AI-Assisted Real-Time Tracking of Subnanometer Strain Relaxation during Heteroepitaxy

Understanding strain relaxation during the initial stages of heteroepitaxy is essential for controlling interfacial structure and film quality, yet direct real-time quantification remains difficult. Here, we report on the demonstration of AutoRHEED, an AI-assisted in situ framework for tracking subnanometer strain relaxation during epitaxial growth. The method integrates image quality filtering, autonomous streak identification, and physics-guided geometric calibration to directly extract quantitative in-plane interplanar spacing (d spacing) evolution from raw reflection high-energy electron diffraction (RHEED) videos. We applied the framework to highly lattice-mismatched AlSb/Si(001) growth and demonstrated that AutoRHEED could capture the transient evolution from the initially strained state to full relaxation with monolayer-level sensitivity. Moreover, the in situ-derived relaxation trajectory shows excellent agreement with ex situ HAADF-STEM, confirming its structural fidelity and temporal reliability.