betaFeedback

WorksPhysics and Astronomy

The Role of Mobile Point Defects in Two-Dimensional Memristive Devices

Abstract

Two-dimensional (2D) layered transition metal dichalcogenides (TMDCs) are promising memristive materials for neuromorphic computing systems as they could solve the problem of the excessively high energy consumption of conventional von Neumann computer architectures. Despite extensive experimental work, the underlying switching mechanisms are still not understood, impeding progress in material and device functionality. This study reveals the dominant role of mobile defects in the switching dynamics of 2D TMDC materials. The switching process is governed by the formation and annihilation dynamics of a local vacancy depletion zone. Moreover, minor changes in the interface potential barriers cause fundamentally different device behavior previously thought to originate from multiple mechanisms. The key mechanisms are identified with a charge transport model for electrons, holes, and ionic point defects, including image-charge-induced Schottky barrier lowering (SBL). The model is validated by comparing simulations to measurements for various 2D MoS2_2-based devices, strongly corroborating the relevance of vacancies in TMDC devices and offering a new perspective on the switching mechanisms. The insights gained from this study can be used to extend the functional behavior of 2D TMDC memristive devices in future neuromorphic computing applications.

Introduction

Artificial intelligence (AI) is regarded as one of the most important emerging technologies of the 2020s and a key enabler for a variety of disruptive innovations. It will continue to have major implications on the worldwide economy. Even if not all optimistic expectations of the early days will be fulfilled, AI will likely change the way people live and work to a greater degree than the introduction of the personal computer, the internet, or the smartphone did in the past [1]. To unfold its full potential, AI requires suitable hardware, which, unfortunately, presents a major obstacle [2]. The conventional hardware for information processing, i.e., computers based on digital CMOS (complementary metal-oxide-semiconductor) logic and the von Neumann architecture, is far from ideal for AI applications. In particular, the energy consumption of information processing in AI systems based on this hardware is unacceptably high. We note that the energy consumption of AI is much more than a matter of merely academic interest. Instead, serious concerns exist about the rapidly growing carbon footprint of AI hardware and its harmful impact on the environment and the global climate. [3, 4, 5] For example, it is estimated that training ChatGPT may have consumed as much electric power as 175,000 Danes within a month [6].

An alternative, significantly more energy-efficient, and thus more sustainable solution for many AI problems is using hardware that processes information similarly to biological brains, i.e., neuromorphic computing [7, 8]. Such hardware can be realized using memristive devices, for short memristors, as a fundamental building block. Memristors permit emulating the operation of synapses and neurons, the basic elements of biological brains, in an elegant way. Their current-voltage (I-V) characteristics are hysteretic; therefore, their resistance can be modulated by applying voltage pulses. So far, most memristors have been vertical two-terminal metal-insulator-metal structures [9, 10]. These devices have been used in nonvolatile memory architectures, frequently referred to as RRAM (resistive random-access memory) or memristive RAM [11].

Recently, memristors made of 2D and quasi-2D layered materials have attracted considerable attention [12, 13]. A promising class of layered materials for memristors are transition metal dichalcogenides (TMDCs) such as MoS2_2, MoSe2_2, WS2_2, and WSe2_2. Both vertical and lateral TMDC memristors have been demonstrated, each offering specific beneficial features. Vertical TMDC memristors comprise a layer sequence of a top contact (metal), active material (TMDC), and bottom contact (metal) [14, 15, 16, 17]. They show excellent scaling potential, can easily be implemented in crossbar structures, and offer memristive switching at low voltages. For example, vertical electroforming-free MoS2_2 memristors with switching voltage swings ΔU\Delta U (ΔU=Uset−Ureset\Delta U = U_{\mathrm{set}} - U_{\mathrm{reset}}, where UsetU_{\mathrm{set}} is the set voltage and UresetU_{\mathrm{reset}} is the reset voltage) in the range 160 - 550 mV have been reported [14, 15, 16], while the lowest reported voltage swings for traditional vertical MIM (metal-insulator-metal) memristor voltages are around/above 1 V [18, 19]. Lateral TMDC memristors (Figure 1), on the other hand, have a larger footprint and require higher switching voltages than their vertical counterparts but offer a higher degree of functionality since lateral structures can be equipped with additional terminals such as top or bottom gates. These gates permit adjusting the device properties and enable several desirable performance features such as improved device linearity and symmetry, increased number of resistance states, controllable synaptic weight, and complex neuromorphic learning [20, 21, 22, 23]. Thus, lateral TMDC memristors, whether equipped with additional terminals or not, show promise for memristive neuromorphic systems for future AI solutions. Unfortunately, the physics of lateral TMDC memristors is not yet understood, and even the mechanisms responsible for resistive switching in these devices are still debated. Indications for three different switching mechanisms were found experimentally.

Schematic of three proposed hysteresis mechanisms in a lateral two-dimensional memristive device

Figure 1. Example of a lateral memristive device structure comprising two electrodes (source and drain) on top of a two-dimensional memristive material. The three illustrated mechanisms have been suggested as the origin of hysteresis in lateral memristive devices based on 2D TMDC materials: (a) the charging and discharging dynamics of trap states [12, 13, 14, 15, 16, 17, 18], (b) the migration of charged point defects in the electric field [34, 36], and (c) the voltage-induced phase change from a stable, low-conductive semiconducting phase to a metastable, high-conductive metallic phase [39, 41].

The first switching mechanism (Figure 1a) is connected with the charging and discharging dynamics of trap states [24, 25, 26, 27, 28, 29, 30]. These states could be introduced by defects far away from the electrodes (in the bulk) or at the semiconductor-electrode interfaces [31, 32]. They can capture and release electrons or holes with a probability depending on the applied voltage, thereby altering the bulk and interface electrostatic potential, free charge carrier density, and contact barriers [33]. Experimental investigations based on current-voltage measurements and analytical approximations resulted in estimated trap area densities between ≈ 101210^{12} cm−2^{-2} [24] and ≈ 101510^{15} cm−2^{-2} [27] for MoS2_2. The second mechanism (Figure 1b) is based on charged point defects sufficiently mobile to drift in the applied electric field. Experimental indications were found for the accumulation and drift of sulfur vacancies along grain boundaries in the conducting channel of lateral monolayer MoS2_2 devices [34]. This observation is consistent with other experimental work [31, 35] and was correlated with hysteresis in the current-voltage characteristics [34]. Elsewhere [36], measurements were performed on exfoliated MoS2_2 multilayers, which were plasma treated to increase the concentration of sulfur vacancies. The homogeneous drift of mobile sulfur vacancies in the conduction channel was confirmed by local Auger-electron spectroscopy and experimentally correlated with hysteresis in the current-voltage characteristics [36]. Sulfur vacancies could behave as n-type dopants in MoS2_2 [37, 38], thereby changing the conductivity in the bulk and modifying the potential energy barrier at the semiconductor-metal interfaces. The third mechanism (Figure 1c) is based on an electric-field-induced transition from a low-conductive semiconducting phase to a highly conductive metallic phase of the TMDC material. Such an effect has been demonstrated in lateral MoTe2_2 devices, where a large electric field caused a transition from the semiconducting 2H phase to a metastable high-conductive 1T’ metallic phase [39], and in 2H-Te2_2 with a transition from the 2H phase to an unstable distorted 2Hd_\mathrm{d} phase [40]. In MoS2_2, this effect was achieved indirectly by a pre-intercalation of Li+^+ ions, which are highly mobile in between the monolayer sheets and induced the 2H to 1T’ transition upon drift and agglomeration in the applied electric field [41]. Despite this experimental progress, the switching mechanism in 2D TMDC devices lack a profound physical understanding.

