End-to-End Machine Learning for Charge Transport Prediction in Heterogeneous Energy Landscapes of Optoelectronic Materials

• Applicant: Professor Dr. David Egger
• Subject Area: Theoretical Chemistry: Electronic Structure, Dynamics, Simulation
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 550206961

We propose methodological developments to theoretically predict charge transport in heterogeneous optoelectronic materials. While first-principles calculations of charge transport in disorder-free and clean systems have long been established, it was shown that they are not predictive for soft optoelectronic materials. Soft lattices imply large and anharmonic displacements of atoms and molecules already around room temperature, which is an important regime for technological applications. Furthermore, non-local structural and electronic effects, such as many-body dispersive and inter-site interactions, were shown to be important in these materials too. Theoretical modeling of charge transport in such scenarios poses a difficult multi-scale problem, for which proof-of-principles studies were demonstrated in recent years. However, today a modeling of heterogeneous optoelectronic materials can only be achieved on a case-by-case basis because of extensive numerical parametrizations and huge computational costs involved in current approaches. Consequently, further important sources of heterogeneity, such as point defects or interfaces, are essentially out of reach for present-day theoretical schemes for charge-transport calculations.Our main goal is to develop an accurate and predictive machine learning approach for charge-transport calculations of heterogeneous optoelectronic materials. Machine learning has empowered accelerated modeling of molecules and materials and can aid the theoretical techniques involved in charge-transport calculations, including inexpensive molecular dynamics simulations at ab initio quality. However, lack of a unified machine learning scheme for predicting charge transport in heterogeneous materials marks an important gap in the field. We propose to develop a seamless and general architecture that integrates the theoretical layers for computing structural, electronic, nonadiabatic and charge-transport properties of optoelectronic materials. Enabled by our complementary expertise we will combine state-of-the-art materials modeling approaches with latest machine learning developments. This will allow for addressing challenging theoretical problems, such as quantifying the impact of complex dynamical effects on carrier mobilities in defective materials. To this end, CLARINET focuses on a well-defined set of organic and molecular semiconductors and their ubiquitous structural inhomogeneities including defects, grains and interfaces. The final aim of our project is an end-to-end machine learning architecture for predicting spectroscopic and transport characteristics of these systems. Implementation and demonstration of our methods will establish an important step in the development of a robust, predictive and efficient computational methodology in the broader field of materials modeling.

• DFG Programme: Research Grants
• International Connection: Belgium, Luxembourg
• Partner Organisation: Fonds National de la Recherche; Fonds National de la Recherche Scientifique - FNRS
• Cooperation Partners: Professor Dr. David Jacques Beljonne; Professor Dr. Alexandre Tkatchenko