Quantitative determination of dielectric thin-film properties on product wafers using infrared reflection-absorption spectroscopy
作者:
Thomas M. Niemczyk,
Lizhong Zhang,
David M. Haaland,
Kenneth J. Radigan,
期刊:
Journal of Vacuum Science&Technology A: Vacuum, Surfaces, and Films
(AIP Available online 1998)
卷期:
Volume 16,
issue 6
页码: 3490-3494
ISSN:0734-2101
年代: 1998
DOI:10.1116/1.581508
出版商: American Vacuum Society
关键词: THIN FILMS;BOROPHOSPHATE GLASS;SILICATES;DIELECTRIC PROPERTIES;CALIBRATION;INFRARED SPECTRA;ABSORPTION SPECTRA;NONDESTRUCTIVE TESTING;REFLECTION;borophosphosilicate glass
数据来源: AIP
摘要:
Process monitoring of borophosphosilicate glass (BPSG) dielectric thin films used in the manufacture of microelectronic devices is currently performed using multivariate calibration models developed from transmission infrared (IR) spectra of the films deposited on undoped monitor Si wafers. It would more be desirable to monitor the BPSG deposition on the actual product or device wafers. Because product wafers are opaque in the IR, reflection rather than transmission spectroscopy must be used to monitor the BPSG films deposited on product wafers. In this article, we demonstrate, for the first time, that IR reflection spectra of product wafers can be used to monitor the boron and phosphorus contents of the film and the film thickness to a precision that is comparable to that found for IR analysis of BPSG deposited on monitor wafers. The cross-validated standard errors of prediction of 0.11 wt %, 0.11 wt %, and 3 nm for B, P, and thickness, respectively, were achieved using multivariate partial least squares (PLS) models applied to the IR reflectance spectra obtained from reference product wafers. The prediction abilities were found to be independent of the position of the infrared spectrum on a given device and independent of which devices of the same structure were examined. The multivariate calibration models could be used to predict the B and P contents and film thickness of BPSG on device structures of different types and even for devices of different feature sizes if the PLS models were adjusted for slope and intercept differences. Therefore, the time and expense of generating calibrations for new device structures can be greatly reduced by the use of a small number of reference samples of the new devices to estimate the required slope and intercept adjustments for the models.
点击下载:
PDF
(85KB)
返 回