Unsupervised machine learning guides the design of MnFeCoNiMg high-entropy alloys for optimizing MoS2 interface properties

The development of high-entropy alloy (HEA) interfaces with two-dimensional materials like MoS 2 holds great promise for next-generation electronics but is hindered by the vast composition space and complex entropy-property relationships. This study introduces an integrated machine learning and experimental framework to efficiently design and optimize MnFeCoNiMg HEAs for MoS 2 heterostructures. Unsupervised clustering revealed compositionally distinct regions characterized by Co/Ni enrichment, and subsequent DFT analysis identified d-orbital hybridization as the key mechanism governing interfacial properties, enabling the simultaneous achievement of an ultralow Schottky barrier of 0.28 eV and a high carrier mobility of 2.8 cm 2 /V·s. A graph neural network model demonstrated exceptional predictive accuracy for interface properties. Guided by these insights, we experimentally synthesized a Mn 0.18 Fe 0.27 Co 0.22 Ni 0.25 Mg 0.08 composition, which exhibited 94% phase purity and a record-high μ×Φ product of 0.82 eV 2 . This work establishes a new paradigm for the data-driven design of entropy-engineered quantum materials.