New numerical models for the design of ductile hard layers

The Deposition of Thin Films and Nanostructures research group at the Department of Plasma Physics and Technology has long excelled in the experimental preparation of thin films. Our scientists have researched and prepared hard protective coatings for over twenty years. Since then, a few scientists have grown into a sizeable scientific group. Most scientists optimize the experimental preparation of thin films by magnetron sputtering by selecting appropriate process parameters and then checking the film's composition, structure, mechanical properties, etc. Dr. Pavel Ondračka focuses on the modeling of thin film materials and predictions of their properties. He is a theoretician who works on material models for thin films. These models help to understand the processes that determine the formation, structure, and properties of thin films.

Dr. Ondračka graduated with a Bachelor's degree in Physics and a Master's degree in Plasma Physics at the Faculty of Science of MU under the supervision of doc. Lenka Zajíčková focusing on optical diagnostics of thin films. He continued his Ph.D. studies under her leadership in theoretical calculations of optical properties, and he successfully graduated in Advanced Nanotechnology and Microtechnology in 2018. He worked in Germany for a three-year postdoc research position at RWTH Aachen University from 2019 to 2021. He has been working at The Department of Plasma Physics and Technology since the end of 2021 in the team of Prof. Petr Vašina.

Marie Skłodowska-Curie Actions (MSCA) is a prestigious European project for Ph.D. students and young scientists to improve their skills or education. Dr. Ondračka applied for this project and received an excellent evaluation. However, he was unsuccessful in obtaining the European funding in the strong competition of other high-quality projects. Based on the superb review, he was able to apply for the MSCA project of the Czech MŠMT, which supports a limited number of unsuccessful European MSCA projects with excellent evaluations, and he was granted a project MSCAfellow5_MUNI (CZ.02.01.01/00/22_010/0003229) from the Operational Programme Research, Development and Education. Dr. Oleksandr Galmiz from the Deparment of Plasma Physics and Technology also applied for MSCA, you can read about his project here

In his project entitled "Material design of moderately ductile hard coatings," Dr. Ondračka aims to develop a model capable of predicting the mechanical properties of amorphous and nanocomposite W-B-C (tungsten-boron-carbon). Ductile hard coatings for protective purposes, of which the W-B-C system is one, have been the focus of researchers at the Department of Plasma Physics and Technology for several years, and they have also been working with industrial partners. Prof. Petr Vašina, an expert in the deposition of W-B-C layers and their applications, acts as a supervisor for the project. Dr. Ondračka is assisted with the technical aspects of modeling by the second supervisor, Assoc. Prof. David Holec from the University of Leoben in Austria.

Dr. Ondračka comments, "Scientists are good at modeling crystalline layers and their mechanical properties. But for amorphous layers, we run into problems. The amorphous structure is, by definition, aperiodic, and huge models are needed to describe it, for which classical quantum-mechanics methods, such as density functional theory, are not very useful because of the demands on computational time, which increases cubically with the size of the model. If you double the number of atoms, the computation time increases eight times. However, an amorphous or nanocomposite (crystalline grains in an amorphous matrix) structure in thin films is very common. Therefore, efficient modeling methods for these systems are essential."​​

Dr. Ondračka develops his model using machine-learning methods describing inter-atomic interactions, the so-called machine-learned inter-atomic potential. The starting point is the assumption that the behavior of an atom depends only on its immediate surroundings, so the potential can be trained on smaller quantum-mechanical models that are easy to calculate. This assumption is reasonably well met, except for exotic materials such as superconductors or some magnetic materials, so the loss of accuracy of the model compared to the calculations on which the model was learned can be minimal for a well-tuned potential. The advantage is the considerable speedup. "We're talking many orders of magnitude, so you can have calculations that would normally take weeks finished in a matter of seconds because of linear scaling. It takes twice as long to model twice the number of atoms, and we can go to models with millions of atoms. The main challenge of this work is that machine learning is not much used for such complex three-element systems. The situation is very complicated considering different structures, local compositions, and the presence of interfaces. So most of the time is spent preparing the learning data, optimizing and testing the potential to work well under all conditions," adds Dr. Ondračka.

Scientific group of doc. Holce is very experienced in modeling related to thin film materials. They work closely with experimenters for whom they make predictions of new materials with interesting properties and they provide an essential theoretical understanding of how existing materials work. A big topic in the group is, among other things, the current issue of hydrogen storage. Dr. Ondračka has come to Leoben, Austria, for a three-month stay as part of an optional project internship, the so-called Secondment, to work with Assoc. Prof. Holec and his colleagues on developing machine learning potential for W-B-C.

The aim of Dr. Ondračka's work is optimizing the potential and generally applying machine learning methods for theoretical numerical calculations of the W-B-C layer structure is. "Until now, no one has been able to model a W-B-C system realistically in sizes relevant to the structure we observe in thin films. With our machine learning potential, the predictions could be directly compared to laboratory-prepared layers. We will be able to predict properties for the amorphous and nanocomposite phases, and help researchers optimize thin films for industrial coatings. In the future, I envision that this methodology could easily be applied to similar systems just coming into the sights of experimental scientists," concludes Dr. Ondračka.