A proposal on optimized scheduling methodology and its application to an actual-scale semiconductor manufacturing problem

Abstract

Semiconductor manufacturing process flows typically include several hundred process steps, using hundreds of machines repeatedly. The production schedule for such a re-entrant flexible flow shop is extremely difficult to be optimized. Furthermore, in recent years, it has become increasingly difficult to use machines effectively because of the product mix increase and also the production lot size decrease. In this paper we apply the well-known cooperative scheduling method, Lagrangian Decomposition and Coordination method, into an actual large scale model in semiconductor manufacturing. Then we show that the proposed method successfully creates a well-performed production schedule as we expected.

Introduction

After the late 1990s, because of the emergence of new business models such as foundry companies in Taiwan and fab-less companies in the U.S., the Japanese semiconductor industry has shifted from low-mix high-volume memory production to high-mix low-volume microcomputer production. Therefore, delivery has come to pose a key challenge that must be solved continuously from the perspective of non-price competitiveness.

However, because semiconductor manufacturing process flows typically comprise hundreds of processing steps using hundreds of machines repeatedly, the production schedule of such a re-entrant flexible flow shop is extremely difficult to optimize. Furthermore, in recent years, it has become increasingly difficult to use machines effectively because product mixes have become much more complex and production lots have become much smaller. Consequently, the number of scheduled jobs has increased dramatically. Therefore, it is indispensable to integrate various scheduling methods into the solution to achieve high productivity. Such methods include flexible lot batching, finely tuned lot dispatching, and adaptability to large-scale models performing comprehensive production-line control [1].

Previous studies have assessed many methods to improve productivity through lot dispatching methods. One proposal specifically examines only aspects of the production process [2]. Another provides decision-making support [3] to improve productivity solely through the production capacity enhancement. Nevertheless, they have not brought comprehensive optimization.

Lagrangian Decomposition and Coordination method, a cooperative scheduling method, was developed by Luh and Hoitomt [4] for a parallel machine scheduling problem. They extended it to a job shop model [5]. Nishi et al. applied this method successfully to a flow shop model [6]. Jiang and Tang [7] applied it into the re-entrant flexible flow shop model such as a semiconductor manufacturing process and demonstrated its effectiveness for a small scale model. However, they left challenges for the large-scale problem. Furthermore, the authors’ previous research [8] was applied to semiconductor production systems, but the objective was to create a schedule capable of facilitating proper maintenance in a small scale model.

For this study, we apply a well-known cooperative scheduling method, Lagrangian Decomposition and Coordination method, to an abstracted small scheduling model based on an actual semiconductor production line. Thereby we verify that the proposed method can create a production schedule effectively. We also verify the effectiveness of the proposed method's application to larger models and clarify issues related to its application to an actual large-scale model.