AI Breakthrough in Nanotechnology Shatters Limits of Precision

At TU Graz, a pioneering research group is leveraging artificial intelligence to drastically enhance the way nanostructures are constructed.

They aim to develop a self-learning AI system that can autonomously position molecules with unprecedented precision, potentially revolutionizing the creation of complex molecular structures and quantum corrals for advanced electronics.

Revolutionizing Nanostructure Construction with AI

The properties of a material are often shaped less by its chemical composition and more by how its molecules are arranged within the atomic lattice or on its surface. Materials scientists harness this principle by positioning individual atoms and molecules on surfaces using high-performance microscopes. However, this process is highly time-consuming, and the resulting nanostructures remain relatively simple.

A research group at TU Graz aims to revolutionize this approach with artificial intelligence. “We want to develop a self-learning AI system that positions individual molecules quickly, specifically and in the right orientation, and all this completely autonomously,” says Oliver Hofmann from the Institute of Solid State Physics, who heads the research group. This advancement could enable the construction of highly complex molecular structures, including nanoscale logic circuits.

The research group, called “Molecule Arrangement through Artificial Intelligence,” has secured €1.19 million ($1.23 million)  in funding from the Austrian Science Fund to turn this vision into reality

Advanced Techniques in Molecular Positioning

The positioning of individual molecules on a material’s surface is carried out using a scanning tunneling microscope. The tip of the probe emits an electrical impulse to deposit a molecule it is carrying. “A person needs a few minutes to complete this step for a simple molecule,” says Oliver Hofmann. “But in order to build complicated structures with potentially exciting effects, many thousands of complex molecules have to be positioned individually and the result then tested. This of course takes a relatively long time.”

AI Integration for Enhanced Precision

However, a scanning tunneling microscope can also be controlled by a computer. Oliver Hofmann’s team now wants to use various machine learning methods to get such a computer system to place the molecules in the correct position independently. First, AI methods are used to calculate an optimal plan that describes the most efficient and reliable approach to building the structure. Self-learning AI algorithms then control the probe tip to place the molecules precisely according to the plan.

“Positioning complex molecules at the highest precision is a difficult process, as their alignment is always subject to a certain degree of chance despite the best possible control,” explains Hofmann. The researchers will integrate this conditional probability factor into the AI system so that it still acts reliably.

The Future of Quantum Corrals

Using an AI-controlled scanning tunneling microscope that can work around the clock, the researchers ultimately want to build so-called quantum corrals. These are nanostructures in the shape of a gate, which can be used to trap electrons from the material on which they are deposited. The wave-like properties of the electrons then lead to quantum-mechanical interferences that can be utilized for practical applications. Until now, quantum corrals have mainly been built from single atoms.

Oliver Hofmann’s team now wants to produce them from complex-shaped molecules: “Our hypothesis is that this will allow us to build much more diverse quantum corrals and thus specifically expand their effects.” The researchers want to use these more complex quantum corrals to build logic circuits in order to fundamentally study how they work at the molecular level. Theoretically, such quantum corrals could one day be used to build computer chips.

Collaborative Research and Expertise Synergy

For its five-year program, the research group is pooling expertise from the fields of artificial intelligence, mathematics, physics, and chemistry. Bettina Könighofer from the Institute of Information Security is responsible for the development of the machine learning model. Her team must ensure that the self-learning system does not inadvertently destroy the nanostructures it constructs.

Jussi Behrndt from the Institute of Applied Mathematics will determine the fundamental properties of the structures to be developed on a theoretical basis, while Markus Aichhorn from the Institute of Theoretical Physics will translate these predictions into practical applications. Leonhard Grill from the Institute of Chemistry at the University of Graz is primarily responsible for the real experiments on the scanning tunneling microscope.

Reference: “MAM-STM: A software for autonomous control of single moieties towards specific surface positions” by Bernhard Ramsauer, Johannes J. Cartus and Oliver T. Hofmann, 6 June 2024, Computer Physics Communications.
DOI: 10.1016/j.cpc.2024.109264