Organic semiconductors (OSCs) are considered as highly feasible for the use in opto- and bioelectronic applications. The relatively low charge carrier mobility limits its current applicability. Insight into electronic structure-property relations is required to understand how to enhance the mobility. Various groups have been working towards a better understanding of the dependence of the mobility both relying on chemical intuition as well as using theoretical approaches covering structural information from molecular dynamics (MD) over transport properties using kinetic Monte Carlo (kMC) [1-3]. Mentioned numerical methodologies provide insight of transport for small perturbations from a crystalline structure. However, there is a lack of description for strongly non-crystalline structures such as solution processed OCSs due to the high computational effort associated with the calculation of all individual transition rates required for precise kMC studies. In this work, we present a machine learning (ML) based ab-initio multi-scale simulation for charge transport within strongly non-crystalline OSCs using pentacene as a case study. ML allows for efficient and effective modeling of transport properties and helps to overcome conventional physical models that rely on empirical parametrizations or semi-empirical calculations. Using MD and density functional theory (DFT), we generate training data for the coupling integrals between adjacent molecules (dimers) of pentacene. The ML algorithm bases on the kernel ridge regression including an optimized set of descriptors for the respective dimers. To cover the range from the coupling integrals to charge mobilities, we couple the ML algorithm with our off-lattice kMC software [4]. The center of mass positions of the molecules within a non-crystalline pentacene film are used to generate the Voronoi tessellation which allows computing charge hopping across amorphous structures. ML helps to estimate coupling integrals within the Marcus theory without the need for a precomputation of each dimer configuration. We overcome existing ab-initio simulations by a tremendous reduction of computational demand and allow a generalized description for a wide range of structural disorder within the organic layer without a need to compute coupling integrals from scratch. We show how the directionality of the charge transport is influenced by varying the degree of crystallinity. Furthermore, we compute the field- and temperature dependence for non-crystalline pentacene layers stacked on top of a silicon oxide substrate, as used for thin-film transistors, and compare our ML-based approach with empirical kMC results.
«
Login
BibTeX