturing system, comprised of a serial processor feeding the batch processor. The system produces different job types, and each job can only be batched together with jobs of the same type. More specifically, we explore the idea of controlling the production of the serial processor, based on the wip found in front of the batch processor. We evaluate the performance of our manufacturing system under several simple control policies under a range of loading conditions and determine which control policies perform better under which conditions. It is hoped that the results obtained from this small system could be extended to larger systems involving a batch processor, with particular emphasis placed on the applicability of such policies in wafer fabrication.