Research On Effective Workload-dependent Release And Idling Dispatch Policy For Wafer Fabrication System

Semiconductor wafer fabrication system(SWFS)is recognized as one of the most complicated manufacturing systems due to its complex and multi-reentrant process flows,various process modes and multi-product mix production.Consequently,its production scheduling and control is vital to achieve shorter average cycle time(AVG_CT),lower work in process(WIP)and higher on time delivery rate(OTD)in a SWFS.The advanced production scheduling approaches pursue the control of the output and the optimization of the key performance indicators of a SWFS by utilizing release policies to control its input and dispatching policies to control the specific flow of jobs within it.Release policy tries to control the time,product type and number of the newly released jobs based on the difference between the current and target workload levels.However,all existing release policies make release decisions based on the workload in name rather than the effective workload.Hence,they are inaccurate.Dispatching policy first sorts the released jobs and then chooses the most desirable job to be processed at the right time.However,all existing dispatching policies subject to the non-idling type(machine must keep on processing until there is no job available for dispatching)and do not take full advantage of the information of workload distribution(namely system states)to control production rates of machines accurately.Hence,they are not the most efficient ones.To address the above two questions,this research focuses on the effective workload-dependent release and idling type dispatch policy for the SWFS.The detail works are listed as follows:(1)Effective workload-dependent release policyThe concept of effective workload is introduced to incorporate the impact of the number of identical parallel machines,the batch capacity,the workstation availability and the capacity utilization level factors on the workload.The workload of a production line is then measured accurately by evaluating its effective workload.We treat the workload of jobs with different product types and the workload of jobs with the same product type but different process steps differentially.A release policy(named EWL-n-Ctrl for short)that tries to maintain the total effective workload of each product in the production line at its corresponding target level is then proposed.Extensive simulation experiments validate the efficiency of the effective workload-dependent release policy(EWL-n-Ctrl)under three different scenarios(e.g.the deterministic and random environments,single and multiple products,different capacity utilization levels and product mix).(2)API-based proactive dispatching decisionTo control the production rates of machines accurately through dispatching,we first derive the information of operation due dates from the target distribution of the workload(i.e.,WIP)based on Little’s law.Then we formulate the dispatcher as the mean absolute deviation(MAD)problem to minimize the sum of absolute deviations of the operation completion times from operation due dates and the decision variables include the process sequence and start processing times.During the dispatching decision-making,we consider proactively the influence of current dispatching decision-making on finishing the other competing jobs punctually and try to minimize it to determine the process sequence and start times.Six dominance properties of this problem are established using proof by the technique of adjacent pairwise interchanges(APIs).A heuristic(named HMAD for short)is designed to solve this problem in real-time.Each time when a machine becomes idle,the proposed dispatcher chooses a target processing job from the competing jobs and assigns it a start time.Simulation study reveals that this approach is robust in keeping the jobs finished on time(neither too early nor too late)and the proposed HMAD achieves shorter AVG_CT,lower WIP level and higher OTD rate in the scenario of tight due dates.(3)Optimal release and dispatching policy for a benchmark RMSTo break the limit that the optimal dispatching policy for the benchmark reentrant manufacturing system(RMS)is assumed to be of non-idling type and further identity the structural property of its optimal release policy,the joint optimization of its release and dispatching policy is addressed.Considering that the optimal dispatching policy can be of idling type,we formulate a Markov decision process(MDP)model for the benchmark RMS to jointly optimize its release policy and dispatching policy.The optimal release and dispatching policy pursuing the maximum discounted profit of this system over an infinite horizon and their structural properties are achieved by using the theoretical derivation and value iteration algorithm.Theoretical extension indicates that the optimal release policy for the benchmark n-RMS(n≥1)is always a base-stock policy with a vector of workloaddependent base-stock levels.Numerical experiments imply that the optimal dispatching policy for the reentrant machine is of idling type while the optimal dispatching policy for the non-reentrant machine is still of non-idling type.Besides,the last-buffer-first-serve(LBFS)dispatching rule is optimal for the 1-reentrant machine and not optimal anymore for the reentrant machine whose degree of reentrancy is larger than one.Moreover,we design independent base-stock(IBS)and coordinated base-stock(CBS)heuristic policies for the benchmark 1-RMS based on the structural properties of the optimal release and dispatch policy.They both lead to a reasonably good approximation to the optimal profit of the concerned system and outperform the optimal non-idling policy.(4)Near-optimal release and dispatching policy for the Mini-Fab modelWe utilize the structural properties of the optimal release and dispatching policy to design some heuristic policies that are more effective than the existing ones.Compared with the benchmark 1-RMS,the process flow of the Mini-Fab model is more complex and the factors of parallel machines and batch processing machines are involved.Based on those structural properties of the optimal release and dispatching policy for the benchmark n-RMS,we design the near-optimal release and dispatching policy for the Mini-Fab model.Specifically,we mimic the step response behavior of the optimal release policy via a linear approximation.Since all the machines in the concerned Mini-Fab model are 1-reentrant machines(including single and batch processing machines),the near-optimal dispatching policies are all of idling type and they all subject to the LBFS manner.Simulation results reveal that the near-optimal release and dispatching policy can achieve better performance compared with the scheduling policy EWLCtrl+HMAD.The research idea that identifies the near-optimal release and dispatching policy according to the structural properties of the optimal release and dispatching policy is effective and promising.