Automated Phase and Orientation Mapping of Multiphase, Polycrystalline Hafnia-Zirconia Thin Films Using 4D-STEM

Ferroelectric materials have attracted significant attention for applications in memory devices and transistors due to their spontaneous electric polarization that can be switched using an external electric field. Among many potential materials candidates, hafnia-zirconia (HZO) has been seen as a prime candidate material due to its ability to maintain its ferroelectric response in condensed geometries (ultra-thin films) and process compatibility with start-of-the-art CMOS technology. Ferroelectricity in HZO has been attributed to the non-centrosymmetric orthorhombic phase (space group Pca21) which cannot be readily formed at room temperature and requires thermal cycling to convert as-deposited nonpolar phases, e.g., monoclinic (P21/c) and tetragonal (P42/nmc) into the desired orthorhombic phase [1]. Additionally, top and bottom electrode layers like TiN are used to promote the formation of the orthorhombic phase during annealing, which has been attributed to phenomena such as stress-induced crystallization and oxygen scavenging [2]. Currently, these fabrication techniques lack the ability to completely convert all non-polar phases to the polar phase and control the growth orientation, resulting in coexistence of randomly oriented, multiple phases. This poses a challenge for characterizing the distribution and orientation of the polar orthorhombic phase, hindering the development of reliable processing-structure-property relationships for HZO films with systematically controlled processing conditions. High-angle annular dark-field (HAADF) imaging in scanning transmission electron microscopy (STEM) has demonstrated success in determining phases and orientation in HZO thin films directly [3]. However, this technique only allows for determination of a certain grain structure that is aligned along a major zone axis, thus failing to offer comprehensive information of multiple grains. As a result, conventional electron microscopy methods are unable to provide statistically robust conclusions on orientation and phase information from many adjacent grains at once, or the missing analysis on how the orientation and grain structure of the electrode layers may impact the formation of the orthorhombic phase. To overcome this issue, we collect 4-dimensional STEM (4D-STEM) data over a large working area to characterize orientation and phase of a 10 nm-thick HZO thin film between TiN electrodes. Our results confirm the coexistence of monoclinic, tetragonal, and orthorhombic regions with detailed information of their orientations. We investigate the grain structure and orientation in HZO by automatically characterizing each diffraction pattern (DP) in the 4D-STEM dataset. Using a probe size of under 2 nm, we are able to collect isolated diffraction patterns mostly from individual grains. Correlation-based pattern-matching using kinematic DP simulation has been shown to be a successful approach for indexing individual diffraction patterns [4,5]. We use normalized cross-correlation for matching the 4D-STEM dataset with over 9000 dynamical DP simulation results generated using the Bloch wave method [6]. To improve the sensitivity of the cross-correlation calculations to high-index zone diffraction patterns dominated by direct beam intensity, we use methods such as logarithmic scaling of diffraction pattern intensity and disregarding regions of patterns which do not differ significantly between patterns, namely the direct beam and background. Figure 1 shows the schematic used for the correlation-based matching procedure. The radial integration of each experimental and simulation diffraction pattern is computed as depicted in Fig. 1(a). The integrated profiles are then compared to find several close matches as depicted in Fig. 1(b). Finally, the original experimental and simulation DPs are directly compared accounting for rotation as depicted in Fig. 1(c) to find the single best match among the matches identified from the radial integration. Figure. 2(a) shows a virtual annular dark-field (ADF) image of the HZO thin film constructed from the 4D-STEM dataset. Individual diffraction patterns from multiple identified grains in the dataset with the simulation patterns which they were most closely matched are displayed in Fig. 2(b). Visual agreement between experimental and simulated patterns further supports the cross-correlation based matching method's accuracy. We observe regions identified as orthorhombic, monoclinic, and tetragonal. Matching ability suffers in regions with overlapping signal from different structures, such as at the edges of grain boundaries and near the top and bottom TiN electrodes as shown in the final phase map in Fig. 2(c). We will discuss how to address this challenge of characterizing overlapped regions. Schematic illustration of matching procedure. (a) Radial integrations are computed for each experimental and simulation pattern. (b) Radial integrations from experimental and simulation are compared to find the best matches. (c) The original patterns from the best matches found using the radial integrations are matched accounting for rotation to find the final match. (a) Virtual ADF image of the HZO thin film scanned with 4D-STEM. (b) Experimental diffraction patterns (above) in the regions marked with orange, yellow, and pink crosses in (a), and the most closed matched simulated electron diffraction patterns (below). (c) Final phase map showing the distribution of the three phases.