Quantitative Microprobe Analysis of Thin Insulating Films

Abstract

AbstractThe analysis of thin insulating films occurring in various stages of the manufacture of integrated circuits has in the past been difficult, if not impossible. Their dimensions usually preclude analysis by conventional chemical techniques, hence very little is known concerning their relative stoichiometrics. However, by microprobe analysis, films as thin as 540 Å (∼18μg/cm2) have been successfully analyzed. In general, two different approaches are taken, depending on the film thickness. Usually, films thicker than 2500 Å (∼50μg/cm2) are analyzed by conventional microprobe techniques. A computer program, called MAGIC, has been written, which converts the raw X-ray intensities to chemical composition. All data are corrected for dead time, background, absorption, atomic number effects, and fluorescence by characteristic radiation, if required. A minimum of input is required, only the chemical symbols. X-ray lines employed, and the accelerating voltage being necessary in addition to the raw X-ray intensities. All constants such as atomic weights, critical excitation potentials, X-ray wavelengths, and absorption coefficients are stored or calculated internally, which reduces the errors and time associated with looking up and key-punching these values. New fluorescent yields are used for K and L radiation. No fluorescence correction is made for M radiation. The use of this correction program generally gives results, accurate to about 2 to 4%, relative to the amount present. For films thinner than 2500 Å, a new model is proposed which allows the X-ray spectra from the film and substrate to be unfolded to give the composition of the film only. To accomplish this, the atomic number correction of Duncumb and deCasa employed in MAGIC has been used in conjunction with a mean electron energy concept. Results obtained to date through the use of this model have been surprisingly good, being of the order of 5% relative to the amount present. Potential errors and uncertainties are discussed, and results given which illustrate the accuracy of the two methods.