Modeling electron cyclotron resonance etching of Ba(Mg1/3Ta2/3)O3 films using neural networks

Complex perovskite materials, Ba(Mg1/3Ta2/3)O3 (BMT) and Ba(Zn1/3Ta2/3)O3, have been reported to have excellent microwave dielectric properties. Using neural networks, in this study, the etching characteristics of BMT films are modelled, which was etched in a SF6/Cl2-based electron cyclotron resonance plasma. The process was characterized by a 23 factorial experiment, in which the factors involved include source power, bias power, and the ratio of SF6 flow rate to Cl2. Two etch responses were modelled, which include the etch rate of BMT and DC bias. Behaviors of neural etch model were in good agreement with experimental results. It was found that Cl species generated from the Cl2/SF6-gas plasma play a dominant role in etching BMT films. By correlating etch rate to DC bias voltage, it was revealed that rather than the DC bias, etch species seemed to have a significant impact on the etch rate.