Model predictive control strategies for supply chain management in semiconductor manufacturing

Supply chain management (SCM) is the 21st century's global operations strategy for achieving organizational competitiveness. The supply networks characteristic of semiconductor manufacturing are highly stochastic, nonlinear and constrained dynamic systems. The stochastic processes associated with semiconductor manufacturing operate on timescales much shorter than the weekly basis considered in strategic planning. Novel decision policies are needed to respond to these stochastic processes on both supply and demand sides to reduce safety stock levels and factory "thrash". The results are with lower supply chain costs and improved levels of customer satisfaction resulting in higher profits.; In this dissertation, a two level hierarchical structure using Model Predictive Control (MPC) as a tactical decision module is presented for semiconductor manufacturing SCM. The inventory control problem is converted into a level control problem by using a fluid analogy. A state estimation based MPC algorithm is developed with three degree-of-freedom tuning that addresses the requirements of inventory target tracking, and forecasted and unforecasted demand fulfillment. Proof-of-concept problems are presented to demonstrate the flexibility of using MPC to generate daily decisions to track inventory targets, meet customer demands, and other associated manufacturing requirements.; The interaction between the strategic outer loop and tactical inner loop is also examined. A strategic planning module is developed using Linear Programming (LP), while the MPC based inner loop manipulates the factory starts based on the targets generated from LP. The benefits of having an MPC inner loop are particularly significant in the presence of forecast error; this is demonstrated in a representative example.; The nonlinear dependence between load and throughput times (TPT) poses a challenge for fixed linear control systems. An adaptive MPC formulation is presented to adaptively schedule the TPT in the controller model based on load in the fab. The performance improvement over fixed linear control is demonstrated with a benchmark problem in which the TPT changes substantially due to load changes.; The novel MPC algorithm presented in this dissertation is suitable not only for SCM in semiconductor manufacturing, but for any high volume discrete parts manufacturing application with long throughput time.