Introduction

Abstract

Memory devices are the key components of information technology. The rapid development of information technology requires memory devices to have high density and high speed. Today, the popular memory devices in market are silicon-based devices. There are three leading memories: flash memories, dynamic random access memories (DRAMs), and hard-disk drives (HDDs) [1, 2]. But all of them have obstacles that are difficult to overcome. Silicon-based flash memories have a structure of a metal-oxide-semiconductor field-effect transistor with a floating gate. They are the state-of-the-art nonvolatile memory because of their high density and low cost. Nevertheless, flash memories have problems of low write–erase speeds that are 0.1–1 ms, limited number of rewrite cycles (about 10⁶) and high write voltage (>10 V). DRAMs have advantages of very high switching speed and large number of write–erase cycles, but they are volatile and need to refresh frequently. Although HDDs can have very high data density and many write–erase cycles, they have a severe problem of slow response to the magnetic field. In order to overcome the disadvantages of flash memories, four random access memories (RAMs) have been proposed: ferroelectric RAMs (FRAMs), magnetic RAMs (MRAMs), phase-change RAMs (PRAMs), and resistive RAMs (RRAMs). Among them, FRAMs and MRAMs also have the miniaturization problem like flash memories because of their large memory cell size. The main obstacle for the commercialization of PRAMs is the high power required for the phase transition between the amorphous and crystalline phases. RRAMs emerge as the promising candidates as the next-generation memories. Their size can be scaled down to below 10 nm [3, 4], and their response time can be in nanoseconds (ns) [3–9]. Therefore, they can potentially solve all the technical difficulties faced by flash memories, DRAMs and HDDs.