Hafnia-based ferroelectric tunnel junction and its application

Abstract

With the advent of the 4th industrial revolution, IoT, big data, mobile, and artificial intelligence are spotlighted as innovative technologies, and data is rapidly increasing throughout our lives. The storage and processing of such vast amounts of data is implemented using semiconductors, accordingly, development of next-generation information processing and storage devices is urgently required in the semiconductor market. The current state of central processing unit (CPU), logic and memory modules are separated, and the on-chip memory module is made up of volatile devices leading to a lot of power consumption. To solve this problem, a non-volatile logic-in-memory architecture in which non-volatile memory devices are integrated in a logic circuit, and furthermore, an ultra-low power neuromorphic computing architecture that mimics the human brain neural network (Neuromorphic Computing) architecture have been intensively studied. In this dissertation, hafnia-based tunnel junction device which suitable for a nonvolatile logic-in-memory architecture and a neuromorphic computing architecture is developed. Ferroelectric materials with two spontaneous polarization states have attracted marked attention for next-generation non-volatile memory. Compared to conventional perovskite ferroelectric materials, hafnia-based materials have wide bandgaps (>5 eV), scalability, and CMOS compatibility which is great advantages for mass production. Among the electronic devices using ferroelectrics, ferroelectric tunnel junction (FTJ) is 2-terminal resistive switching memory which consist of metal-ferroelectric-metal heterostructure. Hafnia-based FTJs have advantages of low power consumption, and nondestructive readout, and scalability compared to other ferroelectric devices. However, so far, the previous studies on the hafnia-based FTJ have shown reduced properties under 10nm, low on/off resistance ratio, poor reliability, and limitation of large-area arrays due to sneak current which is critical issue for practical application. In this study, to solve these problem, ultra-thin hafnia-based FTJ with high-performance and self-rectifying characteristics which applicable to a next-generation logic-in-memory architecture and neuromorphic computing architecture. First, in order to understand the dependence of the bottom electrode, ferroelectric characteristics and resistance switching properties were evaluated according to various conductors and semiconductor materials. It was found that the bottom electrode had a great influence on the growth and crystallization of the ferroelectric thin film. In particular, as thermal coefficient of bottom electrode, α$_{bottom}$, decrease, remnant polarization value increased owing to large tensile stress induced by bottom electrode. As a result, by stress engineering, ultra-thin (<6nm) film exhibit robust ferroelectricity with high P$_r$ value of 30 μC/cm$^2$. In addition, a high on/off resistance ratio (>100) and improved reliability (>10$^8$) were achieved by reducing leakage current due to diffusion barrier. This layer effectively prevents of diffusion of metal atoms. Lastly, we demonstrate self-rectifying FTJ to minimize sneak-path current in synapse crossbar array without external selector. It is attributed to imprint field effect which induces preferential polarization state and coercive voltage shift along one direction. The ferroelectric tunnel junction device proposed in this study can be used as a highly integrated nonvolatile memory device and a synaptic device for neuromorphic computing. For the synaptic application, long-term potentiation and long-term depression characteristics are evaluated. It was confirmed that the conductivity increases or decreases linearly according to the repetitive input pulse. In addition, high pattern recognition accuracy of 90% was achieved using neural network simulation with a multilayer perceptron. The results of this study verified the possibility of hafnia based FTJ for practical application to nonvolatile logic and neuromorphic systems as a next-generation memory device, which is expected to contribute to the development of the future semiconductor industry.