Leveraging machine learning to harness non-parabolic effects in semiconductor heterostructures
DOI
10.1016/j.physe.2022.115513
Abstract
The engineering of optoelectronic devices is closely related to the dependency of quantum wells properties with structural parameters. Obtaining their properties relies on solving Schrödinger’s equation employing several degrees of approximation as the influence of band non-parabolicity effects. In general, the more sophisticated the approximations the longer the development cycle. To efficiently analyze the response of structures equipped with quantum wells, we propose using an artificial neural networktrained with the results from the solution of Schrödinger’s equation. We observe an impressive agreement between the predicted quantum well’s eigenenergies and effective masses, in comparison to the results obtained from the iterative solution of the Schrödinger equation, for an extensive range of structures. The time spent to retrieve the properties of the structures using the artificial neural network was a fraction of the time spent with the solution of Schrödinger’s equation, enabling the neural network to be considered an effective tool in the development and improvement optoelectronicdevices.
Document Type
Article
Publication Date
1-2023
Publisher Statement
© 2022 Elsevier B.V. All rights reserved.
journal homepage: www.elsevier.com/locate/physe
Recommended Citation
da Silva Macedo, G., Rebello de Sousa Dias, M., & Bezzara, A. T. (2023). Leveraging machine learning to harness non-parabolic effects in semiconductor heterostructures corresponding. Physica E: Low-dimensional Systems and Nanostructures, 146. https://doi.org/10.1016/j.physe.2022.115513