Berlin 2024 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 57: New Methods: Theory
O 57.5: Vortrag
Mittwoch, 20. März 2024, 17:15–17:30, MA 043
Adapting Explainable Machine Learning to Study Mechanical Properties of 2D Hybrid Perovskites — •Yuxuan Yao1,2, Dan Han3,4, Hubert Ebert4, Aron Walsh5, David O. Scanlon3, and Harald Oberhofer2 — 1Chair for Theoretical Chemistry, Technical University of Munich — 2Chair for Theoretical Physics VII, University of Bayreuth — 3School of Chemistry, University of Birmingham — 4Department of Chemistry and Center for NanoScience, University of Munich — 5Department of Materials, Imperial College London
The fast increase of power conversion efficiency (PCE) and low-cost preparation of lead-based halide perovskite photovoltaics are of great interest for use in optoelectronic devices. 2D hybrid organic and inorganic perovskites (HOIPs) have been used as capping layers on top of 3D perovskites to increase the stability and PCE. On the other hand, the soft and stable HOIPs are attractive in sustaining flexible electronic devices. In our work, we utilize explainable machine learning (ML) techniques to accelerate the in silico prediction of elasticities of 2D perovskites, as indicated by their Young’s moduli. Our ML models allow us to distinguish between stiff and nonstiff HOIPs and to extract the materials’ features most strongly influencing the Young’s modulus. The Pb-halogen-Pb bond angle emerges as the dominant physical feature with an inverse correlation to the structural non-stiffness. Furthermore, the cations’ steric effect index (STEI) was found to yield rough estimates of non-stiffness. Finally, the deformation of the octahedra strongly affects the mechanical properties, allowing us to perform transfer learning from single layered to multi-layered 2D perovskites.
Keywords: Explainable Machine Learning; Perovskites; Mechanical Properties