Effects of noise level in fitting in-situ optical reflectance spectroscopy data

In an effort to increase the accuracy of the parameters extracted from in situ optical reflectance spectroscopy (ORS) data, the effects of noise and curve-fitting methodology are investigated. Parameters such as the growth rate and the index of refraction of growing films are acquired by fitting ORS data, and the accuracy of those parameters is determined by factors such as noise level and uncertainty in the absolute calibration of the magnitude of the reflectance. This uncertainty is represented by an overall scaling factor in modeling the reflectance with the virtual interface model. Experimentally, the scaling factor is included to adjust for drift in photodetector temperatures, the effects of ORS light source stability, and transparency changes of the window of the growth chamber. Our goal is to determine index of refraction n to an accuracy of 0.0015, or about 0.044 % for Al/sub 0.5/Ga/sub 0.5/As at growth temperature, so that the Al mole fraction x can be extracted to x = /spl plusmn/ 0.002 for Al/sub x/Ga/sub 1-x/As. Curves of reflectance as a function of time were generated using parameters so that the simulated data resemble actual ORS data. Gaussian noise was added to the generated curve. The simulated data curves were fit by means of a Levenberg-Marquardt nonlinear algorithm. The fitting results of different noise levels show that the accuracy of curve fitting depends on noise level and the uncertainty range of the scaling factor.