Data-Driven and Machine Learning-Assisted Design of Functional Defects in Two-Dimensional Materials

Being atomically thin and amenable to external controls, two-dimensional (2D) materials offer a new paradigm for many technologies such as (opto)electronics, energy conversion, and quantum information. In this talk, as an example of functional defect design, I will discuss how data-driven material science can be combined with symmetry-based physical principles to guide the search for spin defects in 2D materials for quantum information technologies and beyond. In our initial work, the use of local bonding symmetry as a material design hypothesis enables the identification of anion antisite defects as promising spin qubits and quantum emitters in six monolayer transition metal dichalcogenides. To enable high-throughput search of functional defects in a vast material space, we propose two machine learning (ML) models that are specially designed for learning localized defect properties, taking advantage of topological objects (Betti numbers) as node features and the Siamese equivariant network architecture. Trained by 5,000~20,000 diverse defected material systems, our models outperform the state-of-the-art models in predicting the formation energies of point defects. This ML capability enables the fast screening of functional defects in 2D materials, and the high-throughput search in all known binary 2D materials led to the identification of more than 45 quantum defect candidates that can be utilized as qubits and/or quantum emitters. At the end of the talk, I will discuss future directions to accelerate the discovery of “defect genome” in a vast space of material systems.