Investigating the influence of microstructure and grain boundaries on electric properties in thin film oxide RRAM devices – A component specific approach

At the beginning of the 21st century, the quest for finding ever more power efficient, densely packed, and multi-bit-level storage for computational applications is still ongoing. Ever increasing demand in low power computing since the advent of the internet of things (IoT), scaling limitations contradicting Moore’s law, the rise of neuromorphic computing and in-memory computing turned a spotlight onto material classes that seem to tick all the boxes of these requirements, such as transition metal oxides. Predicted in 1971, now 50 years ago, Chua described a missing circuit element, that would complete the well-known list of resistor, capacitor and inductor: the memristor. Its behavior is that of a “nonlinear resistor with memory,” lending their names for the contraction naming of the postulated two-terminal circuit element. It took 37 years until Strukov et al. exclaimed “The missing memristor [being] found” in 2008, with their realization of a two-terminal memristor being implemented as a metal-insulator-metal (MIM) stack consisting of a Pt/TiO2 x/Pt stack. By biasing the 5 nm oxide film, the resistivity could be changed between a high and low resistive state in a reversible and non-volatile process. Similar electrical behavior was experimentally observed for organic, amorphous and (single) crystalline oxide thin films as well as for Chalcogenide-metal stacks. Depending on the material class that is involved in the observed “negative resistive change,” several groups of memristive systems (also referred to as RRAM, as they are candidates for resistive random-access memory applications) can be defined. Common between “filamentary switching” material systems is the confinement of the volume of the thin film that takes part in the resistive switching process. Insight into these nanoscale events involved in the localization of these regions is a key step towards the full understanding of the fundamental physical processes involved. The primary goal in the conjunction of research on resistive switching thin films and high-resolution microscopy is to image a working filament. Transmission Electron Microscopy (TEM) methods including the local mapping of structural and spectroscopic properties, Conductive Atomic Fore Microscopy (CAFM) as well X-ray microscopy methods were and still are at the forefront of this ongoing effort. Based on the findings of grain boundaries serving as preferential filament formation sites in oxide thin films, ab initio methods confirmed this structural feature’s unique role in VCM RRAM. Lab scale, photolithography-based device sizes range from several 100 µm² to 10 µm² and finding a predicted filament of only a few square nanometers in square micron sized areas is a “needle in a haystack” problem. One way to facilitate finding the filament is the creation of the filament at a predefined position. In the case of CAFM, this is inherently part of the experimental setup, but in the case of TEM, a whole new set of methods had to be developed to be able to apply a bias on a thin film stack inside a microscope. In this work, parallel to the implementation of a preparation routine enabling the operation of an RRAM device inside a TEM, two components of TiN/HfO2/Pt MIM stacks have been investigated in detail: firstly, the TiN bottom electrode, which is an integral part in a working device due to its defining microstructural features. High substrate temperature growth of TiN on c-cut sapphire substrates showed exceptional room temperature as well as superconducting properties. The low surface roughness and nitrogen deficiency highlight the aptitude for highly textures TiN layers to act as bottom electrodes in resistive switching devices. Secondly, the HfO2 layer itself, whose arrangement of textured grains and resulting grain boundaries have been investigated at nanometer and sub-ångström resolution. It was possible to identify the terminating planes of monoclinic hafnia grains, which subsequently have been used as a basis for ab initio modeling the grain boundary structure. Identification of sub-stoichiometric and stoichiometric phases in 10 nm films was a second achievement obtained in the study of the oxide film. To probe the crystallographic phase, stoichiometry, and orientation of individual grains, the methods chosen in the TEM studies cover techniques fundamentally driven by electron diffraction and incoherent imaging. Two technological revolutions are currently pushing (transmission) electron microscopy into a new era: new in situ specimen holders and 4D-STEM data acquisition. First, modern, micro-electro-mechanical-systems (MEMS) based in situ holders allow the (simultaneous) application of biasing and temperature stimuli as well as the introduction of controlled gaseous and liquid environments to investigate TEM samples as close to real world conditions as possible. Second, electron detector technology is being accelerated by introducing hybrid pixel detectors that allow the fast and high dynamic range capturing of pixelated reciprocal space and spectroscopic data. Both technological developments are addressed as part of this work, as the implementation of preparation routines for in situ experiments and the integration of a 4D-STEM detector were undertaken as part of this work. At the same time, these developments are crucial methodologies that enable the study of functional thin films, with this work focusing on hafnia-based VCM RRAM. 4D-STEM specifically is proving to be an advanced method to investigate the local structure and response to external stimuli to an extent never achieved before. Consequently, the quantities of data created during these experiments (external stimuli add more dimensions to 4D-STEM experiments) in need fundamentally new approaches to data analysis. On the journey towards an all-encompassing experiment, this work aims to establish the necessary tools, both in hardware and software, and the fundamental material scientific understanding to create a unique and state-of-the-art toolset that allows proceeding in the vast field of oxide-based electronic thin films