In particular, the first two mechanisms seem promising for application [42] but are superficially similar because the change in conductance is based on a change in the charge density in both cases. Further, the two mechanisms could be coupled and occur in parallel since point defects can introduce additional electronic states [43, 44]. These similarities make distinguishing the two mechanisms challenging or even impossible with experimental correlation studies. While trap-state-based hysteresis is assumed in most work on lateral devices to explain resistive switching [12, 13, 14, 15, 16, 17, 18], it is unclear whether or not the drift of charged point defects could contribute as well or even dominate the hysteresis in some cases. Note that quantitative physical models could be powerful in separating such effects, which are experimentally difficult to access or distinguish. However, the few reported models of lateral devices are based on analytical approximations of the I-V curve with many fitting parameters [20] or compact-model formulations [45]. While such models can be helpful in circuit simulations and design, they do not capture the complex dynamics of the switching process.

In this work, we analyze the role of mobile point defects in the switching mechanism of 2D memristive devices and demonstrate the dominant impact they can have on the hysteresis and the device characteristics. A charge-transport model is introduced, solved via a finite-volume discretization, and compared to experimental data of lateral memristive devices based on exfoliated MoS2_2. The model self-consistently solves semi-classical equations for the electrostatic potential and quasi-Fermi potentials of electrons, holes, and ionic sulfur vacancies. Additionally, we develop a mathematical formulation of image-charge-induced Schottky barrier lowering (SBL). This effect describes the change of Schottky barriers caused by an applied voltage and a redistribution of charge carriers in the semiconductor [46, 47]. While SBL is often considered negligible [47], Kelvin probe force microscopy measurements on lateral devices demonstrated significant barrier-lowering effects [48]. Hence, SBL could lead to hysteresis by altering the interface-limited conduction. Despite its fundamental character, up to now, there is no mathematical formulation of this effect for charge-transport models, and the role of SBL in the hysteresis is still remains unclear.

Charge Transport Model

This section presents a semi-classical charge transport model. The equations and boundary conditions are detailed in the following, and the underlying assumptions are discussed. First, we present a system of equations without SBL in Subsections 2.1 and 2.2 before including SBL in Subsection 2.3.

Bulk Equations

Without the effect of SBL, the charge-transport model comprises a system of four coupled nonlinear partial differential equations: three drift-diffusion equations for the quasi-Fermi potentials φα(x,t)\varphi_{\alpha}(\boldsymbol{x},t), α∈{n,p,x}\alpha\in\{\mathrm{n},\mathrm{p},\mathrm{x}\} of electrons n, holes p, and mobile defects x, as well as Poisson’s equation for the electrostatic potential ψ(x,t)\psi(\boldsymbol{x},t). The corresponding densities of the charge carriers are denoted by nα(x,t)n_{\alpha}(\boldsymbol{x},t), α∈{n,p,x}\alpha\in\{\mathrm{n},\mathrm{p},\mathrm{x}\}. For the location vector x\boldsymbol{x} and the time tt it is (x,t)∈Ω×[0,te](\boldsymbol{x},t) \in\Omega\times[0,t_{\mathrm{e}}] for the spatial domain of the semiconductor Ω⊂Rd\Omega\subset\mathbb{R}^{d} (d=1,2,3d = 1,2,3) and some given end time te>0t_{\mathrm{e}} > 0.

Continuity equations. The electron, hole, and ionic defect densities nnn_{\mathrm{n}}, npn_{\mathrm{p}}, and nxn_{\mathrm{x}} satisfy the conservation of mass via the continuity equations

qα∂tnα+∇⋅jα=qαrα,withα∈{n,p,x},(1)q_{\alpha}\partial_{t}n_{\alpha}+\nabla\cdot\boldsymbol{j}_{\alpha}=q_{\alpha}r_{\alpha},\quad\text{with}\quad\alpha\in\{\mathrm{n},\mathrm{p},\mathrm{x}\}, \tag*{(1)}

where the charge qαq_{\alpha} of the respective carrier species α\alpha is given by

qα=zαq,withα∈{n,p,x},(2)q_{\alpha}=z_{\alpha}q,\quad\text{with}\quad\alpha\in\{\mathrm{n},\mathrm{p},\mathrm{x}\}, \tag*{(2)}

with the positive elementary charge qq. The charge numbers are zn=−1z_{\mathrm{n}}=-1 and zp=+1z_{\mathrm{p}}=+1 for electrons and holes, respectively, and zx∈Zz_{\mathrm{x}}\in\mathbb{Z} for mobile defects. The source term on the right-hand side of Eq. (1) includes the production/reduction rates rαr_{\alpha}, which can be used to consider recombination and generation processes of charge carriers. The current densities jα\boldsymbol{j}_{\alpha} are proportional to the negative gradient −∇φα-\nabla\varphi_{\alpha} of their respective quasi-Fermi potentials φα\varphi_{\alpha}, and are given by

jα=−zα2qμαnα∇φα,withα∈{n,p,x},(3)\boldsymbol{j}_{\alpha}=-z_{\alpha}^{2}q\mu_{\alpha}n_{\alpha}\nabla\varphi_{\alpha},\quad\text{with}\quad\alpha\in\{\mathrm{n},\mathrm{p},\mathrm{x}\}, \tag*{(3)}

with the charge carrier mobilities μα\mu_{\alpha}. The charge carrier densities nαn_{\alpha} can be linked to the respective quasi-Fermi potentials φα\varphi_{\alpha} and the electrostatic potential ψ\psi via the state equations [49]

