Atomistic insights into the full-cycle switching of Y-doped Sb₂Te₃ phase change memory

Doped Sb-Te phase change materials (PCMs) are promising for memory and neuromorphic computing, but probing dopant mechanisms is challenging, and ab initio simulations are scale-limited. Here, a validated Sb-Te neuroevolution potential (NEP) is extended to Y-Sb-Te by strategically adding a targeted set of Y-doped Sb₂Te₃ configurations. The resulting potential attains density functional theory (DFT)-level accuracy, preserves the original binary fidelity, and accurately predicts unseen Y concentrations along the Sb₂Te₃ line. This approach ensures robustness for large-scale simulations, overcoming the weakness of narrowly trained potentials. The impact of Y doping on Sb₂Te₃ was systematically studied, revealing that Y increases melting and glass–transition temperatures, enhances amorphous stability, reduces diffusivity, and suppresses crystallization. Moreover, large-scale simulations capture collective behaviors inaccessible to ab initio models. Cumulative SET operations show stepwise structural relaxation governing neuromorphic updating. Furthermore, simulating a 49 × 18 × 18 nm³ cross-point-like cell with ∼ 460,000 atoms under unidirectional heating reveals gradient-driven operations, characterized by a kinetic asymmetry between nucleation and growth during the SET process, and a low-power (∼0.022 pJ) re-amorphization during the RESET process. This work delivers an application-ready potential for Y-doped Sb₂Te₃ and a data-efficient dataset extension methodology for robust, scalable potentials for other doped PCMs.