Nunzio Tuccitto oral presentation (FN2-Tue2-1-4)
Novel approaches to data mining in molecular depth profiling: from the simulation to sophisticated data treatment
1 University of Catania - CSGI, V.LE A. Doria 6, 95125 Catania, Italy
2 University of Catania, V.LE A. Doria 6, 95125 Catania, Italy
Since the introduction of cluster beams, allowing the acquisition of molecular depth profiles in dynamic SIMS experiments, many efforts have been devoted to the understanding of the chemical processes involved during the etching of organic materials by energetic ions. Modelling a molecular SIMS depth profile experiment of organic target involves the simulation of the in-depth distribution analysis of trace elements and molecular fragments. We recently introduced a partial differential equation describing the evolution of the concentration profile along the depth of a certain species in the target during a dynamic SIMS experiment. The intimate relationships between energy density deposited by the beam and the induced target reactions is crucial for the detailed modelling of molecular depth profiling experiments. The elucidation and quantification of these relationships would contribute to limit empiricism in the selection of experimental conditions and it could pave the way to the prediction of the best performing primary ion beam. Strictly related to the comprehensive study of the molecular depth profiles is the mining of the information contained in a molecular depth profile. Issues arising in the interpretation of a SIMS molecular depth-profile are linked to the huge amount of data obtained from a single experiment: each point of the profile corresponds to a mass spectrum, which in turn is composed by hundreds or thousands of peaks. We will present a data processing method based on the application of wavelet transform directly to a given depth-profiling raw data for their compression and noise. Compression of data using wavelet transform results in a decrease of the data size and, at the same time, in a noise reduction, allowing – in some extent – a direct and unattended application of Principal Component Analysis (PCA) to the raw data. Examples will be shown where the combination of wavelet compression with multivariate analysis improves the result of feature extraction process, allowing the mining of “hidden” information contained in large experimental data sets.