nα(ψ,φα)=NαFα(ηα),withηα=qα(φα−ψ)+zαEα,0kBT,andα∈{n,p,x},(4)n_{\alpha}(\psi,\varphi_{\alpha})=N_{\alpha}\mathcal{F}_{\alpha}(\eta_{\alpha}),\quad\text{with}\quad\eta_{\alpha}=\frac{q_{\alpha}(\varphi_{\alpha}-\psi)+z_{\alpha}E_{\alpha,0}}{k_{\mathrm{B}}T},\quad\text{and}\quad\alpha\in\{\mathrm{n},\mathrm{p},\mathrm{x}\}, \tag*{(4)}

with the temperature TT, the Boltzmann constant kBk_{\mathrm{B}}, and the effective densities of state of the conduction band and valence band Nc=NnN_{\mathrm{c}}=N_{\mathrm{n}} and Nv=NpN_{\mathrm{v}}=N_{\mathrm{p}}, respectively. For the statistics functions Fn\mathcal{F}_{\mathrm{n}} and Fp\mathcal{F}_{\mathrm{p}} of electrons and holes, we use the Fermi-Dirac integral of order 1/21/2 and for mobile defects the Fermi-Dirac integral of order −1-1, i.e., Fx(η)=1/(exp⁡(−η)+1)\mathcal{F}_{\mathrm{x}}(\eta)=1/(\exp(-\eta)+1). The latter function is selected because it provides an upper limit for nxn_{\mathrm{x}} and can be shown to be thermodynamically consistent [50]. Therefore, NxN_{\mathrm{x}} can be physically interpreted as the maximum possible density of mobile defects prescribed by the material, and Ex,0E_{\mathrm{x},0} as an intrinsic energy level of the defects [50]. Further, Ec,0=En,0E_{\mathrm{c},0}=E_{\mathrm{n},0} and Ev,0=Ep,0E_{\mathrm{v},0}=E_{\mathrm{p},0} denote the intrinsic band edge energies of the conduction and valence bands for electrons and holes, respectively. These energies relate to the electron affinity χe\chi_{\mathrm{e}} and the bandgap EgE_{\mathrm{g}} via Ec,0:=−χeE_{\mathrm{c},0}:=-\chi_{\mathrm{e}} and Ev,0:=−χe−EgE_{\mathrm{v},0}:=-\chi_{\mathrm{e}}-E_{\mathrm{g}}. Consistent with Eq. (4), the electric potential dependent conduction band edge EcE_{\mathrm{c}}, valence band edge EvE_{\mathrm{v}}, and vacancy energy level ExE_{\mathrm{x}} are defined as

Eα:=qnψ+Eα,0withα∈{c,v},(5)E_{\alpha}:=q_{\mathrm{n}}\psi+E_{\alpha,0}\quad\text{with}\quad\alpha\in\{\mathrm{c},\mathrm{v}\}, \tag*{(5)}
Ex:=qxψ−zxEx,0.(6)E_{\mathrm{x}}:=q_{\mathrm{x}}\psi-z_{\mathrm{x}}E_{\mathrm{x},0}. \tag*{(6)}

Poisson equation. Finally, the three continuity equations (Eq. (1)) are self-consistently coupled via the electrostatic potential ψ\psi to the nonlinear Poisson equation

−∇⋅(ε∇ψ)=qCC+∑α∈{n,p,x}qαnα(ψ,φα).(7)-\nabla\cdot(\varepsilon\nabla\psi)=q_{\mathrm{C}}C+\sum_{\alpha\in\{\mathrm{n},\mathrm{p},\mathrm{x}\}}q_{\alpha}n_{\alpha}(\psi,\varphi_{\alpha}). \tag*{(7)}

In this equation, CC is either an effective acceptor or donor background dopant density with dopant charge qC=zCqq_{\mathrm{C}}=z_{\mathrm{C}}q, described by the background dopant charge number zC∈{−1,1}z_{\mathrm{C}}\in\{-1,1\}. The quasi-static electric permittivity ε=ε0εr\varepsilon=\varepsilon_{0}\varepsilon_{\mathrm{r}} is expressed as the product of the electric vacuum permittivity ε0\varepsilon_{0} and the quasi-static relative electric permittivity εr\varepsilon_{\mathrm{r}}.

Boundary Conditions at Metal-Semiconductor Contacts

The metal-semiconductor contacts Γ\Gamma, are assumed to represent physical barriers for mobile defects, and therefore, we impose the zero-flux condition (for x∈Γ\boldsymbol{x} \in\Gamma)

jx(x,t)⋅ν(x)=0,(8)\boldsymbol{j}_{\mathrm{x}}(\boldsymbol{x},t) \cdot\boldsymbol{\nu}(\boldsymbol{x}) = 0, \tag*{(8)}

with the outward directed unit normal vector ν\boldsymbol{\nu} of the semiconductor domain. For electrons and holes, thermionic emission is described with the flux boundary conditions [51]

jn(x,t)⋅ν(x)=qnvn(nn(x,t)−nn,0),\boldsymbol{j}_{\mathrm{n}}(\boldsymbol{x},t) \cdot\boldsymbol{\nu}(\boldsymbol{x}) = q_{\mathrm{n}}v_{\mathrm{n}}\left(n_{\mathrm{n}}(\boldsymbol{x},t)-n_{\mathrm{n},0}\right),
jp(x,t)⋅ν(x)=qpvp(np(x,t)−np,0).(9)\boldsymbol{j}_{\mathrm{p}}(\boldsymbol{x},t) \cdot\boldsymbol{\nu}(\boldsymbol{x}) = q_{\mathrm{p}}v_{\mathrm{p}}\left(n_{\mathrm{p}}(\boldsymbol{x},t)-n_{\mathrm{p},0}\right). \tag*{(9)}

Here, vnv_{\mathrm{n}} and vpv_{\mathrm{p}} are the electron and hole recombination velocities given by (with hh as Planck constant)

vn=4πmn∗(kBT)2h3Nc,vp=4πmp∗(kBT)2h3Nv,(10)v_{\mathrm{n}} = \frac{4\pi m_{\mathrm{n}}^{*}(k_{\mathrm{B}}T)^2}{h^3N_{\mathrm{c}}}, \qquad v_{\mathrm{p}} = \frac{4\pi m_{\mathrm{p}}^{*}(k_{\mathrm{B}}T)^2}{h^3N_{\mathrm{v}}}, \tag*{(10)}

with the effective mass mn∗m_{\mathrm{n}}^{*} and mp∗m_{\mathrm{p}}^{*} of electrons and holes, respectively. The corresponding quasi-equilibrium carrier densities nn,0n_{\mathrm{n},0} and np,0n_{\mathrm{p},0} at the contacts are

nn,0=NcFn(−ϕ0kBT),np,0=NvFp(−Eg−ϕ0kBT),(11)n_{\mathrm{n},0} = N_{\mathrm{c}}\mathcal{F}_{\mathrm{n}}\left(-\frac{\phi_0}{k_{\mathrm{B}}T}\right), \qquad n_{\mathrm{p},0} = N_{\mathrm{v}}\mathcal{F}_{\mathrm{p}}\left(-\frac{E_{\mathrm{g}}-\phi_0}{k_{\mathrm{B}}T}\right), \tag*{(11)}

where ϕ0=ϕ0(x)>0\phi_0 = \phi_0(\boldsymbol{x}) > 0 is an intrinsic Schottky energy barrier constant in time and Eg=Ec,0−Ev,0E_{\mathrm{g}} = E_{\mathrm{c},0} - E_{\mathrm{v},0} the band gap. For the electrostatic potential at the metal-semiconductor contacts (x∈Γ\boldsymbol{x} \in\Gamma) we apply the Dirichlet condition

