Predicting Molecular Self-Assembly on Metal Surfaces Using Graph Neural Networks Based on Experimental Data Sets

The application of supramolecular chemistry on solid surfaces has received extensive attention in the past few decades. To date, combining experiments with quantum mechanical or molecular dynamic methods represents the key strategy to explore the molecular self-assembled structures, which is, however, often laborious. Recently, machine learning (ML) has become one of the most exciting tools in material research, allowing for both efficiency and accuracy in predicting molecular properties. In this work, we constructed a graph neural network to predict the self-assembly of functional polycyclic aromatic hydrocarbons (PAHs) on metal surfaces. Using scanning tunneling microscopy (STM), we characterized the self-assembled nanostructures of a homologous series of PAH molecules on different metal surfaces to construct an experimental data set for model training. Compared with traditional ML algorithms, our model exhibits better predictive performance. Finally, the generalization of the model is further verified by comparing the ML predictions and experimental results of different functionalized molecule. Our results demonstrate training experimental data sets to produce a predictive ML model of molecular self-assembly with generalization performance, which allows for the predictive design of nanostructures with functional molecules.

Two-dimensional self-assembled nanostructures of nucleobases and their related derivatives on Au(111)

The construction of two-dimensional (2D) self-assembled nanostructures has been one of the considerably interesting areas of on-surface chemistry in the past few decades, and has benefited from the rapid development and improvement of scanning probe microscopy techniques. In this research field, many attempts have been made in the controllable fabrication of well-ordered and multifunctional surface nanostructures, which attracted interest because of the prospect for artificial design of functional molecular nanodevices. DNA and RNA are considered to be programmable self-assembly systems and it is possible to use their base sequences to encode instructions for assembly in a predetermined fashion at the nanometer scale. As important constituents of nucleic acids, nucleobases, with intrinsic functional groups for hydrogen bonding, coordination bonding, and electrostatic interactions, can be employed as a potential system for the versatile construction of various biomolecular nanostructures, which may be used to structure the self-assembly of DNA-based artificial molecular constructions and play an important role in novel biosensors based on surface functionalization. In this article, we will review the recent progress of on-surface self-assembly of nucleobases and their derivatives together with different reactants (e.g., metals, halogens, salts and water), and as a result, various 2D surface nanostructures are summarized.

Construction of Two-Dimensional Chiral Networks through Atomic Bromine on Surfaces

Using atomic bromine and 2,6-diphenylanthracene (DPA), we successfully constructed and characterized the large-area 2D chiral networks on Ag(111) and Cu(111) surfaces by combining molecular beam epitaxy with scanning tunneling microscopy. The Br atoms distribute themselves periodically in the network with the maximum number of -C-H···Br hydrogen bonds. Density functional theory calculations demonstrate that the hydrogen bonds contribute to the stability of the Br-organic networks. In addition, by controlling the ratio of bromine atoms to DPA molecules, different patterns of Br-organic networks were obtained on Ag(111) surfaces. Further experiments with 2,6-di(4-cyclohexylphenyl)anthracene on Ag(111) produced analogous atomic bromine guided 2D chiral networks.

Dehalogenative Homocoupling of Terminal Alkynyl Bromides on Au(111): Incorporation of Acetylenic Scaffolding into Surface Nanostructures

On-surface C-C coupling reactions of molecular precursors with alkynyl functional groups demonstrate great potential for the controllable fabrication of low-dimensional carbon nanostructures/nanomaterials, such as carbyne, graphyne, and graphdiyne, which demand the incorporation of highly active sp-hybridized carbons. Recently, through a dehydrogenative homocoupling reaction of alkynes, the possibility was presented to fabricate surface nanostructures involving acetylenic linkages, while problems lie in the fact that different byproducts are inevitably formed when triggering the reactions at elevated temperatures. In this work, by delicately designing the molecular precursors with terminal alkynyl bromide, we introduce the dehalogenative homocoupling reactions on the surface. As a result, we successfully achieve the formation of dimer structures, one-dimensional molecular wires and two-dimensional molecular networks with acetylenic scaffoldings on an inert Au(111) surface, where the unexpected C-Au-C organometallic intermediates are also observed. This study further supplements the database of on-surface dehalogenative C-C coupling reactions, and more importantly, it provides us an alternative efficient way for incorporating the acetylenic scaffolding into low-dimensional surface nanostructures.

The stereoselective synthesis of dienes through dehalogenative homocoupling of terminal alkenyl bromides on Cu(110)

We have successfully achieved the stereoselective synthesis of a specific cis-diene moiety through a dehalogenative homocoupling of alkenyl bromides on the Cu(110) surface, where the formation of a cis-form organometallic intermediate is the key to such a stereoselectivity as determined by DFT calculations.

Confined Synthesis of Organometallic Chains and Macrocycles by Cu-O Surface Templating

The bottom-up construction of low-dimensional macromolecular nanostructures directly on a surface is a promising approach for future application in molecular electronics and integrated circuit production. However, challenges still remain in controlling the formation of these nanostructures with predetermined patterns (such as linear or cyclic) or dimensions (such as the length of one-dimensional (1D) chains). Here, we demonstrate that a high degree of structural control can be achieved by employing a Cu(110)-(2×1)O nanotemplate for the confined synthesis of organometallic chains and macrocycles. This template contains ordered arrays of alternating stripes of Cu-O chains and bare Cu, the widths of which are controllable. Using scanning tunneling microscopy and low-energy electron diffraction, we show that well-defined, ordered 1D zigzag organometallic oligomeric chains with uniform lengths can be fabricated on the Cu stripes (width >5.6 nm) of the Cu(110)-(2×1)O surface. In addition, the lengths of the meta-terphenyl (MTP)-based chains can be adjusted by controlling the widths of the Cu stripes within a certain range. When reducing the widths of Cu stripes to a range of 2.6 to 5.6 nm, organometallic macrocycles including tetramer (MTP-Cu)4, hexamer (MTP-Cu)6, and octamer (MTP-Cu)8 species are formed due to the spatial confinement effect and attraction to the Cu-O chains. An overview of all formed organometallic macrocycles on the Cu stripes with different widths reveals that the origin of the formation of these macrocycles is the cis-configured organometallic dimer (MTP)2Cu3, which was observed on the extremely narrow Cu stripe with a width of 1.5 nm.