Automated analysis of low-energy inverse photoelectron spectra using machine learning

Abstract

The electron affinity (EA) of semiconductors is the energy of the conduction band edge with respect to the vacuum level and represents the electron conduction level. The EAs of organic semiconductors are most precisely determined by low-energy inverse photoelectron spectroscopy (LEIPS). However, data analysis to determine EA from a LEIPS spectrum requires skilled analysts and time-consuming procedures. In this study, we developed a machine learning-based automated analysis method. The vacuum level was determined with an accuracy of 0.01 eV by analyzing the low-energy electron transmission (LEET) spectrum using linear regression. The onset of the LEIPS spectrum was determined by applying Random Forest and Gradient Boosting as learning models to 253 experimental LEIPS spectra derived from measurements of 27 organic semiconductors as the training dataset. By implementing the smoothing process and augmenting the training data in the energy direction, we achieved a prediction accuracy of within ±0.2 eV (±0.1 eV) for 90 % (70 %) and 91 % (74 %) of the test data as the average of six partitions for the Gradient Boosting and Random Forest models, respectively. The fully automated determination of EA is realized by the combined use of LEET and LEIPS spectral onset analyses.