Process Model Construction and Optimization Using Statistical Experimental Design

摘要

A methodology is presented for the construction of process models by the combination of physically based mechanistic modeling and statistical experimental design in order to create smart response surfaces. In contrast to the process independent polynomial fit of the conventional response surface method, smart response surfaces derive their basic shape from the process physics and are then calibrated using designed experiments. This method provides for a surface of better representational accuracy using the same of fewer experimental points. This method has been applied to the development of a model for the low pressure chemical vapor deposition (LPCVD) of polysilicon, a process used in the manufacture of VLSI circuits. A one-dimensional finite difference model of the LPCVD process was constructed. A Taguchi orthogonal array experiment was conducted. A confirming experiment performed at the parameter levels indicated by the Taguchi optimization, served to confirm the validity of the experimental procedure. The experimental results will subsequently be used to calibrate the mechanistic model.