Towards the Prediction of Organic Thin-Film Structures with DFT and Machine Learning

The properties of a material depend on its structure, and no theoretical prediction of the most stable thin film structures through traditional, exhaustive first-principle studies is feasible due to the combinatorial explosion in the number of possible polymorphs.

For monolayers, the machine-learning based SAMPLE approach [1] can already circumvent this problem, by using a few hundred DFT calculations to evaluate the energy of millions of possible polymorphs through Bayesian Linear Regression. It is our intention to extend the applicability of SAMPLE from monolayers to (meta)stable thin films.

Assuming a Frank-van der Merwe growth mechanism is at play, we can investigate thin-film structures by considering consecutive monolayers forming on top of one another.
As a first step, we study the formation of a second layer on top of the most stable monolayer predicted by SAMPLE for a well known system, Benzoquinone on Ag(111).
We evaluate the ways in which the electronic properties of the substrate can promote the formation of different second layers, and we give an assessment of the impact of such a change on the layer-to-layer charge transport rate in the thin film, as one can predict within the hopping regime.

[1] Hörmann et al., Computer Physics Communications 244, 143–155, 2019