A Unified Machine‐Learning Framework for Predicting Barrier‐Free Metal Contacts in 2D Semiconductor Transistors

Achieving low‐resistance, reproducible metal contacts remains a major obstacle to integrating two‐dimensional (2D) semiconductors into deeply scaled CMOS technologies. Despite intensive study of selected semiconductor‐metal (S‐M) interfaces, particularly those based on transition‐metal dichalcogenides, contact optimization continues to rely heavily on experimental, and computational trial‐and‐error, in part due to variations in interface structure and processing that obscure intrinsic trends and hinder direct comparison. To address this challenge, we present a unified, data‐driven framework that connects atomistic interface physics with device‐level metrics using descriptor‐light machine learning. Based on high‐throughput density‐functional‐theory (DFT) calculations spanning 16 665 candidate 2D semiconductors and 80 metals, we construct a record‐scale dataset covering diverse interface chemistries and contact regimes. Trained solely on publicly available database‐derived elemental descriptors, the models predict van der Waals (vdW) barrier formation together with vertical, and lateral Schottky barrier heights (SBHs) within milliseconds, enabling rapid, ab initio‐free screening of arbitrary contacts. Validation against experimental data and quantum transport simulations confirms accurate reproduction of established contact behaviors and resistance trends. The framework further identifies over one hundred previously unexplored near‐barrier‐free interfaces and reveals physically interpretable principles governing carrier injection, establishing a scalable pathway toward intrinsically Ohmic 2D contacts.