Polarization-controlled ultrafast carrier dynamics in sliding ferroelectrics revealed by machine learning-accelerated nonadiabatic dynamics simulations

Sliding ferroelectrics offer a unique pathway to electrically reconfigurable polarization at the atomic scale, offering compelling opportunities for ultralow-power optoelectronic devices. Yet, despite rapid advances in materials discovery, the fundamental question of how polarization reversal dynamically governs photoexcited carrier transport, central to device operation, remains unresolved. This knowledge gap arises primarily from the prohibitive cost of accurately simulating nonequilibrium excited-state dynamics in large systems. Here, we establish a unified framework that combines hybrid functional-level electronic structure accuracy with machine learning-accelerated nonadiabatic molecular dynamics to resolve polarization-controlled carrier dynamics in prototypical sliding ferroelectric heterostructures, ZrTe 2 /HfTe 2 and TiSe 2 /HfSe 2 . Our approach achieves HSE06-level accuracy for periodic systems at more than an order-of-magnitude reduction in computational cost, making largescale excited-state dynamics tractable. We further reveal that polarization reversal exerts deterministic control over photoinduced charge motion, redirecting interlayer carrier-transfer pathways, suppressing electron-hole recombination, and stabilizing long-lived charge-separated states. Strikingly, TiSe 2 /HfSe 2 exhibits ultrafast interlayer electron transfer on a sub-50 fs timescale in both AB and BA ferroelectric states. In sharp contrast, the AB stacking of ZrTe 2 /HfTe 2 supports rapid interlayer transfer of both electrons and holes, whereas the BA stacking nearly eliminates any interlayer carrier motion. This pronounced stacking-dependent asymmetry reveals a polarization-governed on-off switching of photoinduced carrier transport, establishing sliding ferroelectricity as a powerful mechanism for dynamically programmable optoelectronic functionality.