Interpretable structure-property correlation in X-ray diffraction patterns of HfZrO thin films via machine learning

The X-ray diffraction (XRD) patterns of materials contain important and rich information in terms of structure, strain state, grain size, etc. The XRD can become a powerful fingerprint for material characterizations when it is combined with machine learning techniques. Attempts utilizing machine-learning-based methods mainly focus on phase identification for mixture compounds. Herein, we applied a machine-learning-based method linking XRD patterns of HfZrO thin films directly to their electronic properties in experiments. In accordance with conventional understanding, the machine learning model suggests that non-monoclinic (NM) phases of HfO2 and ZrO2 are among the main contributors to higher relative permittivity and lower leakage current. Furthermore, some minor interfacial phases like TiO x and ZrN x are also proposed to be even more important contributors to our target electronic properties. Our research demonstrates that machine learning has the potential to reveal minor XRD signals from sub-1 nm interfacial layers that have long been considered undetectable and thus ignored by human interpretation.