A prediction model of charge density of silicon nitride (SiN) films was constructed by using a generalized regression neural network (GRNN). The SiN film was deposited by a plasma enhanced chemical vapor deposition (PECVD) system and the deposition process was characterized by means of a statistical experiment. The prediction performance of GRNN was optimized by using a genetic algorithm (GA) and yielded an improved prediction of about 63% over statistical regression model. The optimized model was utilized to qualitatively investigate the effect of process parameters under various pressures. A refractive index model was effectively utilized to validate charge density variations. For the variations in process parameters, charge density was strongly dependent on [N-H]. Effects of NH3 or SiH4 flow rates were significant only under high collision rate. Effect of pressure-induced collision rate was noticeable only at higher NH3 flow rate or lower SiH4 flow rate. (C) 2008 Elsevier B.V. All rights reserved.