Neural network modeling of inter-characteristics of silicon nitride film deposited by using a plasma-enhanced chemical vapor deposition

Neural network have been widely used to model a relationship between process parameters (or in situ diagnostic variables) and film qualities. A new neural network model relating inter-relationship between the film qualities, not the process parameters is constructed by using a generalized regression neural network and a genetic algorithm. This approach is applied to the lifetime of silicon nitride films deposited by using a plasma-enhanced chemical vapor deposition system. The lifetime is an important quality that determines the efficiency of solar cells. The other film qualities examined are a deposition rate, a refractive index, and a charge density. For a systematic modeling, the deposition process was modeled by using a statistical experiment. Compared to conventional and statistical regression models, the optimized GRNN model demonstrated an improvement of 73% and 81%, respectively. The model predicted important and useful clues to optimizing the lifetime. It is noticeable that higher lifetime was achieved at lower deposition rate. This was also noted as the charge density was decreased. The refractive index played a critical role in improving the lifetime.

► A new neural network model of inter-relationships between thin film qualities was constructed using a generalized regression neural network and a genetic algorithm. Lifetime of silicon nitride films deposited using a plasma-enhanced chemical vapor deposition system was modeled. Other film qualities involved include a deposition rate, a refractive index, and a charge density. Compared to conventional and statistical regression models, the optimized model demonstrated an improvement of 73% and 81%, respectively. Useful information such as the higher lifetime at the lower deposition rate could be predicted.