ALD/ALE 2022 Session AA2-WeM2: ALD for Memory Applications I

Recent discoveries of the ferroelectric properties of doped HfO₂ opens the possibilities of its integration in Ferroelectric-based non-volatile Random-Access Memories (FeRAMs). Replacing memories based on perovskites materials by ferroelectric HfO₂ has many decisive advantages. Indeed, HfO₂ is readily used in CMOS back end of the line, and its use opens the possibility to grow thinner layers of around 10 nm vs 70-100 nm for the conventional ferroelectric perovskites. Thinner layers are mandatory for integration in advanced sub 100 nm CMOS technology nodes.

Plasma Enhanced Atomic Layer Deposition (PEALD) allows the synthesis of a different range of materials such as oxides and metals with low surface roughness and precise thickness control. TiN metal (M) layers and Gd doped HfO₂ insulator (I) layer were grown in the same PEALD chamber without contact with the air between depositions to synthesize ferroelectric MIM capacitors. Changing ratios between Hf and Gd PEALD cycles inside one supercycle allows choosing an appropriate Gd doping concentration. In its order, during annealing, Gd doping in HfO₂ (Gd:HfO₂) leads to a crystallization of HfO₂ in a metastable non-centrosymmetric orthorhombic phase, which induces the HfO₂ ferroelectric properties. A decrease of the ferroelectric oxide thickness can allow operating lower switching voltages for the low power circuits.

Continuing our previous work [SB], by using the growth of MIM capacitor by PEALD in one batch (i.e. without air break between metal, insulator, and metal layers), we recently demonstrated the ferroelectricity of sub-10 Gd:HfO₂ layers (8.8nm-, 6.6nm, and 4.4nm-thick layers) in TiN / Gd:HfO₂ / TiN stacks and studied the remnant polarization amplitude change with the Gd:HfO₂ layer thickness. Structural measurements (X-ray diffraction and reflectometry) confirmed a HfO₂ transformation to the orthorhombic (ferroelectric) phase. Electrical measurements showed that switching voltage can be decreased for the thinner Gd:HfO₂ layers (hysteresis loop measurements and Positive Up Negative Down measurements). Polarization switching cycling measurements demonstrate ferroelectric endurance of at least up to 10⁸ cycles. This work opens the possibilities for the integration of sub-10nm Gd:HfO₂ in the memory device circuits thanks to the unique PEALD deposition capabilities.

References:

[SB] Applied Physics Letters 117.25 (2020): 252903.

Vanadium oxides are much-researched materials due to their wide range of applications from microelectronics, smart electrochromic and thermochromic windows, metamaterials, gas sensors, programmable critical thermal sensors to battery energy storage. V₂O₅ has been the most widely examined for energy storage and electrochromism, as the Li ions can easily be intercalated between its atomic layers. On the other hand, VO₂ undergoes a reversible transition at 68°C, where a change in light and electrical conductivity occurs, therefore, VO₂ is widely researched as a promising material for resistive switching and thermochromism. Resistive switching is presently at the core of emerging technologies, such as memristors and neuromorphic computing. By the switching between the insulating and metallic phase of VO₂, the switching of signals for nanoelectronical applications becomes possible. Alternative switching mechanisms are also widely researched, such as the field-induced transition. An advantage of the ALD method in the preparation of these films would be the control of the deposition parameters and an easy doping, which allows a better control of material properties (crystallinity, stoichiometry and defects), and, thus, the transition temperature.

The atomic layer deposition of vanadium oxides has been in the focus of much research effort. Several precursors and different reactions can be used for this purpose, but there are some substantial difficulties with all of them. Vanadium oxy-tri-isopropoxide, Tetrakis ethylmethyl-amino vanadium (TEMAV) or VCl₄ are all promising candidates for the process.

Most reports on the ALD of vanadium-oxides present the preparation of V₂O₅ which is the more stable phase and is easier to deposit. An advantage of the ALD method is that the oxidation state of the deposited films can be tuned by the selection of the reactant. More oxidising reagents (e.g. oxygen plasma) will result in films with higher oxidation states, while reactions with hydrogen or ammonia plasma yield lower oxidation states. In this way, by choosing the appropriate reactants and deposition parameters the preparation of VO₂ can be achieved.

The present work examined the ALD process of TEMAV and different oxidants (water, oxygen plasma) a number of annealing procedures in oxidising and reducing atmospheres. The films were annealed to improve the crystallinity and stoichiometry. Their properties were examined with Raman spectroscopy and transmission electron microscopy. The switching properties were analysed electrically and optically, and test structures were demonstrated.

Due to its advantages of massive parallelism, high energy efficiency, and cognitive functions, brain-inspired neuromorphic computing is attracting immense interest. As the basic unit cell for learning algorithms, the implementation of synaptic behavior into memristive devices is the key step toward neuromorphic computing.

Recent advances in the performance of resistive random access memory (RRAM) acting as memristive devices have led to a significant interest in neuromorphic applications. Although RRAM based memory arrays demonstrated excellent performance parameters, the variability is still a critical issue. Controlling this intrinsic phenomenon requires employing program-verify schemes. In this talk, an optimized scheme to minimize resistance state dispersion and to achieve reliable multi-bit operation is evaluated.

However, statistical variations can be tolerated in computing applications like neuromorphic networks. The synaptic behavior memristive devices can be evaluated by applying successive algorithms consisting of set or reset pulses. These algorithms can be used to study the synaptic functionality of memristive arrays.

Nevertheless, there is still a huge gap between the physical implementation and the verification of circuits and systems proposed for memristive devices. The first step, required to fill the gap, is the development of analog simulation tools, which are the base for the successful implementation of digital CMOS circuits with memristive elements. New designs and concepts need to be supported up by physical implementation and verification to be reliable. That means, new simulation tools for memristive devices have to address the following issues: device variability, cycling variability, data endurance, data retention as well as device switching speed. Meaning that memristive device models still have a long way to be completed.