Ionic modulation and ionic coupling effects in MoS2 devices for neuromorphic computing

154

127

and constant data movement are needed while working. In the human brain, huge and complex neural networks composed of gigantic amounts of neurons massively interconnected by an even larger number of synapses are in charge of computing and memory. Unlike a digital computer, information storage and processing happen at the same time in the brain. [3] Artificial intelligence (AI) that could rival biological neural networks is being continuously pursued by human being. There are two approaches to implement AI: one is the software simulation of neural networks with the aid of digital computers, another is developing hardware-integrated circuits of neural networks. In fact, AI based on the software simulation is evolving very fast thanks to the advances in the development of algorithm in recent years, and it even outperforms the human brain in specific complex tasks, such as playing the game of Go. [4] However, software-based AI suffers from critical issues of enormous power consumption and massive data throughput. Hardware-based AI systems are more efficient in terms of speed and power consumption; however, complementary metal oxide semiconductor (CMOS) transistors are inherently different from the basic building blocks of biological neural networks, neurons, and synapses, in terms of operation dynamic and behavior. A circuit consisting of at least ten transistors are required to realize the function of one biological synapse, which is impractical for the implementation of large networks, not to mention the human brain (roughly 10 11 neurons interconnected by ≈10 15 synapses). [5,6] Fortunately, the emergence of memristive devices with innate dynamics resembling biological synapses and neurons provides a feasible and simple way to build hardware-based bio-inspired computing systems. [7,8] For instance, only one compact memristive device with a metalinsulator-metal (MIM) sandwich structure is sufficient to reproduce some functions of a biological synapse. Moreover, the vector-matrix multiplication, which is believed to be the most data-intensive work in neural networks, can be naturally performed in a cross-bar array of memristive devices on the physical level based on Ohm and Kirchhoff laws. [9,10] These features endow the memristive devices ideally suited to realize highly efficient bioinspired computing systems in hardware.To realize highly efficient neuromorphic computing that is comparable to biological counterparts, bioinspired computing systems, consisting of biorealistic artificial synapses and neurons, are developed with memristive devices with native dynamics resembling biological synapses and neurons. Tremendous materials and devices have been successfully used to emulate diverse functions of synapses, as well as neurons, in the last decade. Herein, approaches to realize certain synaptic or neuronal functions are introduced with state-of-art experimental demonstrations. First, the dynamics and working principles of biological synapses and neurons are briefly presented to provide guidance for developing biore...

Section: Heterosynaptic Plasticitymentioning

confidence: 99%

Memristive Synapses and Neurons for Bioinspired Computing

Yang

Huang

Guo

2019

154

127

“…[140] In-plane electric field controls the direction of Li + ions lateral migration and locally triggers the phase transition from 2H to 1T' in MoS 2 channel. Zhu et al developed two-terminal planar MoS 2 memristive devices by intercalating Li + ions into MoS 2 flake.…”

Section: Planar Devicesmentioning

confidence: 99%

2D Layered Materials for Memristive and Neuromorphic Applications

Wang

Meng

et al. 2019

demonstrated that the memristive devices exhibit many promising features, [3][4][5][6][7][8] such as non-volatility, high speed switching, high endurance, high-density integration, CMOS-compatible fabrication, and so on. These features make them ideal candidates for applications in memory devices, [3,5,9] in-memory computing, [3,[10][11][12] and edge computing. [13,14] The working principle of the TMOs-based memristive devices relies on ion drift or diffusion, which resembles motion of ions in the biological neurons and synapses. The ionic motions exhibit dynamic behaviors. Thus, the memristive devices can be used to emulate the synaptic plasticity [15] and neurons. [16][17][18] Compared to traditional silicon synapses [19][20][21] and neurons [22,23] requiring complex circuits, such emerging memristive devices have compact structures and enhanced func-

“…[88] The emulated neural network has a planar structure, composed of multiple adjacent Au/Li x MoS 2 /Au devices as shown in Figure 9a. [88] The emulated neural network has a planar structure, composed of multiple adjacent Au/Li x MoS 2 /Au devices as shown in Figure 9a.…”

