Neural network-based modeling of the plasma-enhanced chemical vapor deposition of silicon dioxide

Abstract

The properties of plasma enhanced chemical vapor deposition (PECVD) silicon dioxide films are modeled using neural networks. This method is simple, extremely useful and readily applicable to the empirical modeling of such complex plasma processes. In characterizing the SiO/sub 2/ films, it is found that the dominant film property is its impurity concentration. The impurity concentration dictates the refractive index and permittivity, two critical figures of merit when these films are used as interlayer dielectric and in optoelectronic applications. The most important parameters in determining the impurity concentration of the films are substrate temperature and pressure. Increasing the substrate temperature causes the impurity concentration to decrease. This drop in impurity concentration causes an increase in refractive index and a decrease in permittivity. Increasing pressure has almost the same effect, causing a decrease in permittivity.<>