Research On Modeling And Scheduling For Vertical Oxidation/Diffusion Furnace

Vertical oxidation/diffusion furnace is available for almost all thermal processes insemiconductor manufacturing, such as oxidation, diffusion, deposition, and so on.Batch processing mode and long cycle time make it one of the main bottlenecks insemiconductor manufacturing system. Therefore, scheduling algorithm has a crucialrole in improving the performance of the furnace control system.In this dissertation, an object-oriented Petri net model for the process of a verticaloxidation/diffusion furnace is constructed. The furnace is divided into three functionunits, stocker, robot and chamber. Object Petri net and objcet communication net areestablished for each unit. Invariant analysis is applied to check the dynamicperformance of object communication nets. The steps in control softwaredevelopment for vertical oxidation/diffusion furnace using object-oriented Petri netare formulated. The character of the process of the vertical oxidation/diffusionfurnace is concluded by modeling and analyzing. What’s more, hybrid Petri nets areused for further analysis of the batch scheduling problem for both single furnace andmultiple furnaces.Secondly, a two-level batch scheduling algorithm for single furnace and athree-level batch scheduling algorithm for multiple furnaces are designed based onparticle swarm optimization. The objective is to minimize the total completion timewith dynamic multi-lot arrivals. The wafer loading/unloading time is considered, aswell as the setup time. In the two-level scheduling algorithm, a search set isdeveloped for batching decision, and particle swarm optimization is applied tosequencing decision. In the three-level scheduling algorithm, a search space is builtfor batching decision as well, while particle swarm optimization is applied todistributing decision, and FIFO is applied to sequencing decision. In both algorithms,the wafer loading/unloading is planned. Simulation experiments based on theparameters of the furnace are carried out to evaluate the computation speed, theeffectiveness and the validity of the proposed algorithms. It is shown by thesimulation results that the proposed algorithms are superior to MBS rule andMBS+SPT rule.Finally, the single-furnace scheduling procedure is tested and embedded in thefurnace control system to make schedule-decisions.