On-line near-real-time scheduling of semiconductor wafer fabrication
Ang, Adeline Tee Hwee
Date of Issue2008
School of Mechanical and Aerospace Engineering
Very little research have considered the combined aspects of scheduling involving (i) a dynamic and stochastic shop floor, (ii) Sequence-Dependent Setup (SDS), (iii) transportation, and (iv) the consideration of user preferences to achieve the best/preferred trade-offs amongst the multiple contradicting objectives. Given these complexities, it is indeed challenging to determine when and how the next task (i.e. transportation, setup, processing of a lot) should be selected and executed on each machine such that the best/preferred trade-offs amongst the multiple conflicting objectives can be achieved based on user preferences in an online, near-real-time manner. In this research, the NEXt Task Scheduling (NEXTS) methodology, made up of a NEXTS time policy and a NEXTS decision policy, is proposed. The NEXTS time policy specifies when to schedule. In essence, the time policy stipulates the appropriate timing (i.e. before and as close as possible to the time when the machine is available) for the selection and transportation of the next lot. The NEXTS decision policy specifies how a scheduling decision is to be made. The decision policy is made up of a multiobjective rule (i.e. weighted aggregation of dispatching criteria/rule) that is applied for the selection of the next lot and a Look Ahead Simulation-based Genetic Algorithm with Desirability function (LASGAD) that optimizes the weights used in the multiobjective rule in accordance with user preferences. The multiple performance measures of interest in this research are average cycle time, standard deviation of cycle time, average tardiness, and standard deviation of tardiness. The NEXTS time policy is shown to perform better than the existing Periodic Scheduling/Rescheduling and dispatching time policies when there are transportation and decision making times involved. Simulation results also show that on-line LASGAD is better at finding the optimal set of weights than the off-line response surface methodology with desirability function method. The novel NEXTS methodology contributes to the advancement of knowledge by making use of real-time data for effective on-line scheduling of a dynamic and stochastic shop floor with sequence-dependent setups, transportation times, and user specified preferences to obtain Pareto optimal performance in the face of multiple conflicting objectives.