A convolutional neural network is trained on a large dataset of suitably randomized film profiles and corresponding elastic energy densities ρɛ, computed by the finite element method. The trained model provides quantitative predictions of ρɛ for arbitrary profiles, surrogating its explicit calculation, and is used for the time integration of partial differential equations describing the evolution of strained films. The close match found between the neural network predictions and the “ground-truth” evolutions obtained by the finite element method calculation of ρɛ, even after tens-of-thousands of integration time-steps, validates the approach. A substantial computational speed up without significant loss of accuracy is demonstrated, allowing for million-steps simulations of islands growth and coarsening. The intriguing possibility of extending the domain size is also discussed.
The surface diffusion and intermixing of Ge ad‐atoms over Si (001) 2 × 1 substrates using classical molecular dynamics (CMD) simulations are characterized here. Several interatomic potentials, parametrizations, and parameter mixing rules are contemplated. A novel simulation scheme is devised to characterize the effective frequency of surface diffusion and intermixing events overcoming the inherent difficulties related to their interdependency in heteroepitaxial systems. The effective energy barriers of these events encompass different atomistic mechanisms weighted by their occurrence probabilities. The overall description of surface diffusion and intermixing based on Stillinger–Weber (SW) potential is in agreement with ab initio calculations and experimental observations, though some atomistic details differ. This study is extended to Si(001) substrates with stressed Ge monolayers grown on top. It is found that Ge ad‐atom dynamics is accelerated with respect to the case of the pure Si substrate and that diffusion across dimer rows is mainly mediated by the atomic exchange of the Ge ad‐atom with a Ge atom on the surface.