Prediction of surface microtrenching by using neural network

Silicon oxynitride films were etched in a C2F6 inductively coupled plasma. A prediction model of microtrenching depth (MD) was constructed by using a neural network and a genetic algorithm. For a systematic modeling, etching data were collected by using a statistical experimental design. The process parameters and ranges were 400–1000 W, 30–90 W, 6–12 mTorr, and 30–60 sccm for source power, bias power, pressure, and C2F6 flow rate, respectively. The root mean-squared prediction error of the constructed model was about 0.019. The model was utilized to generate 3-D plots, which were used to examine etch mechanisms under various plasma conditions. Depending on the plasma conditions, parameter effects on MD were quite different. For most of the parameter variations, MD variations were strongly related to profile angle variations. The effect of bias power on MD seems to be dominated by polymer deposition due to the variations in C2F6 flow rates maintained in the chamber.