Use of a neural network to model SiC etching in a NF3 inductively coupled plasma

Abstract

Using a neural network, a plasma etching of silicon carbide thin films was modelled. The etching was conducted in an inductively coupled plasma etch system and characterized by a 24 factorial experiment. The process parameters involved include radio frequency source power, bias power, pressure and gas ratio defined as a NF3 flow rate divided by the total flow rate. The prediction performance of generalized regression neural network was optimized by means of a genetic algorithm. Etch mechanisms were qualitatively estimated by exploring predicted response surfaces. The etch rate variation with the gas ratio was strongly correlated to that of either the dc bias or expected fluorine concentration. The etch rate increase with the pressure was attributed to dominant chemical etching over polymer deposition. Typical source power effect was transparent only at higher gas ratios and higher powers, presumably due to enhanced radical concentration. The source power effect strongly depended on physical phenomena accompanied by the gas ratio variation maintained in the chamber.