<title>Etch process characterization using neural network methodology: a case study</title>

Abstract

Polysilicon etching in a single-wafer, parallel-plate, magnetically- enhanced RIE tool has been examined using two different approaches to the non-physical modeling of the system characteristics. The behavior of both process responses (polysilicon and oxide etch rates) and plasma parameters (voltage and current metrics) have been examined as a function of five variables (rf power, pressure, magnetic field, gas flow rate, and He backside cooling). The variable-response mapping was examined using both neural network and response surface approaches. The greater fitting power of the former method is demonstrated in a side-by-side, internally consistent comparison of the same data set using these two approaches.