NEURAL PATTERN-RECOGNITION APPLIED TO AES DEPTH PROFILING

Neural pattern recognition was used to analyse the low-energy Auger sp ectra of a thermally annealed Si/Ni/Si layered structure measured duri ng the acquisition of a depth profile. The purpose was to gain informa tion about the chemical state of the elements at the interfaces by pro cessing the data in a way quite similar to conventional target factor analysis (TFA). The new approach, however, has some important advantag es: no standards are required, it is extremely fast and it is fully au tomatic, In principle, there is only one arbitrary parameter, the vigi lance parameter rho, which sets a threshold for the level of similarit y required for assuming two spectra as belonging to the same class of data. However, the requirement that the optimal value for rho should c orrespond to the maximal correlation between the experimental data set and the recalculated spectra makes the system also robust against mis conclusions based on subjective interpretation of the data set, which is not always the case in TFA.