Autonomous AI assistant develops advanced nanostructures

Autonomous AI assistant develops advanced nanostructures
by Robert Schreiber
Berlin, Germany (SPX) Jan 17, 2025

Understanding the properties of materials often requires examining more than just their chemical composition. The spatial arrangement of molecules within atomic lattice structures or material surfaces plays a critical role in determining material properties. By manipulating individual atoms and molecules on surfaces using high-performance microscopes, materials scientists have made significant strides. However, this process remains labor-intensive and limited to constructing relatively simple nanostructures.

A new initiative at Graz University of Technology (TU Graz) aims to revolutionize this process using artificial intelligence (AI). "We want to develop a self-learning AI system that positions individual molecules quickly, specifically, and in the right orientation, all autonomously," explained Oliver Hofmann from the Institute of Solid State Physics, who leads the project. The ultimate goal is to construct highly intricate molecular structures, such as nanometer-scale logic circuits. The Austrian Science Fund has awarded the research group funding of 1.19 million euros for this ambitious project.

Automated molecule positioning with scanning tunnelling microscopes

The project employs a scanning tunnelling microscope (STM) to position individual molecules on surfaces. The STM's probe tip delivers an electrical impulse to deposit a molecule in a specific location. "Currently, it takes several minutes for a person to complete this step for a single molecule," Hofmann noted. "Constructing more complex structures involves positioning thousands of molecules, followed by rigorous testing, which demands substantial time and effort."

The team plans to leverage machine learning techniques to enable a computer to autonomously control the STM. First, AI algorithms will generate an optimal construction plan, outlining the most efficient and reliable sequence for building the desired structures. Self-learning AI will then guide the STM's probe tip to place molecules with precision. Hofmann highlighted the challenges of this process: "Aligning complex molecules precisely is inherently probabilistic. Our AI system will account for these uncertainties to ensure reliable performance."

Quantum corrals and logic circuits

The researchers aim to construct advanced quantum corrals - nanostructures shaped like gates - using their AI-driven STM. Quantum corrals can trap electrons on a material's surface, enabling quantum-mechanical interference effects that may have practical applications. Traditionally, quantum corrals have been built using single atoms. Hofmann's team intends to construct these structures with complex molecules to create a broader range of quantum corrals and expand their functionalities.

"Our hypothesis is that using complex-shaped molecules will enable the construction of more diverse quantum corrals, thereby enhancing their effects," Hofmann said. The team plans to utilize these structures to develop molecular-scale logic circuits and explore their fundamental mechanisms. In the long term, this research could contribute to the development of molecular-level computer chips.

Interdisciplinary collaboration

This five-year program draws expertise from diverse fields, including artificial intelligence, mathematics, physics, and chemistry. Bettina Konighofer from the Institute of Information Security leads the development of the machine learning model, ensuring the AI system does not inadvertently damage the nanostructures it assembles. Jussi Behrndt from the Institute of Applied Mathematics focuses on theoretical analyses of the structural properties, while Markus Aichhorn from the Institute of Theoretical Physics translates these predictions into practical methods. Meanwhile, Leonhard Grill from the University of Graz's Institute of Chemistry oversees experimental applications with the STM.

Related software

The team has also developed MAM-STM, a software solution designed for autonomous control of molecular placement on surfaces, detailed in the publication:

Research Report:MAM-STM: A software for autonomous control of single moieties towards specific surface positions