| AVS 55th International Symposium & Exhibition | |
| Applied Surface Science | Tuesday Sessions |
| Session AS-TuA |
| Session: | Practical Surface Analysis |
| Presenter: | M. Suzuki, ULVAC-PHI, Inc., Japan |
| Authors: | M. Suzuki, ULVAC-PHI, Inc., Japan S. Fukushima, NIMS, Japan S. Tanuma, NIMS, Japan |
| Correspondent: | Click to Email |
XPS is widely used for the analysis of surface regions to determine elemental composition and chemical state. Peak identification is usually conducted with sophisticated software that is delivered from a manufacturer. In the VAMAS/TWA2/A9 project, we have artificially prepared basic test XPS spectra, corresponding to Au-Ag-Cu ternary alloys with three different compositions, and we also superposed noise onto them. Noise was generated by random number with amplitudes that were numerically defined from total intensity of the basic test spectra.1 Three basic and noise-superposed spectra were distributed to participants in a round robin test (RRT) to evaluate visual peak detection and peak identification using software in daily use. We will demonstrate peak identification efficiency, referring the results for visual peak detection2 for the test spectra. We have, first, prepared the reference sets of peak energies and elemental transitions, where we have chosen the peaks with detectabilities more than 75% (9 persons / 12 RRT participants) or with S/N (background-subtracted peak intensity / background deviation at both peak sides) ratios more than 10. The nine persons among twelve RRT participants reported results of automated peak identification for noise-superposed test spectra. 6, 2, and one persons used software of MultiPak (ULVAC-PHI), Casa XPS (Casa Software), and Spectral Data Processor (XPS International), respectively. Analyst A (MultiPak user) showed about 60% of “positive efficiency (PE)” and several % of “negative efficiency (NE)”. Here PE means the ratio of the number of identified peaks to that in a set of reference, but NE corresponds to the percentage of peak number not-included in a set of reference. Analyst J (Casa user) reported about 70% PE and 100% NE for the same test spectrum. For one other noisy test spectrum, analysts A and J answered about 60% and 70% PE, and 5-10% and 45-50% NE, respectively. These efficiencies did not depend on the noise amplitudes. The elements assigned by analyst A were only Au, Ag, and Cu, but analyst J reported several other elements. The results evidently prove the difference in design concepts of software. In the presentation, efficiencies depending on software and noise amplitude will be discussed in details.
1 M. Suzuki, S. Fukushima, and S. Tanuma, J. Surf. Anal. 14, 104 (2007).
2 M. Suzuki, S. Fukushima, and S. Tanuma, Surf. Interface Anal. Accepted.