Section: Heterosynaptic Plasticitymentioning

confidence: 99%

“…[88] The emulated neural network has a planar structure, composed of multiple adjacent Au/Li x MoS 2 /Au devices as shown in Figure 9a. [88] The synaptic competition effect was first demonstrated in such a system consisting of two adjacent synaptic devices with a limited amount of Li + ions. Similar to PRP diffusion, the diffusion of Li + ions allows ionic coupling among multiple devices, which is enabled by the layered van der Waals structure in the MoS 2 film that allows the Li + ions to move relatively freely in between the MoS 2 layers.…”

Section: Heterosynaptic Plasticitymentioning

confidence: 99%

“…[53,91] For example, a monolayer MoS 2 -based multiterminal structure has been used to emulate a neural circuit, where the conductance of the MoS 2 film between two inner electrodes (as pre-and postneurons) can be tuned by applying programming pulses on the two outer electrodes (as modulatory neurons). [88] Copyright 2019, Springer Nature. [91] Compared to the implementation of homosynaptic/heterosynaptic plasticity using digital circuits or software-based algorithms, these recent studies show that memristive systems can directly mimic the underlying ionic process in biology and enable diverse synaptic functions with low power and high integration density.…”

Section: Heterosynaptic Plasticitymentioning

confidence: 99%

See 1 more Smart Citation

Nanoionic Resistive‐Switching Devices

Zhu

Lee

Lü

2019

Self Cite

devices have opened up promising opportunities for electronic circuit applications beyond energy storage. One representative example is ionic resistive-switching (RS) devices (also called memristive devices or memristors), [3][4][5] which can be used as a nonvolatile memory element due to its excellent performance such as high switching speed (<1 ns), [6] good retention (>10 years), [7] good endurance (>10 9 ), [8] as well as high device scalability. [9] The RS effect refers to the phenomenon that the resistance of a dielectric thin film, typically sandwiched between two electrodes, can be reversibly modulated by an electric field. [2] In ionic RS devices, the mechanism originates from the rearrangement of atoms/ions in the dielectric thin film through ionic drift/ diffusion and electrochemical processes that lead to the formation/rupture of nanoscale conductive paths (filaments) between the two electrodes. [10] Controlling these internal ionic processes will enable the design and implementation of better engineered devices with improved performance and reliability. Additionally, recent studies show that the internal ionic processes during RS can be used to faithfully emulate many biological effects that are critical for learning and memory, allowing efficient neuromorphic systems to be implemented using solid-state devices and networks. [11][12][13] Controlled ionic processes can also be used to directly modify the composition and/or structure of the material itself at the atomic scale, allowing new nanostructures to be built on the fly, in a reconfigurable fashion.In this review, we will first discuss the switching mechanism of ion-induced RS (memristive) effects in nanoscale solid-state thin films, followed by discussions on controlling and utilization of the ionic effects to implement biological synaptic functions, and to reduce device variation and improve device stability. New concepts of material tuning enabled by controlled ionic process will be introduced at the end. Ion-Induced RS EffectSo far, ion-driven RS effects have been observed in a large variety of materials, ranging from oxides, halides, to chalcogenides, etc. [2,[13][14][15] Based on the types of ions involved during the RS process, the switching mechanism can be divided to three categories: cation-driven RS effects, anion-driven RS effects, and cation and anion jointly driven RS effects.Advances in the understanding of nanoscale ionic processes in solid-state thin films have led to the rapid development of devices based on coupled ionic-electronic effects. For example, ion-driven resistive-switching (RS) devices have been extensively studied for future memory applications due to their excellent performance in terms of switching speed, endurance, retention, and scalability. Recent studies further suggest that RS devices are more than just resistors with tunable resistance; instead, they exhibit rich and complex internal ionic dynamics that equip them with native information-processing capabilities, particularly in the temporal domain. RS effec...