Accelerating chemical vapour deposition discovery and development with generative machine learning

Research question

Thin layers, or films, of different materials play a vital role in everyday life. They keep food fresh, make cutting tools more durable, and form the foundation of all electronic devices. One of the main ways to create these films is chemical vapour deposition (CVD), which uses reactive gases carrying the atoms needed for the film.

Despite CVD’s technological importance, developing new processes still relies heavily on trial and error and in-house expertise, because current modelling methods are slow and resource-intensive. In this project, we aim to combine CVD modelling with generative machine learning to achieve a deeper, more reliable understanding of CVD and to speed up the sustainable discovery of new processes.

Sustainability aspects

A combined use of computational fluid dynamics (CFD) and density functional theory offers a powerful way to model CVD at the atomic level. However, this approach is highly labor-intensive. The only real alternative to costly and resource-heavy CVD experiments — which consume significant electricity, cooling water, gases, and raw materials, and generate large amounts of hazardous waste gases — is such advanced modelling.

The main limitation of traditional CFD methods (e.g., finite element analysis) is that they do not use data-driven insights. We therefore propose integrating probabilistic machine learning models into our approach to overcome CVD’s computational challenges.