The development of novel devices, such as organic solar cells (OSC) and organic transistors (OFET) requires detailed understanding of the used materials. We are researching on how the material structures affect device properties. Accurate simulations of organic materials require computationally very time-consuming quantum chemical models. In order to keep high accuracy but reduce the computational cost, we develop machine learning methods to predict material properties (e.g., charge transfer coefficients) from as-small-as-possible training sets. Such machine learning models are used to couple simulation methods from different length scales (such as molecular dynamics, kinetic Monte Carlo, and Drift-Diffusion) to obtain more accurate simulations of device behavior. Challenges are (i) the generation of suitable training data, (ii) the development of good fingerprints, and (iii) the optimization of machine learning parameters to improve prediction accuracy.
Covered Topics: Machine Learning, Data science, Molecular Dynamics, Kinetic Monte Carlo