Semiconductor manufacturing fab is one of the most sophisticated man-made system, consisting of hundreds of very expensive equipment connected by highly automated material handling system. Operation schedule has huge impact on the productivity of the fab. Obtaining efficient schedule for numerous equipment is a very complex problem, which cannot be solved by conventional optimization techniques. Hence, heuristic dispatching rules combined with fab simulation is often used for generating fab operation schedule. In this paper, we formulate the fab scheduling problem as a semi-Markov decision process and propose a reinforcement learning method used in conjunction with the fab simulator to obtain the (near-)optimal dispatching policy. Resulting schedule obtained by the proposed method shows better performance than heuristic rules whose parameters are tuned by human experts.

Most manufacturing systems in which jobs arrive dynamically and their processing times have variations use dispatching rules to obtain production schedules. In LCD manufacturing, several dispatching rules that reflect the knowledge of the fab operator have been developed and prioritized to select a unique job for processing on a machine. However, engineers rank the dispatching rules based on their experiences without any systematic analysis method. Hence, there is a great need for a tool that can analyze how the order of dispatching rules affect the key performance indicators (KPIs) of schedules. Therefore, we provide a framework for the performance analysis of dispatching rules so that engineers can examine the KPIs for a given order of dispatching rules and find the best order of dispatching rules.

Presented in this paper is a dispatching rule to achieve the on-time delivery and throughput for a semiconductor wafer fabrication (FAB) with dedication constraint. It is necessary to improve throughput of production system as well as on-time delivery to make semiconductor companies more competitive. To be successful in the globalized competition, most of the companies have applied a concept of the production target which is defined as the quantity of production to be achieved for each day. Also, it is necessary to consider natural bias that significantly affects the alignment of patterns between different photolithography steps. As the natural bias has a negative effect on the quality and yield of products, most manufacturers have applied dedication constraint. If the dedication constraint is no managed properly, it may decrease the utilization of the photo machines related to the performance of wafer FAB. To overcome the problem, we propose the two boundary based dispatching rule considering the dedication constraint. The simulation model based on MIMAC6 was developed to prove the performance of this proposed dispatching rule, and conducted a simulation by using MOZART. The simulation results clearly show the advantages of the proposed dispatching rule over the other dispatching rules.

This paper proposes rescheduling algorithms for improving schedules obtained by dispatching rules with a commercial software program, MozArt, developed by VMS Solutions Co., Ltd.. Schedules for flexible flow shops with sequence-dependent setup times and job splitting are analyzed. The objective of the algorithms is to reduce the completion time by decreasing the number of setups and setup times. We first identify four types of problems with badly assigned job sequences and unnecessary idle times in given schedules derived from dispatching rules, and solve the problems by changing the sequence of jobs or splitting jobs. The performance of the proposed algorithms is tested with randomly generated instances based on real data from a factory in Korea.

This paper addresses a semiconductor wafer fabrication (FAB) scheduling problem with dedication constraint. Under dedication constraint, a fabrication lot must be processed using the same photo machine at all photolithography (photo) steps. To solve the utilization decrease of photo machines by dedication, we propose a dedication load as the sum of the workload of lots dedicated to each photo machine. When a photo machine becomes available to process a new lot, if its dedication load is less than the average of similar machines, then the photo machine will be assigned to process the first step of a new lot in the event that one is available. To prove the performance of this proposed dispatching rule, we developed a simulation model based on MIMAC6, and conducted a simulation by using MOZART. The proposed dispatching rule was implemented and outperformed conventional dispatching rules.

Presented in this paper is a dispatching rule for engineering lots by a wafer FAB. The proposed rule uses the concept of reservation to ensure greater capacity for the engineering lots. Although a reservation based rule was previously proposed by one of the authors, it has remained difficult to ensure the necessary capacity for the engineering lots because of variations in the processing time for each step of the process. In this paper, we enhanced the previously proposed rule to reflect the fact that batch tools often cause bottlenecks because of their long processing times. We developed a FAB model using the Measurement and Improvement of Manufacturing Capacity (MIMAC) dataset 6, and performed simulations with MozArt. The simulation results clearly show the advantages of the enhanced reservation based dispatching rule over the previous version of the rule and that it is vital to identify an appropriate reservation range.

Diversified and complicated manufacturing sites make optimal scheduling of production lines difficult. Under current manufacturing processes, it is almost impossible for schedulers to consider all the constraints of production processes. A strategy of simulation-based advanced planning and scheduling (APS) is employed to overcome difficulties that interfere with satisfactory on-time delivery and commitment to the current status. In simulation-based scheduling, key performance indices (KPIs) are important for selecting optimal dispatching rules in scheduling. In cases involving complex processes, in which the identification of appropriate KPIs is limited to selection among existing KPIs, KPIs should be chosen and modified carefully to optimize process management and to reflect all of the existing constraints of production. However, the existing methodologies for modifying KPIs are misplaced in complex manufacturing environments such as job-shop processes. We propose a new method to design and select appropriate KPIs that meet the characteristics of any given process, and verify with empirical analysis whether or not the KPIs meet requirements from experts of production lines.

Presented in this paper is a dispatching rule to achieve the on-time delivery for an order-driven FAB involving high priority orders. We classify orders into two groups; regular orders (RO), high priority orders(HPO). HPO lots have shorter cycle times, tighter target due date and higher margins than RO lots. The proposed rule introduces the concept of reservation for HPO lots, which means the provisional allocation of capacity to meet the on-time delivery of HPO lots. Since the rule considers the due date of HPO lots first, RO lots might be tardy. To minimize the tardiness of RO lots, the proposed rule considers tool utilization as well as on-time delivery of HPO lots. We developed a simulation model based on MIMAC6, and conducted experimentations with MOZART. The experimentation results show that the proposed dispatching rule can achieve the on-time delivery of HPO lots with the minimum tardiness of RO lots.