Use of neural network to model the FTIR spectra of PECVD silicon nitride films for cardiovascular pressure sensor applications

In this paper, the empirical process models based on neural network are applied to discover the relationship between inputs and outputs of the plasma enhanced chemical deposition (PECVD) silicon nitride process. The design of experiments are based on a 2^6-2 fractional factorial experiment with four center replicate on six factors which are 1) the SiH₄ flow rate, 2) the NH₃ flow rate, 3) the N₂ flow rate, 4) the chamber pressure, 5) the radio frequency (RF) power/distance between the wafer base and shower gas, and 6) the deposition temperature. Once these experiments are performed, different neural networks are applied to identify these six inputs to the chemical bonding information from the FTIR measurements. The best performances of neural networks for each response are selected based on the smallest prediction error. Then the three-dimensional surface plots are generated to qualitatively interpret factor effects. The corrosion testing in saline solution based on a potentiostatic measurement of an aluminum film with the protective PECVD silicon nitride is demonstrated. This measurement could be used as a tool to identify the best dense PECVD silicon nitride film in order to improve the protective performance of the existing recipe. The silicon nitride film is used as the final passivation layer of the cardiovascular pressure sensor.