Inverse Design of Semiconductor Heterostructures Using First-Principles Modelling and Machine Learning Approaches

Nanofabrication techniques have attained great control over the growth of the seminconductor Heterostructures. Nevertheless, the fabricated materials are strongly affected by the growth process, and their structural properties show high variability. It is essential to acquire a comprehensive understanding of the relationship between structural parameters and electronic properties of materials, to optimize their performance. Ab initio computational approaches enable prediction of materials properties with minimal experimental input; however, these approaches often require large computational costs. It remains a challenge to model electronic properties of technologically relevant heterostructures incorporating full structural complexity, stemming from the vast fabrication dependent parameter space. In this talk, I will discuss our machine learning (ML) based approaches that extend the applicability of ab initio techniques for predicting electronic properties of fabricated heterostructures. We develop an inverse approach that predicts the structural features of a given heterostructure for target electronic band structures. We train the ML models with first-principles electronic property data of silicon/germanium superlattices of varied period and composition. We use structural descriptors and unfolded effective band structures or spectral functions as training data. I will discuss the importance of accurate identification of local structural features on the performance of our ML models. This approach establishes a direct connection between experimental and theoretical results. For example, our model predicts the atomic scale structural data of the system that results in a given spectra obtained with ARPES techniques. We provide the ARPES image of bulk Si and δ-doped Si as input. Our inverse approach successfully converts the data into a set of structural and atomic features necessary to describe the bulk systems. The ML models reveals key physical insights regarding the relationships between atomic configurations and their contributions to electronic properties.