Phase identification of nano grains in epitaxial HfO2-based ferroelectric materials using STEM image deep learning

Abstract

Doped HfO2-based ferroelectric materials have attracted a great deal of attention because they have ferroelectric properties under a film thickness of 10 nm unlike existing ferroelectric materials such as Pb(Zr, Ti)O3, BaTiO3 and BiFeO3. And they are a promising candidate for future ferroelectric field effect transistors and non-volatile memories, as well as ferroelectric tunnel junctions. The most stable structure of HfO2-based materials is monoclinic structure without ferroelectric property, so it is important to stabilize the orthorhombic structure with spontaneous polarization. However, most of the studied HfO2-based materials were polycrystalline films with a mixture of monoclinic, orthorhombic or tetragonal phases, which makes it difficult to keep the uniformity of the properties. Therefore, epitaxially grown orthorhombic single phase HfO2-based film should be acquired. For that it is necessary to analyze the phase distribution, but it is difficult to simply identify the phase using XRD due to similar lattice parameters between phases. Also, direct identification of atomic positions in high resolution scanning transmission electron microscope (HRSTEM) image and further comparison with simulated templates has limitation to be applied since it needs labor of trained human experts for statistical analysis. In this study, we developed HRSTEM image deep learning method to identify phase structure of epitaxial growth HfO2-based materials. For epitaxy growth on a Si substrate, Y-doped ZrO？2(YSZ, Yttria-stabilized zirconia) having a similar lattice constant is applied as a buffer layer, and the YSZ thin film is epitaxy grown on the Si substrate using the PLD process. Afterwards, an HfO2-based ferroelectric thin film was grown epitaxy on YSZ/Si at a deposition temperature of 500~700℃ using an RF sputtering process capable of large-area deposition. And HRSTEM images were obtained using aberration-corrected STEM (Titan S80-300, FEI) to apply for image deep learning, and then we evaluated the crystallinity and electrical properties depending on the growth process conditions.