From Experimental Values to Predictive Models: Machine Learning-Driven Energy Level Determination in Organic Semiconductors

Abstract

The precise determination of ionization energy (IE) and electron affinity (EA) is crucial for the development and optimization of organic semiconductors (OSCs). These parameters directly impact the performance of organic electronic devices. Experimental techniques to measure IE and EA, such as UV photoelectron spectroscopy (UPS) and low-energy inverse photoelectron spectroscopy (LE-IPES), are accurate but resource-intensive and limited by their availability. Computational approaches, while beneficial, often rely on gas-phase calculations that fail to capture solid-state phenomena, leading to discrepancies in practical applications. In this work, machine learning methods are used to develop a chained model for estimating solid-state IE and EA values. By implementing a transfer learning strategy, the challenge of limited experimental data is effectively addressed, utilizing a large database of intermediate properties to enhance model training. The efficacy of this model is demonstrated through its performance achieving mean absolute errors of 0.13 and 0.14 eV for IE and EA, respectively. The model has also been tested on an external validation dataset comprising newly measured molecules. These findings highlight the potential of machine learning in OSC research, significantly enhancing property accessibility and accelerating molecular design and discovery.