Data‐Driven Approach to Designing Two‐dimensional Van der Waals Heterostructures: Misjudgment of Band Alignment Type and its mechanism

Data‐driven approach, including database building and high‐throughput design, is rapidly applied to various fields, ranging from physics to materials. Predicting the band alignment type at interfaces, which is based on Anderson's rule (AR), is crucial for implementing the data‐driven approach in the field of 2D van der Waals heterostructures (vdWHs). The validity of the AR‐based data‐driven approach, however, is ambiguous, particularly when it involves band alignment. Based on a large‐scale calculation (1231 heterostructures), it is explicitly demonstrated that the current AR‐based data‐driven approach have prediction errors for certain band alignment type, including the inability to predict type‐other band alignment and the partial misclassification of type II band alignment as type III. Such prediction errors are quantitatively explained by geometric deformation, interlayer orbital coupling, and charge redistribution effects. As a result, the current data‐driven design approach is insufficiently reliable for type II, type III, and type‐other heterostructures, which may result in inaccurate or incomplete design results. Herein, a flaw in the current AR‐based data‐driven approach is elucidated as well as important information for reasonably designing 2D vdWHs is provided in general.