Berlin 2018 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
O: Fachverband Oberflächenphysik
O 80: Poster: Scanning Probe Techniques - Method Development
O 80.19: Poster
Wednesday, March 14, 2018, 18:15–20:30, Poster A
Reinforcement learning for automatic SPM-based single molecule manipulation — •Philipp Leinen1,2, Kristof Schütt3, Klaus-Robert Müller3, Ruslan Temirov1,2, F. Stefan Tautz1,2, and Christian Wagner1,2 — 1Peter Grünberg Institut (PGI-3), Forschungszentrum Jülich, Germany — 2JARA-Fundamentals of Future Information Technology, Jülich, Germany — 3Institut für Softwaretechnik und Theoretische Informatik, Technische Universität Berlin, Germany
To realize the vision of molecular electronics, the precise control of molecular conformations is critical. The scanning probe microscope (SPM) is the tool of choice for the controlled manipulation of single molecules on surfaces. Finding the right manipulation protocol for a given task can be complicated due to the large number of degrees of freedom of such systems. The computationally demanding search for correct manipulation trajectories via simulation can be avoided if a clear manipulation goal can be formulated. In this case the experimenter can learn to achieve this goal in a trial and error approach [1,2]. Here we go one step further and delegate the learning process to a computer. We use reinforcement learning to train a neural network for the task of removing individual PTCDA (3,4,9,10-perylene-tetracarboxylicdianhydride) molecules from a monolayer on Ag(111). The feasibility of our approach is proven using simulated manipulation and first steps are shown of how to employ the method in an experimental SPM setup. [1] M. F. B. Green et al. Beilstein J. Nanotechnol. 5, 1926 (2014) [2] P. Leinen et al. J. Vis. Exp. (116), e54506 (2016)