Challenges and Applications of Emerging Nonvolatile Memory Devices

Banerjee

et al. 2021

Self Cite

Electrochemical metallization cell–based threshold switching (TS) devices are promising candidates for selectors in high‐density cross‐point memory arrays. However, TS characteristics in density‐ and stoichiometry‐engineered solid electrolyte systems have not been studied. By adopting TS‐based stoichiometric and substoichiometric solid electrolyte HfO2 layers, the localized atomic scale movement of Ag ions can be effectively controlled in ultrathin bilayers. The stoichiometric HfO2 thickness is crucial to this. This study proposes defect‐ and density‐engineered bilayer TS with a 0.5 nm critical thickness of the stoichiometric HfO2 layer, which maximizes various switching characteristics. The unstable filament in the ultrathin stoichiometric HfO2 layer prevents the formation of stable Ag clusters owing to limited Ag injection into the dense and stoichiometric HfO2 layer. In addition, the substoichiometric HfO1.91 (1.5 nm) buffer layer prevents direct injection of Ag ions from the top electrode into the dense HfO2 layer. These factors enable the bilayer design to achieve a high turn‐off speed of 100 ns, excellent endurance above 107 cycles, low off current of ≈pA, and tight Vth distributions even for sub‐2 nm devices. These exceptional results demonstrate the possibility of designing density‐graded bilayer TS devices with high stability, fast switching, and high endurance.

Section: Improved Threshold Switching and Endurance Characteristics Umentioning

confidence: 99%

Improved Threshold Switching and Endurance Characteristics Using Controlled Atomic‐Scale Switching in a 0.5 nm Thick Stoichiometric HfO₂Layer

Banerjee

et al. 2021

Self Cite

“…It is widely accepted that 6–8 bit‐level RRAM devices are sufficient for a wide range of RAC tasks. [ 29–31 ]…”

Section: Resultsmentioning

confidence: 99%

Memristive Devices with Multiple Resistance States Based on the Migration of Protons in α‐MoO₃/SrCoO_2.5 Stacks

Wang

Huang

Guo

2021

Memristive devices are building blocks for neuromorphic computing. However, non‐ideal properties of memristive devices, such as bad retention, small number of resistance states, and nonlinear pulse programming hinder the development of neuromorphic computation. Based on proton migration in the α‐MoO3/SrCoO2.5 stack, a Pt/α‐MoO3/SrCoO2.5/Nb‐SrTiO3 memristive device is developed with multiple resistance states and excellent nonvolatility. When protons migrate from α‐MoO3 to the SrCoO2.5 lattice, both layers undergo a resistance increase, due to a reduced doping level in α‐MoO3 along with the loss of protons, and a larger direct bandgap of SrCoO2.5 resulted from the insertion of protons. While protons migrate from SrCoO2.5 to α‐MoO3, the device resistance decreases, because of the increased proton concentration in α‐MoO3 and the decreased proton concentration in the SrCoO2.5 layer. The device also realizes nearly linear potentiation and depression under appropriate pulse schemes. A three‐layer backpropagation neural network constructed with the memristive devices acquires an accuracy of 94.3% for the recognition of MNIST handwritten digits.

“…The left panel of Figure 14a illustrates the configuration of a 1S1R integrated cell. Recent years, various selector devices based on various mechanism have been proposed, [68] including Schottky diodes, [64b] Mott metal-insulator transition, [69] ovonic threshold switching, [70] mixed-ionic-electronic conduction, [71] fieldassisted superlinear threshold [72] and redox-diffusive threshold switching. One Schottky diode cascaded with one memristor is confined to unipolar RRAM due to the unilateral conductivity of diodes.…”

Section: Selectorsmentioning

confidence: 99%

Design of a Controllable Redox‐Diffusive Threshold Switching Memristor

Sun

Song

Yin

et al. 2020

In most cases, the current flows uniformly through the device in the HRS and is restricted to a local region with high conductance known as a conducting filament (CF) in the LRS. [3] Among them, a specific memory device is termed as conductivebridge RRAM (CBRAM), where the formation/rupture of metallic conductive filaments are dominated by cation migration and redox processes. [4] Meanwhile, volatile resistance switching behaviors are commonly observed in CBRAM as threshold switching (TS). Analogous to the nonvolatile electrochemical metallization mechanism in terms of materials and structures, [5] the electrical resistance of such a threshold switching memristor (TSM) could decrease by orders of magnitude when an electric field is applied due to the formation of CFs with active metal (such as Ag or Cu) atoms. [6] Differently, the resistance returns back spontaneously after the termination of the external bias, yielding a superior I-V nonlinearity and unique temporal conductance evolution dynamics. [7] Recent years, the threshold switching memristors based on active metals are also called "diffusive memristors" to emphasize the diffusion dynamics of the metal species. [8] To be more comprehensively, redox-diffusive threshold switching memristor (RDTSM) might be the best choice to demonstrate the overall switching behavior. [1a] RDTSM has gained significant attention due to its similar advantages to RRAM, such as the simple structure, great fabricability and integrability, and compatibility with conventional CMOS technology. More importantly, it has great potential in many vital applications, such as two-terminal selectors with high nonlinearity, [9] high-powerefficient synaptic or neuronal devices with novel functions. [8,10] A specific application requires a correspondingly proper device. Figure 1 displays the measurement schematic of the key parameters of RDTSM, including selectivity (or nonlinearity), compliance currents, switching voltages (including threshold voltages and hold voltages),TS mode (unidirectional and bidirectional, as shown in Figure 1a,b, respectively), switching slopes (Figure 1c), response time (including delay and relaxation, as shown in Figure 1d) and endurance, which are all up to the material composition and device structure. [11] Therefore, With the rapid development of information technique in the big-data era, there is an extremely urgent demand for new circuit building blocks, represented by resistive switching memristors with high speed, high-density integration, and power-efficiency, to overcome the limitations of electronic device scaling and thus achieve non-von-Neumann neuromorphic computing. Redox-diffusive threshold switching memristors, based on the volatile formation/rupture of metallic conductive filaments, are attracting great attention for many novel applications, ranging from selectors to synaptic and neuronal devices. Here, how to design a proper redox-diffusive threshold switching memristor is comprehensively introduced, with particular focus on the effect of the devic...