Modeling and optimization of plasma chemical etching of polysilicon in HBr/Cl2 mixture with data-driven approaches

A complex model of the plasma chemical etching process, which describes both the physicochemical properties of plasma and heterogeneous processes on the surface was proposed. The densities of plasma species, their energies and the surface fluxes of ions and chemically active particles using a 0-dimensional model were calculated. The resulting etching profile using cellular automata model and Monte Carlo method was calculated. Pressure, the power supplied to the plasma, the composition of the plasma-forming mixture, and the bias were used as control parameters. The outputs were etching rate, selectivity, and anisotropy. Model verification was based on experimental data on polysilicon etching in HBr/Cl₂ plasma. Initially unknown probabilities of heterogeneous processes, which provided the agreement between model and experiment in respect to etching rate were determined. A machine learning model that makes it possible to predict the shape of the etching profile based on the set of control parameters was proposed. The architecture based on a combination of polynomial regression and four-stage preprocessing of the etching profile was implemented. A method of optimal dataset generation for training the above-described model was described.