ψ(x,t)=ψ0(x)+ψa(x,t),(12)\psi(\boldsymbol{x},t) = \psi_0(\boldsymbol{x}) + \psi_{\mathrm{a}}(\boldsymbol{x},t), \tag*{(12)}

with ψ0=(ϕ0−Ec,0)/qn\psi_0 = (\phi_0 - E_{\mathrm{c},0})/q_{\mathrm{n}} denoting an intrinsic electrostatic potential barrier, and ψa\psi_{\mathrm{a}} denoting an applied electrostatic potential at the contact. The intrinsic Schottky barrier ϕ0\phi_0 is considered as a material and device parameter, i.e., for a given device, ψ0\psi_0 and ϕ0\phi_0 depend only on the location x\boldsymbol{x}. Zero-flux conditions can be used at all non-contact boundaries.

Image-Charge Induced Schottky Barrier Lowering

The two subsections above describe a system where the electrostatic potential at the boundary is prescribed by an intrinsic Schottky barrier and only the applied electrostatic potential ψa\psi_{\mathrm{a}} adds a time-dependent contribution. However, any point charge present in the semiconductor induces a charge of opposite polarity (image charge) in the metal electrode, and this adds a contribution ϕi\phi_{\mathrm{i}} to the potential energy of the charge (image-charge potential energy ϕi\phi_{\mathrm{i}}) with an energy minimum at the contact interface [51] as schematically illustrated in Figure 2a. The image-charge potential energy ϕi\phi_{\mathrm{i}} superposes with the residual potential energy $\phi_{\mathrm{r}} := qnψrq_n\psi_r (ψr\psi_r: residual potential), which corresponds to the potential energy without image charge. As a result, the barrier maximum of the total potential energy at the interface is reduced by a value Δϕ\Delta\phi compared to the case without image-charge contribution (Figure 2b). While the image-charge potential energy is constant in time, ϕr\phi_r depends on the charge distribution within the semiconductor and the applied potential ϕa≔qnψa\phi_a \coloneqq q_n\psi_a, which can correspondingly influence Δϕ\Delta\phi.

Schematic illustration of an electron point charge and image-charge-induced barrier lowering

Figure 2. Schematic illustration of (a) an electron point charge qnq_{\mathrm{n}} at the semiconductor-metal interface inducing an image charge −qn-q_{\mathrm{n}} in the metal electrode, which results in an attractive electrostatic potential energy ϕi\phi_{\mathrm{i}} (image-charge potential energy), and (b) illustration of the superposition of the residual potential energy ϕr\phi_{\mathrm{r}} and the image-charge potential energy ϕi\phi_{\mathrm{i}}, which reduces the interfacial potential energy barrier maximum by Δϕ\Delta\phi relative to the value of ϕr\phi_{\mathrm{r}} at the interface. The boundary value of ϕr\phi_{\mathrm{r}} at the interface is prescribed by the sum of the intrinsic Schottky energy barrier ϕ0\phi_0 and the applied potential energy ϕa\phi_{\mathrm{a}} (ϕa=qnψa\phi_{\mathrm{a}} = q_{\mathrm{n}}\psi_{\mathrm{a}}, with the applied electrostatic potential ψa\psi_{\mathrm{a}}). Note that an offset of approximately 0.16 eV is added to ϕi\phi_{\mathrm{i}} for illustrative purposes and ϕi=0\phi_{\mathrm{i}} = 0 eV far away from the electrode, i.e., at x→−∞x \to-\infty in this example.

Including image-charge-induced barrier lowering requires modifying the bulk equations as well as the boundary conditions. First, we modify the boundary condition for the electrostatic potential (Eq. (12)) at the metal-semiconductor contact (x∈Γ\boldsymbol{x} \in\Gamma) by adding the change Δψ≔Δϕ/qn\Delta\psi\coloneqq\Delta\phi/q_n of the potential barrier induced by the image charge. The new boundary condition reads

ψ(x,t)=ψ0+ψa(x,t)+Δψ(x,t),x∈Γ.(13)\psi(\boldsymbol{x},t) = \psi_0 + \psi_a(\boldsymbol{x},t) + \Delta\psi(\boldsymbol{x},t), \qquad\boldsymbol{x} \in\Gamma. \tag*{(13)}

Within the classical approximation [51], the change Δψ\Delta\psi of the potential barrier induced by the image charge is only defined for an upwards-bent conduction band edge (∇vψr<0\nabla_{\boldsymbol{v}}\psi_r < 0, x∈Γ\boldsymbol{x} \in\Gamma) and can be expressed as a function of the projected gradient of the residual electrostatic potential ∇vψr≔∇ψr⋅v\nabla_{\boldsymbol{v}}\psi_r \coloneqq\nabla\psi_r \cdot\boldsymbol{v}, (x∈Γ\boldsymbol{x} \in\Gamma),

Δψ={q∇vψr4πεi,∇vψr<0 (upwards bent Ec),0,∇vψr≥0 (constant and downwards bent Ec).(14)\Delta\psi= \begin{cases} \sqrt{\dfrac{q\nabla_{\boldsymbol{v}}\psi_r}{4\pi\varepsilon_i}}, & \nabla_{\boldsymbol{v}}\psi_r < 0\ \text{(upwards bent }E_c\text{)},\\ 0, & \nabla_{\boldsymbol{v}}\psi_r \ge0\ \text{(constant and downwards bent }E_c\text{)}. \end{cases} \tag*{(14)}

In this equation, we refer to εi≔Re⁡{ε∗(fi)}\varepsilon_i \coloneqq\operatorname{Re}\{\varepsilon^*(f_i)\} as the image-charge dielectric constant, defined as the real part Re⁡{ε∗(fi)}\operatorname{Re}\{\varepsilon^*(f_i)\} of the complex dielectric permittivity ε∗\varepsilon^* at the frequency fi=1/Tif_i = 1/T_i, which is connected to the time TiT_i an electron needs to travel from the barrier maximum to the metal-semiconductor interface (see Methods). The residual electrostatic potential ψr\psi_r is defined as the electrostatic potential obtained as a solution to Poisson’s equation

