Translating Overall Production Goals into Distributed Flow Control Specifications for Semiconductor Wafer Fabrication

• Main Authors: Ming-Der Hu, 胡明德
• Other Authors: Shi-Chung Chang
• Format: Others
• Language: en_US
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/77722718568322523117

博士 === 國立臺灣大學 === 電機工程學研究所 === 90 === Manufacturing systems are often operated under distributed control. To achieve the global goals of a manufacturing system in a coordinative way, the local controllers, either computers or human decision-makers, are often given local production flow control targets to follow or go beyond by superior controllers. How to translate overall production goals, such as system output rate and cycle time, into local flow control targets, such as work-in-process (WIP) levels and part release intervals, for internal operations has been a significant and challenging research topic. This dissertation defines and studies the production goal translation problem for semiconductor wafer fabrication plants (fabs), where there are multiple part types, failure prone machines, and deterministic re-entrant process flows. Such a system is first modeled as a failure-free, batch-free, re-entrant and open queueing network (OQN) with an aggregated part type. In conjunction with the re-entrant OQN model and the decomposition-based approximation method, a backward queueing network analysis (BQNA) is designed. Taking a reversed viewpoint of relations among variables in the results of the decomposition analysis, BQNA leads to two sets of linear equations with production output goals and nodal service parameters such as mean and variability of service times given. We apply BQNA to derive flow control parameters in terms of means and variations of input processes for each machine group. These derived control parameters can be utilized to obtain managerially tangible requirements for cycle times and WIP levels and to create targets for distributed flow control leading to the desired overall production goals. Comparisons with simulation results over two example wafer fab models demonstrate the accuracy and computational efficiency of BQNA and its potential for real applications.