Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 71: Mini-Symposium: Machine learning applications in surface science I
O 71.4: Vortrag
Mittwoch, 3. März 2021, 14:45–15:00, R1
Active Discovery of Organic Semiconductors — •Christian Kunkel1,2, Johannes T. Margraf1, Ke Chen1, Harald Oberhofer1, and Karsten Reuter1,2 — 1Chair for Theoretical Chemistry and Catalysis Research Center — 2Fritz-Haber Institut der Max-Planck-Gesellschaft
Improving charge-transport of organic semiconductors (OSCs) for electronic applications is usually tackled by empirical structural tuning of promising compounds. Howver, the versatility of organic molecules generates a rich design space whose vastness dictates efficient search strategies. We thus here present an active machine learning (AML) approach that explores this virtually unlimited design space iteratively. Judging suitability of OSC candidates by charge injection and mobility-related descriptors, the AML approach iteratively queries first-principle evaluation on well-selected molecules. We first optimize the approach in a fully characterized, but truncated molecular test space, gaining deep methodological insight about its exploratory behavior. Outperforming a conventional computational funnel, the devised algorithm can thereby successfully leverage its gradually improving knowledge and focus on promising regions of the design space. When subsequently lifting the artificial truncation, high-performance candidates are constantly found while the algorithm meanders ever more deeply through the endless OSC design space. The demonstrated high efficiency in the detection of candidate compounds with superior charge conduction properties highlights the usefulness of autonomously operating systems for a targeted OSC design.