−∇⋅(ε∇ψr)=qCC+∑α∈{n,p,x}qαnα(ψ,φα),x∈Ω,(15)-\nabla\cdot(\varepsilon\nabla\psi_r) = q_C C + \sum_{\alpha\in\{n,p,x\}} q_\alpha n_\alpha(\psi,\varphi_\alpha), \qquad\boldsymbol{x} \in\Omega, \tag*{(15)}

which is now linear in ψr\psi_r because nαn_\alpha is a function of ψ\psi. At the contact boundaries, ψr\psi_r is obtained using the Schottky contact boundary condition

ψr(x,t)=ψ0(x)+ψa(x,t),x∈Γ.(16)\psi_r(\boldsymbol{x},t) = \psi_0(\boldsymbol{x}) + \psi_a(\boldsymbol{x},t), \qquad\boldsymbol{x} \in\Gamma. \tag*{(16)}

Further, the equations for the equilibrium charge carrier densities at the contacts (x∈Γ\boldsymbol{x} \in\Gamma) in Eq. (11) must be altered to

nn,0=NCFn(−ϕ0+ΔϕkBT),np,0=NVFp(−Eg−(ϕ0+Δϕ)kBT),(17)n_{n,0} = N_C\mathcal{F}_n\left(-\frac{\phi_0+\Delta\phi}{k_B T}\right), \qquad n_{p,0} = N_V\mathcal{F}_p\left(-\frac{E_g-(\phi_0+\Delta\phi)}{k_B T}\right), \tag*{(17)}

to consider the change of the barriers in equilibrium. If the effect of SBL is considered, we consequently solve a system of five coupled partial differential equations (Eq. (1), Eq. (7), and Eq. (15)) for five solution fields, namely the three quasi-Fermi potentials φn\varphi_n, φp\varphi_p, φx\varphi_x, and the two electrostatic potentials ψ\psi and ψr\psi_r. The equation system is supplemented with the equilibrium solution as initial conditions.

Application to Lateral Memristive Devices

We implemented all equations and numerical examples in the Julia programming language to extend the software package ChargeTransport.jl [50]. It was developed to simulate charge transport in semiconductors and is based on a finite volume method for spatial discretization [52, 53]. In the following, the equations from Section 2 are applied to model representative examples of lateral TMDC-based memristive devices that allow us to gain general insights into the hysteresis mechanism.

Model Geometry and Boundary Conditions

An illustration of the geometry of a typical lateral memristive device is shown in Figure 3a, along with its cross section in the x-y plane in Figure 3b. The devices comprise a TMDC flake on a SiO2_2/Si substrate, and the flake is laterally sandwiched by two Ti/Au electrodes. Typical TMDC flakes in lateral memristive devices have a thickness DD of some nanometers and a width WW and length LL of the order of a few micrometers (e.g., [29, 32, 39, 54, 55, 56]). We further simplify the model geometry and represent it by a one-dimensional (1D) conducting channel (Figure 1c) defined by the flake length LL between the electrodes and the electrode contact area A=DWA = DW. Hence, the 1D model comprises only the semiconductor (TMDC) domain of length LL with two metal-semiconductor contacts at the locations x1x_1 and x2x_2, where we apply the corresponding boundary conditions. For the applied potential, we assume ψa(x1)=0\psi_a(x_1) = 0 V at the left contact and $\psi_a(x_2) = \psi_a(x_1) + U(t)U(t) at the right contact. The voltage U(t)U(t) is a piecewise linear interpolation with four support points within the period of length tct_c and the extrema Umax⁡U_{\max} and −Umax⁡-U_{\max} (Figure 1d). The source term in Eq. (1) is set to rα=0r_{\alpha}=0 for all simulations. This is justified by the large electron and hole mobilities (section below) and the small voltage sweep rates of the order of some seconds per volt [36].

Illustration of the modeled memristive device geometry and applied voltage

Figure 3. Illustration of (a) the geometry of the modeled device in three dimensions with indicated SiO2_2/Si substrate, TMDC flake, and Ti/Au electrodes. (b) 2D view of the x-z cut plane of the memristive device, and (c) simplification of the geometry for the one-dimensional (1D) model with indicated channel length LL, and locations x1x_1 and x2x_2 of the two contact boundaries Γ\Gamma. (d) Illustration of the piecewise linear voltage U(t)U(t) as a function of the time tt over a period of length tct_c, and with a maximum voltage of Umax⁡U_{\max} and a minimum voltage Umin⁡=−Umax⁡U_{\min}=-U_{\max}.

Material Parameters

To gain general insights into the hysteresis mechanism and thoroughly validate the model, we use MoS2_2 as a well-known example material for the TMDC domain. Auger-electron spectroscopy measurements revealed a large density of sulfur vacancies sufficiently mobile to drift under an applied electric field in this material [36]. The vacancies can be introduced during the sample fabrication with a plasma treatment step [36]. Because of the formation energy of sulfur vacancies of up to several eV depending on the conditions [43, 44, 57], we assume that the number of sulfur vacancies remains constant during the operation of the device. While the role of sulfur vacancies for doping and their stable charge states have been debated [43, 58, 59], for large concentrations, they are expected to behave as n-type dopants [58]. Therefore, we assume that the mobile ionic species x represents donor-type sulfur vacancies, single-positively charged with zx=+1z_x=+1 in their ionized state. The maximum density of vacancies NxN_x is estimated from the density of sulfur sites to Nx≈4/V≈4⋅1028 m−3N_x\approx4/V\approx4\cdot10^{28}\,\mathrm{m}^{-3} (see Methods).

The electronic properties of MoS2_2 are well investigated and depend significantly on the number of monolayers and other extrinsic and intrinsic factors, such as the stress state and the temperature [37, 60, 61]. In the following, we review and estimate the range of material parameters expected for the most stable phase 2H-MoS2_2, a layer thickness of D=10−15D=10-15 nm at a temperature of T=300T=300 K.

