Unified Physics-Guided Discovery of Ultrahigh- and Ultralow-Thermal Conductivity Two-Dimensional Materials

Abstract

Abstract As thermal management approaches the low-dimensional limit in post-Moore electronics, two-dimensional (2D) materials offer a versatile platform for engineering lattice thermal conductivity (κL). However, a unified framework capable of discovering both ultrahigh- and ultralow-κL materials remains absent because these extremes originate from fundamentally different phonon-transport physics. Here, we develop a unified physics-guided framework integrating a modified 2D Slack-type model with hierarchical random forest learning for accurate full-range κL prediction. The framework establishes a physically interpretable descriptor space, achieves high predictive accuracy (R2 = 0.962), and reveals distinct descriptor spaces governing ultrahigh- and ultralow-κL transport. Screening 16,488 2D materials identifies 48 ultrahigh-κL candidates exceeding most conventional metals and 12 ultralow-κL candidates below liquid-like systems, establishing a unified paradigm for discovering and engineering thermal conductivity extremes in 2D materials.