Modeling of silicon oxynitride etch microtrenching using genetic algorithm and neural network

A prediction model of etch microtrenching was constructed by using a neural network. The etching of silicon oxynitride films was conducted in C"2F"6 inductively coupled plasma. The process parameters that were varied in a statistical experimental design include radio frequency source power, bias power, pressure, and C"2F"6 flow rate. The etch microtrenching was quantified from scanning electron microscope images. The prediction accuracy of optimized neural network model with genetic algorithm had a root mean-squared error of 0.03nm/min. Compared to conventional model, this demonstrates an improvement of about 32%. The constructed model was used to infer etch mechanisms particularly as a function of pressure. Roles of profile sidewall variations were investigated by relating them to the microtrenchings. The pressure effect was conspicuous at lower source power, lower bias power, or higher C"2F"6 flow rate. Microtrenching variations could be reasonably explained by the expected ion reflection from the profile sidewall. The pressure effect seemed to be strongly affected by the relative dominance of fluorine-driven etching over polymer deposition initially maintained in the chamber.