With increasing thickness, the conduction band minimum (CBM) and the valence band maximum (VBM) shift from the K point (in the monolayer) to a midpoint between Γ\Gamma and K (CBM) and the Γ\Gamma point (VBM) (in the bulk) [62]. A transformation occurs from a direct semiconductor to an indirect one [54, 63, 64, 65], with a gradual shift of the electronic bandgap from Eg=1.8E_g=1.8 eV (monolayer) to Eg=1.2E_g=1.2 eV (bulk). This shift has been explained theoretically by an increasing contribution of interfacial stress to the overall atomic configuration as the material gets thinner [66] and is consistent with ab initio calculations for the influence of stress on the electronic properties of MoS2_2 monolayers [67, 68]. Assuming no external or internal strain of the MoS2_2 layer, e.g., from the fabrication process [69] or defect clusters [59], simulations predict values of Eg≈1.3E_g\approx1.3 eV for a layer thickness of 15 nm [66]. The changing band structure with layer thickness also alters the effective masses mn∗m_n^* and mp∗m_p^* of electrons and holes. For 2H-MoS2_2 monolayers, values at the K point of mn∗=(0.45−0.57)m0m_n^*=(0.45-0.57)m_0 and mp∗=(0.59−0.64)m0m_p^*=(0.59-0.64)m_0 are reported [62, 68, 70], with geometric means of mˉn∗≈0.54m0\bar{m}_n^*\approx0.54m_0 and mˉp∗≈0.61m0\bar{m}_p^*\approx0.61m_0 [70] considering the slight anisotropy (m0m_0: electron rest mass). This is slightly smaller than the effective masses of the bulk with values of mn∗≈0.55 m0m_n^*\approx0.55\,m_0 and mp∗≈0.71 m0m_p^*\approx0.71\,m_0 [62]. For a layer thickness of 15 nm, we expect effective masses close to the values in the bulk, and use them to estimate the effective density of states NcN_c and NvN_v of electrons and holes with the three-dimensional parabolic band approximation (ℏ\hbar: reduced Planck constant).

Nc=2(mk∗kBT2πℏ2)3/2,Nc=2(mp∗kBT2πℏ2)3/2.(18)N_{\mathrm{c}} = 2\left(\frac{m_{\mathrm{k}}^{*}k_{\mathrm{B}}T}{2\pi\hbar^{2}}\right)^{3/2},\qquad N_{\mathrm{c}} = 2\left(\frac{m_{\mathrm{p}}^{*}k_{\mathrm{B}}T}{2\pi\hbar^{2}}\right)^{3/2}. \tag*{(18)}

Experimental and theoretical studies on the in-plane quasi-static dielectric permittivity of monolayer MoS2_2 report a large scatter of values of εr≈3.7\varepsilon_{\mathrm{r}} \approx3.7 [71, 72, 73], and εr≈10−17\varepsilon_{\mathrm{r}} \approx10-17 [60]. Generally, εr\varepsilon_{\mathrm{r}} tends to increase as a function of the film thickness and values of εr≈10\varepsilon_{\mathrm{r}} \approx10 were measured for layer thicknesses of D=10−15D = 10-15 nm (close to εr≈12\varepsilon_{\mathrm{r}} \approx12 in the bulk) [74]. While these results are in excellent agreement with ab initio simulations [75], other calculations yield smaller values of εr=7.1\varepsilon_{\mathrm{r}} = 7.1 for the bulk [73]. For the image-charge dielectric constant, the frequency dependency of the complex electric permittivity becomes relevant, but our estimation shows εi≈εr\varepsilon_{\mathrm{i}} \approx\varepsilon_{\mathrm{r}} as a reasonable approximation (see Methods). Further, electron affinities of χe≈3.7\chi_{\mathrm{e}} \approx3.7 eV [76], χe≈4.0\chi_{\mathrm{e}} \approx4.0 eV [77, 78] and χe≈4.3\chi_{\mathrm{e}} \approx4.3 eV [79, 80] are reported for 2H-MoS2_2. A summary of the parameters and their estimated range is provided in Table 2 (see Methods).

Table 2: Summary of MoS2_2 material parameters collected from the literature for a layer thickness of ≈15\approx15 nm and comparison with the values used in the simulations. The electron effective mass at rest is denoted by m0m_0, and the effective masses mn∗m_n^*, mp∗m_p^* are approximated by their bulk values. Values for the fit parameters μn\mu_n, μp\mu_p and ϕ0\phi_0 depend on the respective dataset and are provided in Table 4 and for EaE_a in Table 1.

Parameter NameSymbolRangeValue usedUnit
Band gapEgE_g-1.31.3 [66]
Electron affinityχe\chi_e3.73.7-4.34.3 [76, 77, 78, 79, 80]4.04.0eV
Relative el. permittivityεr\varepsilon_r-1010 [74]1
Electron/hole effective massmn∗,mp∗m_n^*,m_p^*-0.55,0.710.55,0.71 [62]m0m_0
Image-charge dielectric constantεi\varepsilon_i≥10\ge10 [97, 98, 99, 100]10101
Maximum vacancy concentrationNxN_x≤4⋅1028\le4 \cdot10^{28}1⋅10281 \cdot10^{28}m−3^{-3}
Electron/hole mobilityμn,μp\mu_n,\mu_p0.1−2000.1-200 [37, 69, 81, 82, 83]-cm2^2/(Vs)
Schottky barriers (values for equilibrium)ϕ(x)\phi(x)0.050.05 [85], 0.30.3 [36]-eV
Activation energy of diffusionEaE_a0.60.6 [36], 0.8−2.50.8-2.5 [57]-eV

Table 2.

While the abovementioned parameters are obtained directly from the literature and used for all simulations, other parameters are expected to be highly sensitive to the sample's microstructure, defect composition, and interface quality. These sample-dependent model parameters include the mobilities μα\mu_{\alpha}, the intrinsic Schottky barriers ϕ0\phi_{0} and the defect energy level Ex,0E_{\mathrm{x},0} or the corresponding vacancy density nxn_{\mathrm{x}}. For the electron and hole mobilities, a large range of μn,p≈0.1−200\mu_{\mathrm{n,p}} \approx0.1-200 cm2^{2}/(Vs) [37, 69, 81, 82, 83] is reported. This range has been explained by the scattering of charge carriers with interfacial Coulomb impurities [81]. As a result, the mobilities tend to increase with the layer thickness and depend significantly on the gate dielectric and interface quality of the respective device [69, 81]. Ab-initio studies find activation energies of diffusion of Ea≈0.8−2.5E_{\mathrm{a}} \approx0.8-2.5 eV for sulfur vacancies in MoS2_2, depending strongly on the spatial configuration of the point defects [57]. This energy barrier is expected to be reduced along grain boundaries or interfaces [34], and a previous estimation based on Auger-electron spectroscopy measurements resulted in a much smaller value of Ea=0.6E_{\mathrm{a}} = 0.6 eV [36]. These measurements were performed on lateral memristive devices with vacancy densities of approximately nx=1026−1027n_{\mathrm{x}} = 10^{26}-10^{27} m−3^{-3} [36], assuming Nx=4⋅1028N_{\mathrm{x}} = 4 \cdot10^{28} m−3^{-3} (see Methods). With a work function of approximately 4.3 eV for Ti [84], slightly rectifying Schottky barriers around 0.3 eV are expected at the MoS2_2/Ti contact according to the Schottky-Mott rule [36]. However, experimental values are smaller, with ≈0.05\approx0.05 eV for Ti-MoS2_2 contacts in equilibrium [85].

