Estimating viscosities of semiconductor-manufacturing gases

Semiconductor manufacturing demands an accurate delivery of gases to the process chamber. To achieve this, accurate gas viscosities are needed. Hence, this paper compares viscosity models applied to pure gases and operating conditions relevant to semiconductor manufacturing and develops a method to design neural-network/multilayer-perceptron models of viscosity. Overall, the perceptron models give the smallest root-mean-square errors in comparison with experimental data, followed closely by a simplified variation of well-known models. Based on these findings, this paper uses a perceptron model to give viscosities of several semiconductor-manufacturing gases that are unavailable and gives recommendations on how to estimate these viscosities.