| This dissertation analyzes the dispatching and releasing policies for semiconductor fabs and aims to devise better production control policies measured by mean cycle time and variance of cycle time at a given throughput rate. An adaptive, parsimonious scheme using real-time feedback is also proposed to predict the parameters necessary for applying the proposed policies. Simulation studies prove that these integrated policies perform better than other popularly known policies efficiently and stably.; The adaptive scheme utilizes exponential smoothing to predict mean flow time from every route step to factory exit and converges in a short time stably. The flow time estimates are then used to calculate job slack. The job with the least slack is chosen to be processed next when a machine is free (Least Slack dispatching policy). Average lateness, calculated from job slack, determines when a new lot is released into the system (Lateness Release policy). Unlike traditional release rules, the proposed release policy does not require any pre-specified threshold for a given obtainable throughput rate nor any detailed simulation to determine its parameters. It guarantees to achieve the targeted, obtainable throughput rate. The simulation results show that Least Slack dispatching policy and Lateness Release policy, with inputs from the adaptive scheme, achieve lower WIP levels than the best previously known methods.; One important factor in applying these policies is setting the due date, which is viewed as controllable variable. Two due date setting rules suitable for both single product and multi-product fabs are proposed. One is aimed to reduce mean cycle time and the other is to reduce variance of cycle time. Simulation results show that due date setting has a significant impact on the performance of the proposed scheduling policies.; Taking advantage of the computer-integrated manufacturing environments, the proposed adaptive scheme and scheduling policies are able to estimate all the parameters from the most immediate data, thus reflecting changes in factory, such as product mix and machine reliability, and making necessary adjustments timely. The work presented in this dissertation provides new insights into future studies taking advantage of the real-time feedback systems. |
