Abstract
A TFT-LCD (thin film transistor-liquid crystal display) manufacturing consists of three main processes: TFT Array, Cell and Module process, and TFT Array is the most important of all. Being the first manufacturing stage, the throughput of TFT Array has a significant impact on the subsequent processes. In order to prevent material shortages in Cell plants, TFT Array plants must produce the right products with correct quantities to comply with the daily throughput schedule. In addition, the resource utilization in TFT Array process has been placed high attention due to expensive equipments. As a result, an effective release plan is necessitated for TFT Array process to achieve the daily throughput schedule and simultaneously exploit the finite resource. Many manufacturing characteristics and constraints in TFT Array process lead to a great complexity in TFT Array shop floor scheduling. Firstly, TFT Array process is a re-entrant flow, in which a similar sequence of processing step is repeated for five times. Further, the photolithography stage, the bottleneck in TFT Array process, requires a second resource in addition to machines (steppers or scanners). Photo masks and scanners are both indispensable while executing exposure operation in the photolithography stage. Every product with different re-entrant layer requires a unique mask and consequently each has a set of mask with five different pieces. Moreover, there is a tool eligibility issue between masks and scanners. Masks are approved to be used in specific scanners for quality considerations. Since TFT Array process is characterized as reentry and mask constraints, it is regarded as a complicated scheduling problem and is more difficult to solve than classical ones. In this study, we decompose the complex problems and present a two-phase scheduling method to acquire a suitable release plan for TFT Array process, which are masks allocation and a DBR-based bottleneck operations scheduling. In the first phase, masks allocation problem is quickly solved by mathematical programming and each mask is designated to one or more scanners. Phase one is similar to a rough-cut capacity planning. Afterward the detail bottleneck operations scheduling and the release plan is established according to DBR (drum-buffer-rope) scheduling method in accordance with the mask allocation result from the previous phase. Finally, relative scheduling performances of the two-phase scheduling method are provided by a simulation model extracting from a real TFT-LCD manufacturing company. The release plans developed for the tardiness and resource utilization criteria reveal that the proposed methodology quickly solves the difficult scheduling problem and performs well.