Deep-learning analysis of simulated RBS spectra of impurities in SiO2 layers: data, trained networks and code

Simulated and measured Rutherford backscattering spectrometry (RBS) spectra, trained neural networks, results and code of the article "Deep-learning analysis of simulated Rutherford backscattering spectra of impurities in SiO2 layers: accuracy, physical limits and robustness" (S. Unal, submitted to Measurement Science and Technology). The record contains 145,835 RUMP spectra of SiO2 layers with one impurity element (66 elements, 10-1,960 nm, X/Si ratio 0.001-0.101; 3.0 MeV 4He, 171°), new simulations with off-grid values, unseen elements and buried layers, a library of 13,886 spectra of Au in SiO2 for 2.0 MeV 4He at 166.2°, four measured spectra of Au-implanted quartz, the trained networks (MLP, plain and residual CNN, CNN-GRU, transformer) and the training and analysis code. Thicknesses in the data files are in angstrom (see README.md).