Closing the Loop: Simulations Reveal Why a Small Chemistry Change Reshapes Thin-Film Growth
During molecular beam epitaxy, the order in which atoms arrive at a surface, and the exact ratio between them, can decide whether a film grows as a scattering of isolated islands or as a smooth, continuous layer. A new study shows why, tracing this sensitivity for the layered semiconductor tin selenide (SnSe) grown on magnesium oxide (MgO) all the way down to how a single tin or selenium atom behaves once it lands.
To connect what happens on the growth surface to what shows up under the microscope, the team combined three approaches. Collaborators at Pennsylvania State University and Georgia Institute of Technology grew SnSe films by molecular beam epitaxy while systematically varying the tin-to-selenium flux ratio and the order in which the two elements were supplied. The atomic-scale interpretation was led from PDI: Nadire Nayir, head of the Computational Materials Science group, modelled the same growth process atom by atom using reactive molecular dynamics and applied unsupervised machine learning to the simulated surfaces to pick out morphology trends independently of any human-imposed categories, work carried out together with Wesley Reinhart and Adri van Duin at Penn State.
The simulations trace this back to a straightforward chemical asymmetry. Selenium-rich conditions passivate the growing surface and encourage atoms to spread out sideways, while tin-rich conditions favour rapid clustering of tin atoms into taller, three-dimensional mounds. Shifting the flux ratio and deposition sequence tips the balance between these two competing behaviours, and the simulations reproduce the same transition, from rough island growth to smooth films, that shows up experimentally in the AFM images.
The machine learning analysis adds a second, independent line of evidence. Applied to images rendered directly from the simulation snapshots, with no prior labelling or physical assumptions built in, it distils the surface morphologies into a low-dimensional representation that captures the same growth trends seen in both the simulations and the experiments.
The stakes go beyond a tidier film. Nanoscale roughness in SnSe shapes how well it scatters charge carriers and phonons and how stable its domains are, properties that matter directly for electronic and energy devices built from this material. The chemistry-guided picture developed here offers a concrete route to controlling that roughness by design.
Taken together, the results close a loop that runs from growth conditions to atomic-scale mechanism to morphology to the structure observed in the lab, and back again to guide the next growth run. SnSe is the demonstration here, but the underlying framework, pairing MBE with reactive molecular dynamics and unsupervised analysis, is not specific to this material system. Nayir's group already has this computational capability in place at PDI, and is open to extending it to other materials, interfaces, deposition processes, and growth questions across the institute. A further framework aimed at computation-guided control of film thickness is already in preparation.
Title: Local Chemistry-Guided Molecular Beam Epitaxy Growth of SnSe on MgO via Combined ReaxFF Modeling and Machine Learning
Authors: M. Wang, I. A. Moses, J. R. Chin, Q. Zhang, M. Hilse, S. Law, L. M. Garten, W. F. Reinhart, N. Nayir, A. C. T. van Duin
Source: ACS Appl. Mater. Interfaces 18 (33), 45729–45741 (2026)
DOI: 10.1021/acsami.6c08764