Vertically Aligned WS₂ Layers for High‐Performing Memristors and Artificial Synapses

Abstract

functionalities could be mimicked by using and fine tuning (by spikes) the conductance of two-terminal memristive devices. [4,6,7] Therefore, a broad range of materials, for instance, metal oxides, organic/inorganic perovskites, 2D layered materials, etc., have been used to build memristive devices, which were further employed as artificial synapses. [8][9][10][11] Despite the achievements made so far, the immediate development in particular for metal-oxide-based artificial synapses faces several major problems: a great degree of variability (both from device to device and cycle to cycle) and trivial range of linearly programmable conductance states. [12] All these obstacles must be resolved to implement error-free neuromorphic functioning. These objectives could be achieved by either exploring new material architecture or improving the performance of existing ones with a detailed understanding of fundamental change dynamics.The realization of memristive properties in metal oxide hinges on the distribution of well-known intrinsic defects or ions, mainly oxygen vacancies (O V ). [10,11,13,14] Generally, under the influence of applied electric field, local oxygen vacancy density distribution changes and thus modifies the total (two-terminal) resistance of the device. [6,15,16] Therefore, to realize the reproducible (cycle-to-cycle) and stable (device-to-device) performance, it is essential to have a better control on the oxygen vacancy/ion distribution and its movement with applied field. As a matter of fact, researchers have made several attempts to confine the oxygen ion dynamics along the preferential sites. For instance, the insertion of metal nanodots or nanoparticles, and embedded nanotip electrodes have been found to be effective in improving the cycle-to-cycle uniformity. [17][18][19] However, large size and random distribution of metal nanoparticles generate hindrance to realize reproducible performance from device to device over a larger area. Indeed, the key challenges to design metal-oxidebased artificial synapse are to have reproducible and robust (against electric pulses) performance along with large number of linearly programmable states, which is yet to be achieved.As a promising strategy, the insertion of 2D layered materials into memristive device structure offers a new possibility to improve the performance. In this scenario, few attempts have been made; however, most of them used planar configurations, which occupy relatively large space and are difficult to stack in 3D Inspired by the human brain, the quest for high-performing neuromorphic architecture has recently gained more attention, which can be achieved by two-terminal memristors. However, due to random and uncontrolled filament formation during a typical switching process, conventional memristors suffer from severe shortcomings such as temporal/spatial reproducibility as well as trivial sensitivity against applied spikes, however all these properties are crucial for accurate and quick information processing. Here, reproducible and robust...

Two-Dimensional Near-Atom-Thickness Materials for Emerging Neuromorphic Devices and Applications

“…Kumar et al. demonstrated robust two-terminal memristors based on zinc oxide (ZnO) and WS 2 layers prepared by a radio frequency sputtering process ( Kumar et al., 2019 ). The interlayer separation between ZnO and WS 2 layers provided an effective porous medium for the growth of defective ZnO (Zn metal-rich and non-uniformly distributed oxygen vacancies), which provided a unique charge trapping/detrapping platform layer ( Kumar et al., 2019 ).…”

“… (E) Demonstration of STDP learning rule in a Ag/ZnO/WS 2 /Al synapse indicating the viability of 2D material-based synapses for unsupervised learning. Reprinted with permission from ( Kumar et al., 2019 ). Copyright 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.…”

“…STDP learning is associated with the unsupervised learning of the ANNs. A representative plot of the STDP, as demonstrated in a two-terminal ZnO/WS 2 /Al synaptic device, is shown in Figure 6 E ( Kumar et al., 2019 ). Another study on memristive switching in Ag/MoO x /MoS 2 /Ag vertical stacks reported their synaptic plasticity characteristics like STP and LTP, achieved at an operating voltage of 0.1 V ( Bessonov et al., 2015 ).…”

“…These devices also exhibited the STDP learning rule required for unsupervised learning. Furthermore, synaptic characteristics such as pulse paired facilitation (PPF), STP, LTP, and STDP were observed in graphene/WSe 2 /WO 3-x /Ag ( Huh et al., 2018 ) and ZnO/WS 2 /Al memristive devices ( Kumar et al., 2019 ). However, all the aforementioned device demonstrations employed mechanically exfoliated 2D flakes of small dimensions.…”

“…Their desirable properties, including atomic thickness, dangling-bond-free surfaces, mechanical strength, high integration density, tunable electrical transport, and optical properties, as well as low energy consumption, make them ideal candidates for applications in a wide range of electronic devices ( Gupta et al., 2015 ; Mas-Ballesté et al., 2011 ; Xia et al., 2017 ). More recently, the applications of 2D materials have been extensively studied for energy-efficient and high-performing artificial synapses ( Arnold et al., 2017 ; Chen et al., 2019c ; Dev et al., 2020 ; Hu et al., 2019 ; Jiang et al., 2017 ; Kalita et al., 2019 ; Kim et al., 2019c ; Krishnaprasad et al., 2019 ; Kumar et al., 2019 ; Li et al., 2018 ; Liu et al., 2019 ; Mao et al., 2019 ; Paul et al., 2019 ; Pradhan et al., 2020 ; Xie et al., 2018a , 2018b ; Xu et al., 2019 ; Yan et al., 2019a ; Yi et al., 2018 ; Zhu et al., 2019 ). Furthermore, owing to their dangling-bonds-free surface and atomically thin nature, a variety of 2D materials-based heterostructures have been developed in spite of their lattice mismatch ( Novoselov et al., 2016 ).…”