Discoveries in machine learning expose the hidden properties of slender atomic layers

In a groundbreaking development, researchers led by Turlo at Empa have successfully simulated amorphous aluminum oxide on a computer for the first time. This milestone is set to revolutionise the field of hydrogen membranes, particularly in the separation of hydrogen from oxygen.

The simulation, which combines experimental data, high-performance simulations, and machine learning, offers an atomically precise understanding of this promising material. It considers trapped hydrogen atoms in amorphous aluminium oxide, a factor that has been notoriously difficult to measure and model due to hydrogen's status as the smallest element in the periodic table.

The researchers used an innovative spectroscopy method called HAXPES to characterise the chemical state of aluminum in the thin layers. By deriving the hydrogen distribution in aluminum oxide for the first time using the simulation, they have opened up new avenues for understanding the atomic-level processes in protective thin films and membranes.

Hydrogen binds to oxygen in the material at a certain concentration and influences the material properties. In the case of amorphous aluminum oxide, this binding causes the material to become less dense. This finding is crucial for future material design and applications, particularly in the production of green hydrogen.

Turlo sees the greatest potential for amorphous aluminum oxide in this area, as it could significantly improve the efficiency and sustainability of hydrogen production. The Empa researchers are now looking to produce targeted aluminum oxide membranes based on the simulations, with the aim of advancing the field of green hydrogen production.

The simulation process, which previously would have taken billions of years, now only takes about a day thanks to machine learning. This accelerated process allows for more specific optimisation of material properties, paving the way for improvements in all applications of amorphous aluminum oxide and potentially other amorphous materials over time.

This research underscores the power of computational modelling in materials science and the potential it holds for shaping the future of sustainable energy production.