R. B. Marinenko, J. Armstrong, D. L. Kaiser, J. J. Ritter, P. K. Schenck, C. Bouldin, J. E. Blendell, and I. Levin

Objective: To assess the accuracy of wavelength dispersive (WD) electron probe microanalysis of BST films.

Problem: BST is one of a few high dielectric materials considered for the next generation gate dielectric layers where thicknesses down to 1 nm will be required. Since the properties and performance of these materials are strongly dependent upon thickness and composition, analytical methods that can determine these characteristics accurately are critical. The electron probe microanalyzer (EPMA) was used to quantify the composition of the 5 nm to 400 nm thick BST films. Accurate and reproducible quantitative analysis of films with thicknesses greater than 20 nm can generally be done readily. But the analysis of films that are only a few nanometers thick may be more difficult to quantify with EPMA due to the low x-ray count rates resulting from the very small sample volumes.

Approach: Three commercial BST films (54, 75, and 400 nm thick) on Si or Pt/Si substrates prepared by sputtering or MOVCD were analyzed. Eight films prepared at NIST (5 nm - 100 nm thick) were also analyzed. These were prepared by spin coating onto Si wafers using a BSTAL (BST acetate lactate) solution with (Ba + Sr)/Ti = 1. EPMA was done at 8-10 different points on each specimen using a 5-micrometer diameter beam at three different excitation potentials, 8, 15, and 25 KeV or 8, 12, and 15 KeV for the thinnest films, at 100 nA current. Peak overlap corrections were made for BaLa and TiKa on the PET crystal and for SrLa and SiKa on the TAP crystal. Specimens were quantified from data taken from well-characterized microanalysis standards using two different thin-film data reduction procedures.

Results: Uncertainties in the EPMA data taken from the randomly-selected points on each specimen were approximately 2% relative or less for the six thickest films (40 nm to 400 nm), but these uncertainties increased to as much as 20% relative for films less than 20 nm thick. This indicates that the thicker films are more uniform on the micrometer scale than the thinner ones although the poorer statistics observed for the thinner films also contribute to this uncertainty. The (Ba+Sr)/Ti atomic fraction ratio was within ±1-4% relative of the 1.0 value predicted for the NIST-fabricated films except for the thinnest, the 5 nm film, that had a (Ba+Sr)/Ti of 1.17. The Ba/Sr atomic fraction ratio was not as close to the predicted 2.4 value; it varied by as much as ±15% from this value for the NIST-fabricated specimens. Thickness calculations were also consistently 20-30% less than values predicted for those measured by other techniques. The 5.7 g/cc density used for the expected composition, Ba1-xSrxTiO3 (x = 0.29), is probably incorrect. TEM analysis showed that portions of two of the films were in part amorphous, not crystalline.

Future Plans: Improvements in statistics for the thinner films will be needed for EPMA to be useful for characterizing the thinnest films. Other materials proposed for gate exide layers, such as the oxides of zirconium, hafnium, and aluminum and silicon oxynitrides may be evaluated with EPMA. A conference proceedings publication on this work is in press.

Web Contact micro@nist.gov