LEAD: Literature Enhanced Ab Initio Discovery of Nitride Dusting Layers for Enhanced Tunnel Magnetoresistance and Lower Resistance Magnetic Tunnel Junctions

Abstract

Abstract Magnetic tunnel junctions (MTJs) using magnesium oxide (MgO) tunnel barriers face challenges of high resistance‐area product (RA) and low tunnel magnetoresistance (TMR). To discover alternative materials, Literature Enhanced Ab initio Discovery (LEAD) is developed as a framework that combines language models with first‐principles screening. A domain‐specific Word2Vec model is used to extract correlations from materials science literature, identifying promising candidates such as hybrid nitride‐based barriers. Additionally, a Bidirectional Encoder Representations from Transformers based masked language model is trained on the same corpus to capture deeper contextual relationships, identifying tantalum nitride (TaN), vanadium nitride (VN), and titanium nitride (TiN) as candidate barrier materials. These predictions are subsequently evaluated using density functional theory (DFT) simulations in QuantumATK, benchmarking the predicted materials' TMR and RA against MgO and scandium nitride (ScN). Results show that MTJs featuring a monolayer dusting of ScN, or monolayer or bilayer dusting of TiN, on either side of MgO have similar or lower RA while achieving higher TMR than pure MgO junctions. LEAD offers a scalable method for discovering tunnel barriers and highlights nitride dusting layers as promising for next‐generation MTJ performance.