Search

We examine the problem of minimizing the mean cycle time of a batch processor with incompatible job families and future job arrivals. Batch processors frequently found in wafer fabrication include oxidation and diffusion ovens. When the problem is deterministic with a fixed number of job arrivals, the problem is NP-hard; an optimal dynamic program formulation is developed and empirically tested on small problem instances. For larger problem instances with possibly infinite job arrivals, an online heuristic based on model predictive control (MPC) is proposed. The proposed heuristic has two parameters: the number of job families considered f, and the number of future job arrivals L. Statistical analysis of simulation results shows that the MPC-based heuristic has significantly lower mean cycle time than a benchmark look-ahead method under the assumption that the arrival time is deterministic and job families of consecutive job arrivals are uncorrelated. These observations also hold true when the predicted arrival times can be erroneous, indicating that the suggested MPC-based heuristic is as robust as the benchmark with respect to errors in forecasted arrival times. When the job families of future job arrivals are positively correlated, the mean cycle time of jobs passing through the batch processor is significantly reduced, regardless of control policy. We infer that the introduction of correlation generates less improvement for policies that foresee events longer into the future. For this reason, when the correlation is high, the MPC-based heuristic, which considers events that occur farther into the future, may have worse performance than the benchmark. The results suggest two ways to improve the performance of the batch processor. First, we can use a heuristic that has better mean performance, which may entail much larger computational costs. Another option is to continue using a simple look-ahead policy at the batch processor, and focus attention on controlling the upstream processors, such that the batch processor arrival process exhibits correlated job families between consecutive arrivals. This introduction of correlation causes a significant reduction of cycle time regardless of system parameter and batch processor control policy.

This paper validates a proposed job release methodology and experimentally investigates the impact of production control methodologies and system factors on wafer fab performance in terms of average cycle time, standard deviation of cycle time, average lateness, WIP inventory and fab output by simulating a wafer fab of Chartered Semiconductor Manufacturing Ltd (Chartered) and statistically analyzing the experimental results using t-test and ANOVA. A full factorial design of experiment is conducted to evaluate the performance of three job release methodologies, three dispatching rules and three greedy levels of batching policy under different system environmental settings differentiated by fab output level and machine unreliability level. Based on the experimental results, the proposed job release methodology WIPLOAD control (WIPLCtrl) appears to be very efficient to be able to potentially improve all the considered performance measures simultaneously. The advantage of WIPLCtrl is robust to the change of system environmental conditions. In contrast, the improvement brought by a dispatching rule on a certain performance measure might cost the deterioration of other performances. Considering the relative impact on the fab performance, job release control appears to be the most important production control factor in comparison with dispatching and batching policy, especially when the system is operating on a high output level and/or with a high system variability level caused by machine unreliability.

A discrete event simulation based "online near-real-time" dynamic multiobjective scheduling system has been conceptualized, designed, and developed to achieve Pareto optimal solutions in a complex manufacturing environment of semiconductor back-end. Our approach includes the use of linear weighted aggregation optimization approach for multiple objectives and auto simulation model generation for online simulation. Developed concepts are implemented at a semiconductor back-end site and are in use. The impact of the system includes a better customer delivery achievement, consistent cycle time with narrower distribution, improved machine utilization, reduction in the time that planners and manufacturing personnel spend on scheduling, and more predictable and repeatable manufacturing performance. In addition, it enables managers and senior planners to carry out "what now" analysis to make effective current decisions and "what if" analysis to plan for the future.