Run-to-Run Process Adjustment Using Categorical Observations

When quality characteristics cannot be measured on a continuous scale due to either inherent or outside constraints, qualitative observations can be collected alternatively. Under this situation, most conventional run-to-run (R2R) process-control algorithms that are developed based on quantitative measurements cannot be implemented. In this paper, we develop a run-to-run control scheme that uses qualitative information for process adjustments. A two-phase modeling and adjustment strategy is introduced and demonstrated by a real example from a deep reactive ion etching (DRIE) process: model building and parameter estimation is performed in Phase I, and a latent-model control law, a categorical R2R controller, is developed for process regulation in Phase II. Simulation results show that the proposed algorithm exhibits competitive control performance, less adjustment effort, and a larger stability region than the conventional exponentially weighted moving average (EWMA) controller.