Time bound sequences are constraints deemed necessary to ensure product quality and avoid yield loss due to time dependent effects. Although they are commonly applied in production system control they cause severe logistical challenges. In this paper, we evaluate the effects of time constraints in combination with batching on a real metallization work center of an opto-semiconductor fab. We use simulation to analyze the impact of these production constraints and point out potentials to increase work center performance. We have a closer look at the required planning horizon, the influence of dedication, the capacity loss due to time bounds and the effects of batching strategies on wafer cost. Our results show the importance to tackle these issues. Furthermore, we will discuss actions taken in response to the experiments.
Time bound sequences are a common constraint in semiconductor manufacturing. These constraints represent time bounds in which a number of succeeding process steps should be performed. Violating these constraints usually necessitates rework or, even worse, the scrapping of the affected wafers. The reason for these constraints is usually to keep particle contamination and surface reactions to a level where it does not influence the process quality. To avoid violating these time bounds effective dispatching or scheduling is used to keep them to a statistical minimum. The effect on the system is usually that some lots, batches or wafers are on hold until sufficient resources are available to ensure that they can be processed before the time runs out. While trying to ensure non-violation release strategies we basically trade equipment utilization for cycle time. There are several approaches in literature to tackle this issue. Robinson (1998) and Robinson and Giglio (1999) presented a basic approach to capacity planning with time bound constraints and calculations to estimate time bound violations. Scholl and Domaschke (2010) proposed a Kanban-type approach where tool capacity is limited directly. Klemmt and Mönch (2012) proposed a heuristic and an MIP based approach of scheduling lots in a time bound sequence. In general, there is always some loss of capacity to ensure as few as possible violations.
In this paper, we want to present a first study of a coating work center and its time bound sequences to evaluate the impact of different logistical characteristics on the system. The system at hand is a group of batch coating equipment with a number of tools supporting preprocessing and handling steps. The time bounds considered are a mix of time bound sequences with and without intermediate steps using a number of different time targets. The system is modeled from a productive manufacturing line of an opto-semiconductor manufacturer which in contrast to traditional semiconductor manufacturers faces a broader spectrum of materials used for coating. This, in turn, increases the number of different recipes, tool dedications and processes in the coating workshop and therefore its complexity.
With the recent rise in demand for LED, Osram Opto Semiconductor is drastically increasing its efforts to collect and use fab data to improve their logistic processes. Although these efforts have come far, data availability has to be further improved to compete with leaders from traditional semiconductor fields. Therefore, the completely automated generation of a workshop model is not feasible at the moment. At the same time, a completely manual approach to model building is similarly unfeasible as the amount of information to create a simulation model for a specific workshop is near to impossible to maintain in an environment of permanently changing products, product mix and tool set. Hence, we use a semi-automated approach to generate models. The approach is visualized in Figure 1. Furthermore, considering increased involvement of users, an Excel frontend was chosen as a familiar user interface to increase acceptance and compatibility with input data not yet available in data bases. In our concept, we provide the user with an Excel template which is able to import raw fab data from a number of data bases. In a second step, this data can be adapted and missing data is added to create a full dataset. This is done in an Excel spreadsheet to provide the users with their daily work environment to reduce usability issues and training time. From this user-interaction based format we generate a second Excel file translating all information into a machine readable format which is compatible with our simulation meta model. During this first transformation we make the model machine readable and transform values and measures from the units as they are used on a day-by-day basis to a set of systematic units to standardize data.