Machine Learning Prediction of Electronic Structure for High-Throughput Inverse Design of Functional Organic Materials

Materials for electronic devices such as organic light emitting diodes need to have tailored optoelectronic properties and be synthetically viable. As devices are composed of organic thin films with different functionality, the optoelectronic properties of materials need to be designed in concert and the role of the interfaces between thin films must be understood. Computational high-throughput screening of molecular excited states can greatly facilitate this complex multi-objective design problem. I will present our recent work on deep machine learning models that are able to predict structure, electronic structure, and excited states of organic molecules and materials in general. Our models predict optical excitations, the fundamental gap, electron affinity and ionisation potential for large organic molecules of diverse composition. [1] While the models are trained on first principles electronic structure data, the prediction process requires no recourse to computationally expensive ab initio calculations. The accuracy and transferability of the models can be assessed against photoemission spectroscopy data. I will further showcase how such models can be used in combination with generative machine learning to discover novel organic compounds that satisfy specific optoelectronic properties. [2] We explore different applications with our approach from the design of organic electronics to plasmonic sensors and tailored nanoparticles. [1] Chem. Sci. 12, 10755-10764 (2021), [2] Nature Comp. Sci. 3, 139–148 (2023);