Abstract: Liquid metal embrittlement during hot rolling of recycled steels, caused by tramp elements accumulated through repeated scrap recycling, represents a critical challenge for sustainable steel production. Here we develop interatomic potentials for Fe-Cu-X (X = Sn, Sb, As, Pb, and Bi) ternary systems and employ molecular dynamics simulations to elucidate atomistic mechanisms governing grain boundary penetration of liquid metals. Systematic investigation of binary Fe-X and Cu-X systems establishes that penetration is controlled by wetting thermodynamics depending on both the mixing enthalpy and the solute concentration dissolved in the liquid phase. Extension to ternary systems reveals three distinct penetration regimes determined by the disparity in mixing enthalpies between binary pairs relative to entropic stabilization: (i) Sn and Sb promote cooperative penetration where tramp elements lead Cu into grain boundaries, with penetration efficiency increasing systematically with tramp element concentration in the liquid; (ii) Pb and Bi suppress penetration as their unfavorable Fe interactions exclude them from the solid-liquid interface, maintaining baseline Fe-Cu behavior; and (iii) As exhibits decoupled penetration where its extreme Fe affinity induces liquid phase separation, suppressing Cu penetration despite independent As infiltration. These findings establish a thermodynamic framework based on the difference between binary mixing enthalpies that enables prediction of tramp element effects on grain boundary penetration and provides guidance for prioritization of elements for industrial control.

Abstract: Computer simulations have long been key to understanding and designing phase-change materials (PCMs) for memory technologies. Machine learning is now increasingly being used to accelerate the modelling of PCMs, and yet it remains challenging to simultaneously reach the length and time scales required to simulate the operation of real-world PCM devices. Here, we show how ultra-fast machine-learned interatomic potentials, based on the atomic cluster expansion (ACE) framework, enable simulations of PCMs reflecting applications in devices with excellent scalability on high-performance computing platforms. We report full-cycle simulations-including the time-consuming crystallisation process (from digital "zeroes" to "ones")-thus representing the entire programming cycle for cross-point memory devices. We also showcase a simulation of full-cycle operations, relevant to neuromorphic computing, in a mushroom-type device geometry.