Modelling plasma etching of AI thin films using neural network

Abstract

Plasma etching of aluminium thin films is characterised using a neural network. For this, etch experiments were designed by means of a statistical experimental design. Relationships between process parameters and etch rate were captured by a back propagation neural network (BPNN). Thepredicted performance of the BPNN model was optimised as a function of training factors. Model predictions were experimentally validated. Parameter effects were examined under a variety of plasma conditions. Radio frequency (rf) power affected the etch rate in different ways, depending onthe Cl₂ flow rate. The etch rate variation with rf power (or Cl₂ flow rate) was conspicuous only at higher Cl₂ flow rates (or rf power). The noticeable effect of BCl₃ flow rate at lower N₂ flow rate was attributed to an increased concentrationof bombarding ions.