Machine Learning Electronic Transport Properties of Multilayered Semiconductor Nanostructures

Computing components are being aggressively inserted into semiconductor architectures to perform operations at high rates, for a broad range of applications. The contact interfaces between these components dictate performance, especially as device dimension approaches nanoscale. Ab initio methods become expensive and infeasible to predict electronic properties of systems with large number of configurational degrees of freedom. In this study, we employ machine learning (ML) algorithms to predict electronic structure and transport of non-ideally fabricated multilayered thin film Si/Ge nanostructures. The algorithm is trained on inexpensive ~200 DFT calculations of SixGe1-x substitutional alloys, by exploiting the relationship between local atomic environments and electronic properties. The predictor variables are obtained with Voronoi tessellation approach and the response variables are calculated with decision tree regression algorithm. Our model has shown remarkable ability to predict band structures and Onsager transport coefficients of large non-ideal superlattices. The ML framework will facilitate the development of inverse design approach to engineer interface profiles for desired performance of integrated semiconductor architectures.