Asynchronous Transition Across the Crystal‐Melt Interface Revealed by Machine Learning Potentials

Why It Matters

By linking atomic‑scale dynamics to macroscopic facet selection, the study offers actionable insight for controlling silicon crystal growth, a cornerstone of semiconductor manufacturing.

Key Takeaways

• ML potential reveals >12 Å asynchronous interface zone.
• Structural order changes sharply, density shifts nearer liquid.
• Interfacial free energy anisotropy: γ111 < γ110 < γ100.
• Lower interfacial energy with supercooling reduces nucleation barrier.
• Predicted crystal shapes match observed silicon facet morphology.

Pulse Analysis

Understanding crystal‑melt interfaces has long been a bottleneck in materials engineering, especially for high‑purity silicon used in microelectronics. Traditional empirical potentials struggle to capture the subtle balance of forces at the atomic level, leading to oversimplified models of a sharp, single‑step transition. The advent of machine‑learning potentials trained on first‑principles data provides the fidelity needed to resolve nuanced variations in order, density, and mobility across the interface, opening new avenues for predictive simulations of solidification processes.

The study’s central revelation is that the silicon interface is not a monolithic boundary but a multi‑stage zone where structural ordering, thermodynamic signatures, and kinetic mobility decouple spatially. Using the \(\bar q_6\) order parameter, the authors pinpoint a relatively abrupt structural shift, yet the accompanying entropy, enthalpy, and diffusivity transitions occur farther into the liquid. This asynchronous behavior suggests a staged melting mechanism, where atoms first lose crystalline order before gaining liquid‑like mobility, refining our theoretical picture of phase change dynamics and informing more accurate nucleation models.

Perhaps most consequential for industry is the quantified anisotropy of interfacial free energy—γ111 < γ110 < γ100—which directly rationalizes the prevalence of (111) facets in silicon crystals. The reduction of interfacial energy under supercooling conditions lowers the nucleation barrier, enabling finer control over crystal habit during melt‑growth techniques such as the Czochralski process. By integrating these insights into process simulations, manufacturers can tailor thermal gradients and cooling rates to engineer desired facet orientations, improving wafer quality and yield. Future work will likely extend this methodology to alloy systems and multicomponent melts, further bridging atomistic insight with commercial crystal‑growth optimization.