Data-Driven Doping for Semiconductors: Identifying Top Dopant Candidates for Complex Crystals

Stoichiometric crystals, while often insulators, pose a challenge for technological applications that require excess electrons (n-type) or holes (p-type) through p-n junctions. Complex crystals, such as ternary and quaternary systems, offer numerous dopant possibilities, making it difficult to determine the optimal substituting element. This research utilizes a data-driven approach to identify the top 10 potential dopants for multicomponent materials.<br/><br/>Our "doper" code utilizes chemical similarity metrics based on structure analysis and machine-learned representations. It generates n-type p-type cation and anion potential dopants based on the input species of multicomponent materials. We validate our approach using density functional theory (DFT) calculations, considering solubility energy and defect levels. Our methodology is implemented within the Semiconducting Materials from Analogy and Chemical Theory (SMACT) framework, which offers rapid screening tools based on chemical element data.<br/><br/>Our results demonstrate a strong correlation between the data-driven top dopant candidates generated by "doper" and DFT calculations. This validates the efficacy of our approach in identifying promising dopant elements for semiconductors. By leveraging machine learning and structure analysis, our data-driven technique accelerates materials discovery and optimization, significantly reducing time and cost in dopant selection.