Neural network modeling of PECVD silicon nitride films

摘要

In this paper a neural network based technique has been developed to model a plasma enhanced chemical vapor deposition (PECVD) silicon nitride process. The study covers the range of normal input parameters used for PECVD silicon nitride films. These film compositions range from nitrogen-rich to silicon-rich including stoichiometric. This study emphasizes on modeling the process and is application independent. The purpose of this model is to predict the deposition rate and refractive index with joint variation of four process parameters viz., rf power, silane:ammonia gas flow-ratio, pressure and substrate temperature. Two separate networks have been used to predict the two outputs. The training data-sets for the networks has been generated by designing the experiments with the help of factorial design technique. The response surface and contour plots, generated by the model, are conforming to the physics of the process.