Report on deep learning approaches to analyse complex RBS/EBS spectra

Deep learning algorithms find widespread use in various fields where modelling algorithms are unsuitable In the domain of ion beam analysis and ion beam modification of materials we have identified two fields where these artificial intelligence approaches will particularly enhance the performance of ion beam methods One field of application is to solve the inverse problem for Rutherford backscattering spectrometry RBS where one has to solve the ambiguity of elemental and depth information mixed together in a single energy spectrum of backscattered particles These data may be supported by multiple technique approach where additional information is obtained from complementary analysis techniques KUL SUR By training the system with the large set of already measured and wellknown samples one can expect the evaluation of new measured energy spectra by the deep learning algorithms This will allow an easier analysis of energy spectra even by users that are not as experienced in data evaluation Using advanced learning methods and network architectures as well as by simultaneously combining spectra from multiple detectors WP21 T223 this approach will be implemented for more challenging measurements eg where signals overlap where the spectra are very noisy due to a low count rate or where the samples exhibit extreme roughness Finally we aim at implementing such networks in elastic backscattering spectrometry EBS where the contribution of nonelastic scattering events makes the operatorbased spectrum analysis very tedious and timeconsuming KUL JYU IST SUR