Model Predictive Control of ${\hbox{Si}}_{1-x}{\hbox{Ge}}_{x}$ Thin Film Chemical–Vapor Deposition

This paper integrates in situ robust and efficient fundamental models and noninvasive optical sensors with state-of-the-art estimation and model predictive control techniques in order to grow unusual and aggressive Si
_1-x
Ge
_x
alloy films. A model predictive controller is presented that utilizes a dynamic process model and feedback from a spectral ellipsometer to reconstruct the current state of a Si
_1-x
Ge
_x
growing film in real time. Si
_1-x
Ge
_x
films grown utilizing feedback from a spectral ellipsometer are compared to films grown using open-loop recipes, which is the current industrial practice. These films are compared quantitatively utilizing the offline characterization techniques, auger spectrometry, and secondary-ion-mass-spectrometry analysis. The model predictive controller presented in this paper detects and rejects unmeasured disturbances allowing for precise control over film qualities. In this paper, films grown utilizing feedback from an optical sensor are shown to be up to 51% truer to desired profiles, when compared with similar films grown using open-loop recipes. The experimental results presented in this paper provide the first demonstration of feedback control using online optical film measurements to grow aggressive alloy composition profiles in which flow rates of several component gases and reactor temperatures must be varied simultaneously in order to achieve the profile of interest