Optimal etch time control design using neuro-dynamic programming

This paper focuses on using a new learning algorithm, namely neural dynamic programming (NDP), to design the optimal etch time control system for a reactive ion etch process. First a predictive neural network model is built. This model represents the relation between some state variables and the resulting thickness remain. The NDP is employed to determine the optimal etch time based on the predictive film thickness remain model. Simulation results show that NDP is a viable learning optimization tool. The controlled film thickness remains have smaller variances in a few tested lots of wafers than those measured from 89 wafers during production.