Use of Neural Network to Control a Refractive Index of SiN Film Deposited by Plasma Enhanced Chemical Vapor Deposition

Abstract

Refractive index of silicon nitride film deposited in a plasma enhanced chemical deposition system is modeled by using neural network. The deposition process was characterized with a full factorial experiment. Additional 12 experiments were conducted to test model appropriateness. Predicted model behaviors were in good agreement with actual measurements. Deposition mechanisms were qualitatively examined especially with respect to the pressure. Possible interactions between the pressure and other factors (SiH4, NH3, radio frequency (RF) power, and substrate temperature) were examined on the basis of SiH/NH bond ratio. The refractive index increased with increasing either SiH4 flow rate or RF power. In contrast, the refractive index decreased with increasing NH3 flow rate. Little interactions between the pressure and RF power were observed. Pressure effect on the refractive index was quite different depending on the level of SiH4 flow rate or substrate temperature. In general, increasing the pressure increased the refractive index. Meanwhile, the refractive index was insensitive to parameter variations at relatively high pressures. The most complex interaction occurred as the pressure interacted with the temperature. Useful clues to control the refractive index were revealed from the predictive model.