A comparison of statistically-based and neural network models of plasma etch behavior

A neural network modeling methodology is applied to the removal of polysilicon films by plasma etching. For a polysilicon etch in a CCl₄/He/O₄ plasma, the etch rate, uniformity, and selectivity to both silicon dioxide and photoresist were modeled as a function of RF power, pressure, electrode spacing, and the three gas flows. Neural process models were subsequently compared to models derived by response surface methodology (RSM) for the same data. It was demonstrated that the neural models possess significantly superior performance. Furthermore, the derivation of accurate neural models was shown to require fewer training experiments. As a result, neural network modeling promises to be a faster, more efficient, and less expensive method of process characterization