Toward nanofabrication with molecular building blocks

Christian Wagner¹

¹ Peter Grünberg Institut (PGI-3), Forschungszentrum Jülich, 52425 Jülich, Germany

The control of single-molecule junctions with a scanning probe microscope (SPM) not only directly benefits molecular electronics and molecular machines, but is also a stepping stone to the more ambitious goal of (supra)molecular nanofabrication. We use the versatile concept of two-contact manipulation, where the SPM tip actuates a molecule via a single chemical bond, while the surface provides a second (weaker) fixation. The biggest challenge of this approach is the lack of information about the atomic configuration of the molecule during manipulation. I present the components of an SPM-based single-molecule manipulation setup, specifically (1) a simulation with a machine-learned model of the tip-molecule-surface junction based on the message-passing neural network PaiNN [1] trained on DFT data, (2) a probabilistic search that compares experimental and simulated force gradient data to find the best molecular configuration estimate [2], and (3) an immersive virtual reality interface. A complementary approach to molecular nanofabrication is the use of an autonomous agent to replace the decision making of a human experimenter. I will describe such an agent, which is based on reinforcement learning principles and performs a nanofabrication task even in the face of large uncertainties, sparse feedback, and without further knowledge of single-molecule mechanics [3].

[1] K. Schütt et al., Proceedings of the 38th International Conference on Machine Learning, 9377 (2021) [2] J. Scheidt et al., J. Phys. Chem. C 127, 13817 (2023) [3] P. Leinen et al., Sci. Adv. 6, eabb6987 (2020)