Systematic Investigation of Grain Size Modulation and Its Impact on Electrical Characteristics of Ferroelectric Hafnia via Deep-Learning-Enabled Analysis

While grain size is a fundamental determinant of ferroelectricity in doped HfO 2 thin films, establishing a unified understanding of its evolution has been hindered by both methodological challenges and divergent experimental reports. Conventional segmentation techniques often struggle with the complex contrast of plan-view electron micrographs, leading to statistical uncertainties that obscure the relationship between the microstructure and its ferroelectric performance. In this work, we address this characterization bottleneck by leveraging a deep neural network (DNN) approach, enabling a highly reliable statistical quantification of grain size distributions across diverse processing regimes. Contrary to the conventional expectation that a higher thermal budget necessarily drives grain growth, our findings reveal that increased deposition temperature ( T dep ) triggers a transition toward a significantly finer and more uniform grain structure. Interestingly, postmetallization annealing (PMA) was found to exert a minimal influence on the established grain size distribution, suggesting that the initial nucleation stage is more decisive than the subsequent thermal treatment. This microstructural transition is directly reflected in the electrical characteristics in that a notable reduction in remnant polarization ( P r ) was observed as grain size decreases, a trend that deviates from the commonly reported grain size effects of ferroelectric hafnia. Meanwhile, the coercive field ( E c ) and its variability show only marginal differences, and the fine-grained morphology correlates with a distinct increase in leakage current density. These findings highlight the necessity of reconsidering microstructural optimization strategies to ensure the performance and scalability of next-generation ferroelectric devices.