AVS 52nd International Symposium
    Applied Surface Science Wednesday Sessions
       Session AS-WeM

Paper AS-WeM9
Automated Peak Identification in a TOF-SIMS Spectrum

Wednesday, November 2, 2005, 11:00 am, Room 206

Session: Essential Tools for Surface Analysis
Presenter: H. Chen, College of William and Mary
Authors: H. Chen, College of William and Mary
E.R. Tracy, College of William and Mary
W.E. Cooke, College of William and Mary
M.W. Trosset, College of William and Mary
D. Malyarenko, INCOGEN Inc.
D.M. Manos, College of William and Mary
M. Sasinowski, INCOGEN Inc.
Correspondent: Click to Email
Although the high mass resolution, imaging capability and the high-throughput capability of mass fingerprint measurements have made TOF-SIMS one of the standard tools for research in surface analysis. A bottleneck is that TOF-SIMS produces very large raw data sets that must be preprocessed to identify the mass peaks for further analysis, especially when complex biological samples produce a large number of peaks. The accuracy of the mass assignment, which is critical when comparing mass fingerprints with databases, can be another limitation. Under survey conditions, the positions of the desired mass peaks are commonly not known beforehand, and TOF-SIMS peak-picking requires a procedure to distinguish mass peaks from background noise. Often, those peaks and their positions are identified manually. This introduces a subjective error due to the asymmetric peak line shape and to Poisson (counting) noise which has larger variance at larger peaks. This results in a degradation of the apparent machine performance and an inconsistency in the peak identification. We have developed an automated peak picking algorithm based on a maximum likelihood approach that effectively and efficiently detects peaks in a TOF-SIMS spectrum. The algorithm takes into account the underlying characteristic Poisson process and asymmetric peak line shape and produces maximum likelihood estimates of peak positions and amplitudes. It also simultaneously develops estimates of the uncertainties in each of these quantities. With this approach, we avoid the ambiguities involved in manual peak picking and mass assignments. We use the estimated peak positions, amplitudes and their uncertainties to align different spectra more accurately than is possible by using a few known calibrants. This precise peak summary is crucial for further multivariate analysis.