Results

In the following, we analyze the origin of hysteresis, the influence of vacancy dynamics, and image-charge-induced Schottky barrier lowering in general for lateral memristive devices. We use parameters within the physically realistic range of MoS2_2 (§ 3) as representative material and discuss the cases of vanishing Schottky barriers and significant Schottky barrier heights. Finally, we compare additional simulations with various experimental data sets to demonstrate the model's validity.

For the first case, we consider a parameter set S1_1 (Table 1, Methods) with omittable Schottky barriers ϕ0(x1)=ϕ0(x2)=1\phi_0(x_1) = \phi_0(x_2) = 1 meV. The experimental data used to obtain the sample-dependent model parameters are taken from [36] and are shown in Figure 4a together with the simulation. The current magnitude ∣I∣\lvert I\rvert is a smooth function of the applied voltage UU without visible discontinuities and with approximately equal maximum current magnitudes at U=13U = 13 V and U=−13U = -13 V. While ∣I(U)∣\lvert I(U)\rvert is overall symmetric around U=0U = 0 V, the right hysteresis branch (U>0U > 0 V) has a slightly larger area compared to the left hysteresis branch (U<0U < 0 V). As indicated, the direction of the hysteresis is clockwise and counterclockwise in the right and left hysteresis branch, respectively. The simulations of eight consecutive voltage cycles are shown in Figure 4b, revealing a significant difference in the I-V characteristics of the first cycle compared to the following cycles. In particular, the I-V curve of the first cycle is highly asymmetric, with an initial rise in the current (Figure 4b, 1), which quickly saturates and almost remains constant until the maximum voltage is reached. In the left hysteresis branch (Figure 4b, U<0U < 0 V) a crossing of the current is visible, resulting in a change of the hysteresis direction from clockwise (Figure 4b, 3-6) to counterclockwise (Figure 4b, 4-5). The other voltage cycles (Figure 4b, Cycle 2-8) show a different but highly reproducible I-V behavior and reflect the characteristics of the measured I-V curve mentioned above very well, including the hysteresis direction and the transitions from the low-resistive state (LRS) to the high-resistive state (HRS) and vice versa.

Measurement versus simulation, simulations, simulated band diagrams, and vacancy-density configurations

Figure 4. Simulations using the parameter set S1_1 (Table 4, Methods). (a) Comparison of measurements with simulations of the second I-V cycle. (b) All eight consecutively simulated current-voltage cycles. (c) Simulated band diagram of the initial equilibrium configuration at t=0t = 0 s, and an applied voltage U=0U = 0 V, and (d) of the nonequilibrium configuration at the beginning of the second cycle at t=10.4t = 10.4 s (U=0U = 0 V). (e-j) Example configurations of the vacancy density nx(x)n_{\mathrm{x}}(x) as a function of space xx for various positions in the I-V curve shown in (a), together with the quasi-static equilibrium configuration of nxn_{\mathrm{x}} (gray), i.e., the configuration that would be reached at a constant applied voltage for t→∞t \rightarrow\infty. The measured data are taken from [36].

Parameter NameExpressionValues (S1_1)Values (S2_2)Unit
Mean maximum electric fieldEmax⁡E_{\max}≈1⋅10−8\approx1 \cdot10^{-8}≈3⋅10−7\approx3 \cdot10^{-7}V/m
Vacancy mobilityμx\mu_{\mathrm{x}}5⋅10−145 \cdot10^{-14}1.15⋅10−131.15 \cdot10^{-13}m2^{2}/ (Vs)
Activation energy of diffusionEaE_{\mathrm{a}}0.53±0.010.53 \pm0.010.503±0.0010.503 \pm0.001eV

Table 1. Overview of results for the activation energy of diffusion and the parameters extracted from the simulations to calculate the activation energy of diffusion from the vacancy mobilities obtained from the fits in Figure 4a and Figure 6a.

The origin of the two I-V characteristics is the different initial configurations of the system at the beginning of the respective cycle. At the beginning of the first cycle (t=0t = 0 s, U=0U = 0 V), the modeled system is in equilibrium, i.e., all quasi-Fermi levels are constant qnφn=qnφp=qxφx=0q_n\varphi_n = q_n\varphi_p = q_x\varphi_x = 0 eV, as shown in the reduced band diagram (Figure 4c). During the first cycle and at the beginning of the second cycle (t=10.4t = 10.4 s, U=0U = 0 V) the electrons and holes remain close to their quasi-static equilibrium configuration, i.e., the configuration that would be reached at a constant applied voltage for t→∞t \to\infty. The small mobility of vacancies on the other hand results in a substantial deviation from equilibrium (Figure 4d). This deviation from equilibrium is apparent from the notable drop in the vacancy quasi-Fermi level qxφxq_x\varphi_x at the left metal-semiconductor contact around x=0x = 0. Such a nonequilibrium configuration is reproducibly reached after each cycle following the first one. This phenomenon appears to be superficially similar to electroforming in filamentary devices. There, the formation of a filament is induced during the initial voltage cycles and then modulated in the subsequent cycles [86]. However, in our model, vacancies drift homogeneously distributed over the cross section of the conducting channel, and we will show in the following that the difference between the first and subsequent hysteresis loops is just a special case of a more general dynamic phenomenon. In the following, we focus the discussion on the second cycle as a more reproducible and representative device characteristic.

In Figure 4e-Figure 4j, the vacancy density is shown at various voltages in the second hysteresis cycle. Additionally, its quasi-static equilibrium configuration at the respective voltage is shown (gray, faint). At the beginning of the cycle, the drop in qxφxq_x\varphi_x (Figure 4d) results in a local depletion of vacancies at the left electrode (nx≪1010 m−3n_x \ll10^{10}\ \mathrm{m}^{-3}, Figure 4e). This depletion zone initially limits the current, but it vanishes quickly as the voltage increases (Figure 4e-Figure 4f). Therefore, the device is in a high-conductance (LRS) state at the beginning of the second cycle. As the voltage is further ramped up, the current increases continuously until a depletion zone forms and grows at the right electrode (Figure 4g). It limits the current just before the maximum voltage is reached. The depletion zone continues growing at the right electrode and reduces the current as the voltage is ramped down (Figure 4g-Figure 4i). Hence, the device is in the HRS. At U=0U = 0 V, the system strives to restore equilibrium where no depletion zone is present. Because of the small vacancy mobility, the depletion zone is not entirely annihilated at the beginning of the left hysteresis branch (Figure 4j, U<0U < 0 V) leading to a vacancy density configuration similar to the one at the beginning of the right cycle. Therefore, the I-V curve is overall very symmetric around U=0U = 0 V and the device starts in a high conductance state (LRS) at the beginning of the left hysteresis branch.

