TESTING OF A RAPID FAULT-DETECTION MODEL FOR QUALITY-CONTROL - BOROPHOSPHOSILICATE GLASS THIN-FILMS MONITORED BY INFRARED-ABSORPTION SPECTROSCOPY

Infrared absorption spectra of 108 borophosphosilicate glass (BPSG) th in films produced in a multiple-wafer low-pressure chemical vapor depo sition (LPCVD) reactor were collected to enable the development and te sting of a rapid and inexpensive method for determining if films are w ithin the desired specifications. Classification of samples into good and bad product categories was made by applying principal component an alysis to the spectra. Mahalanobis distances were used as the classifi cation metric. The highest overall percentage of correct classificatio n of samples based upon their spectra with two-step classification was 95%. The misclassified samples were, however, within the error of the reference methods that were used in making the original classificatio n against which the infrared (IR) classification methods were tested. The classification errors are thus just as likely to be a result of mi sclassification by the reference method rather than errors by the IR c lassification. Although reference measurements were used in this artic le for the original classification of the samples, these expensive and time-consuming reference methods can be eliminated simply by building classification models on samples determined to produce a product with in the correct device specifications. The IR classification methods pr esented here hold great promise as a tool for rapid quality control of BPSG deposition. (C) 1997 American Vacuum Society.