Machine learning uncovers the mystery of atomic scale amorphous alumina thin films

Aluminum oxide or alumina is a fruit fry in materials science. It has been thoroughly researched and well understood. This compound, simple chemical formula al₂o₃It frequently occurs in the form of mineral corundum and its famous color variations in sapphire and ruby, and is used for a variety of purposes, including electronics, chemical industry, and technology ceramics.

A special feature of aluminum oxide is its ability to take on a variety of structures while maintaining the same chemical composition. All of these variants are well understood with one exception. In addition to some crystalline forms, aluminum oxide can also exist in amorphous, i.e., in a disordered state. Amorphous alumina has particularly advantageous properties for some high-tech applications, particularly in the form of a uniform protective thin film coating or ultra-thin agitating layer.

Despite its extensive use and the know-how available to handle it, amorphous alumina remains an atomic level mystery. “Crystalline materials are made up of small, periodically repeated subunits,” explains Empa Researcher VladySlav Turlo, Thun's Advanced Materials Processing Laboratory.

Therefore, it is relatively easy to examine them to the level of a single atom. The same goes for modeling on a computer. After all, if you can calculate the interaction of atoms in single crystal units, you can also easily calculate a large crystal made up of many units.

Amorphous materials do not have such a periodic structure. The atoms are cluttered together. It is difficult to inspect and even more difficult to model. “When simulating a thin film coating of amorphous alumina grown from scratch at the atomic level, calculations take longer than the age of the universe,” says Turlo.

However, accurate simulations are key to effective material research. They help researchers understand the material and optimize its properties.

Experiments satisfy the simulation

For the first time, EMPA researchers led by Turlo have managed to simulate amorphous alumina quickly, accurately and efficiently. A model that combines experimental data, high-performance simulations, and machine learning provides information on the atomic arrangement of amorphous ALs₂o₃ It's the first of the layers and their kind. Researchers published their results in the journal NPJ Calculation Materials.

A breakthrough has been made possible thanks to the interdisciplinary collaboration between the three EMPA labs. Turlo and his colleague Simon Gramatte were the first authors of the publication and are based on models based on experimental data. Materials and nanostructure mechanic researchers used atomic layer deposition to produce amorphous aluminum oxide thin films, and studied with colleagues at Duvendorf's participating technology and corrosion research.

One of the great strengths of the model is that it also takes into account the integrated hydrogen atoms, in addition to the alumina and oxygen atoms in alumina. “Amorphous alumina contains varying amounts of hydrogen depending on the manufacturing method,” explains co-author Ivo Utke. Hydrogen, the smallest component of the periodic table, is particularly difficult to measure and model.

An innovative spectroscopy called HAXPES, which can only be achieved in Switzerland with EMPA, allowed researchers to characterize the chemical state of aluminum with various thin films, incorporate it into simulations, and for the first time reveals the distribution of hydrogen within alumina.

“Along with certain content, we were able to demonstrate that hydrogen bonds to the oxygen atoms of the material and affects the chemical state of other elements in the material,” says co-author Claudia Cancellieri. This will change the properties of the material. Aluminum oxide is “fluffy,” meaning that it results in a lower density.

Potential breakthroughs of green hydrogen

This understanding of atomic structure paves the way for new applications of amorphous aluminum oxide. Turlo sees its greatest potential in the production of green hydrogen. Green hydrogen is produced by using renewable energy to divide water. Separating hydrogen from the oxygen produced during water splitting requires an effective filter material so that only one of the gases can pass through.

“Amorphous alumina is one of the most promising materials for such hydrogen films,” says Turlo. “Our model allows us to better understand how the hydrogen content of a material prefers the diffusion of gaseous hydrogen to other large molecules.”

In the future, researchers would like to use the model to develop better films made of alumina.

“Understanding materials at the atomic level allows us to optimize material properties that are related to mechanics, optical, or transmission in a much more targeted way,” says materials researcher Utke. This model can lead to improvements in all application areas of amorphous alumina and can also be transferred to other amorphous materials over time.

“We have shown that it is possible to accurately simulate amorphous materials,” Turlo summed up. And thanks to machine learning, this process only takes about a day in billions of years.