Model predictive control and its application to plasma-enhanced chemical vapor deposition

This thesis addresses the development of model predictive control (MPC) technique and its application to plasma enhanced chemical vapor deposition (PECVD)--one of the many steps in the integrated circuits (IC) manufacturing processes.;A new approach to guaranteeing asymptotic stability for MPC called stability constrained model predictive control (SCMPC) is proposed as the main result. In contrast to other approaches which impose constraints on the terminal state of the predicted trajectory or rely on a long/infinite prediction horizon, a stability constraint is computed as the maximal magnitude of the state for a controllable form realization along the entire predicted trajectory. This stability constraint is carried forward as a constraint on the magnitudes of the predicted state for the next prediction and reduced (if possible) at each stage. It is shown that the on-line optimization is always feasible and asymptotic stability is guaranteed for the case of multi-input linear time invariant (LTI) systems with full state feedback or a stable state estimation. Moreover, stability is achieved for any prediction horizon and optimization cost function. Relationships to other existing approaches are discussed also. Features of the algorithm are illustrated by several simulation examples. By augmenting the original plant model and applying the SCMPC, results obtained for a regulation problem are extended to handle output tracking and disturbance rejection control. It is shown that zero steady state error is achieved for a constant setpoint change or a constant disturbance occurrence.;One of the important features of MPC technique is its ability to incorporate input/output constraints explicitly in the on-line computation. Mixed input and output constraint handling in the SCMPC framework is also addressed in this thesis. By modifying the SCMPC slightly and making use of a pre-defined feasible stability constraint region and an admissible initial condition set, it can be shown that the constraints imposed on the system inputs and outputs can be incorporated into the SCMPC on-line computation without violating the feasibility of the optimization. It is also shown that the originally imposed stability constraint can be satisfied after an appropriate number of time steps. Thus, asymptotic stability of the constrained closed-loop system is preserved.;The applicability of the proposed SCMPC method is demonstrated by experimentally controlling a plasma enhanced chemical vapor deposition (PECVD) system which is commonly seen in the semiconductor manufacturing process industry. The results show that the process species are successfully controlled to their desired setpoints and the control performance is comparable to what can be achieved by an LQG controller. The motivation for implementing a multivariable feedback control on the PECVD system and hardware and software setups are also described. Comparisons to other control designs are made from several perspectives.