A Model Predictive Control Approach for Managing Semiconductor Manufacturing Supply Chains under Uncertainty

AbstractA two level architecture using Model Predictive Control as a tactical decision module ispresented for supply chain management in the semiconductor manufacturing industries. Thestrategic and inventory planning steps in the outer loop provide the inventory targets and ca-pacity limits by solving an optimization problem that maximizes pro ts. These decisions areusually weekly or monthly. The MPC-based tactical decision module takes advantage of thesetargets, capacity limits and demand forecasts to make daily decisions on starts at the variousmanufacturing nodes. Fluid analogies are used to model the supply chain dynamics in semi-conductor manufacturing which facilitates the application of Model Predictive Control. Severalbenchmark problems which contain distinguishing features of semiconductor manufacturing,such as nonlinear and stochastic throughput times and customer demands, are examined. All ofthese problems involve two types of manufacturing nodes, Fab/Test1 and Assembly/Test2, andthree types of inventories, Assembly-Die Inventory, Semi-Finished Goods Inventory and Fin-ished Components Inventory. Both supply side uncertainty, including varying throughput timesand yields, and demand side uncertainty are addressed. The nonlinear relationship between thethroughput time and load is considered in each case. The e ects of judiciously picking tuningand model parameters to achieve performance, robustness and improved customer satisfactionare studied by comparing the variance in starts, inventories, and load as well as the percentageof un lled orders. Increasing move suppression and choosing the nominal throughput times ataverage values usually gives better performance with lower variance and less backlog. The exi-bility provided by the choice of tuning and model parameters in MPC to achieve more e ectivesupply chain management in semiconductor manufacturing is demonstrated in each case study.