To analyze the dependency of the I-V curve on the vacancy mobility in detail, we performed simulations using the same parameter set S1 (Table 4, Methods) as before but with different values for the vacancy mobility, ranging from μx=1⋅10−16\mu_x = 1 \cdot10^{-16} m2^2/(V s) to μx=1⋅10−1\mu_x = 1 \cdot10^{-1} m2^2/(V s) for two different voltage sweep rates f1=5f_1 = 5 V/s and f2=2.5f_2 = 2.5 V/s. We define the hysteresis areas AHA_\mathrm{H} of the left and right hysteresis branches as the integral over the respective branch and consider the sign of the integral to distinguish the direction of the hysteresis. The hysteresis areas of the second cycle are plotted in Figure 5h as functions of μx\mu_x for f1f_1 and f2f_2. Example values of AHA_\mathrm{H} calculated from the hysteresis curves in Figure 5a-Figure 5f are indicated. As expected, AH=0A_\mathrm{H} = 0 for μx→0\mu_x \to0 and μx→∞\mu_x \to\infty because the vacancy density either remains stationary or in quasi-static equilibrium. A significant hysteresis occurs only in the regime between approximately μx=10−15\mu_x = 10^{-15} m2^2/(V s) and μx=10−12\mu_x = 10^{-12} m2^2/(V s).

Simulation results illustrating the dependency of the I-V characteristics on vacancy mobility and voltage sweep rate

Figure 5. Simulation results illustrating the dependency of the I-V characteristics on the vacancy mobility μx\mu_x and the voltage sweep rate ff using the parameter set S1 (Table 4, Methods) (a-f) Example hysteresis loops of the second cycle for various vacancy mobility values simulated with the voltage sweep rate f1=5f_1 = 5 V/s. (g) Negative cross-over voltage −Uc-U_c normalized to the maximum voltage Umax⁡=13U_{\max} = 13 V for the first twenty voltage cycles as functions of μx\mu_x. The values from the second-cycle hysteresis curves in (a-e) are indicated with dots. (d) Hysteresis areas AHA_\mathrm{H} defined as the integrals over the right (bold) and left (dashed) hysteresis branch of the second sweep cycle for two different voltage sweep rates f1=5f_1 = 5 V/s and f2=2.5f_2 = 2.5 V/s. The values from the hysteresis curves in (a-f) are indicated with dots.

Table 4: Sample-specific parameter sets S1_1 and S2_2 obtained from the fit of the simulation to the experimental data in Figure 4a (S1_1) and Figure 6a (S2_2). For all fits a sweep rate of f=5f = 5 V/s was used.

Parameter NameSymbolS1_1S2_2Units
Left Schottky barrierϕ0(x1)\phi_0(x_1)1⋅10−31 \cdot10^{-3}0.1440.144eV
Right Schottky barrierϕ0(x2)\phi_0(x_2)0.1100.110
Electron/hole mobilityμn,μp\mu_n,\mu_p2.5⋅10−42.5 \cdot10^{-4}2.15⋅10−32.15 \cdot10^{-3}m2^2/(Vs)
Vacancy mobilityμx\mu_x5⋅10−145 \cdot10^{-14}1.15⋅10−131.15 \cdot10^{-13}m2^2/(Vs)
Intrinsic defect energyEx,0E_{x,0}−4.32-4.32−4.33-4.33eV
Voltage amplitudeUmax⁡U_{\max}13131010V

Table 4.

As the mobility exceeds small values μx≤1⋅10−15\mu_x \leq1 \cdot10^{-15} m2^2/(V s), the hysteresis starts to open (AH≈0A_\mathrm{H} \approx0) and is initially clockwise oriented in both branches (Figure 5a-Figure 5b). Because the depletion zone is still present when U<0U < 0 V is reached the device is in the low conducting state at the beginning of the left hysteresis branch. A transition of the hysteresis direction occurs in the left hysteresis branch from clockwise (AH<0A_\mathrm{H} < 0) to counterclockwise (AH>0A_\mathrm{H} > 0) at intermediate mobilities around μx≈3⋅10−14\mu_x \approx3 \cdot10^{-14} m2^2/(V s) (Figure 5c). During the transition, the current in the left hysteresis branch intersects at a voltage UcU_c (Figure 5c, indicated), which continuously moves towards U=0U = 0 V as the mobility increases until the entire hysteresis branch is directed counterclockwise (Figure 5d). Origin of this transition is the shorter time required for vacancies with a larger mobility to reach quasi-static equilibrium, which causes a faster annihilation of the depletion zone before U<0U < 0 V is reached. Further increasing the vacancy mobility to large values (Figure 5d-Figure 5f) results in the formation of distinct maxima at voltages U<Umax⁡U < U_{\max} (and U>Umin⁡U > U_{\min}), an increasingly symmetric I-V curve, and an overall reduced maximum current. This behavior reflects the formation dynamics of the depletion zone. In the high-mobility regime, the depletion zone can form before the maximum positive or negative voltage is reached, which results in a drop in the current, leading to the formation of the maxima. Simultaneously, the vacancy density is close to its quasi-static equilibrium configuration, which results in the pronounced symmetry around U=0U = 0 V and Uc≈0U_c \approx0 V. The dependency of −Uc/Umax⁡-U_c/U_{\max} as a function of the vacancy mobility is detailed in Figure 5g for the first 20 voltage cycles as a measure for the symmetry of the I-V curve. Over the entire range of μx\mu_x, the voltage UcU_c always reaches Uc=0U_c = 0 after a sufficiently large number of voltage cycles. While >20 cycles are required to reach the symmetric I-V curve for μx<2⋅10−14 m2/(Vs)\mu_x < 2 \cdot10^{-14}\ \mathrm{m^2/(Vs)}, one to two cycles are enough for values close to μx=1⋅10−13 m2/(Vs)\mu_x = 1 \cdot10^{-13}\ \mathrm{m^2/(Vs)}.

The results show that the hysteresis and its symmetry are governed by the formation and annihilation dynamics of the vacancy depletion zone, i.e., the deviation of the vacancy density from its quasi-static equilibrium configuration. Hysteresis occurs in a small vacancy mobility range where the vacancies are sufficiently mobile to follow the potential gradient but slow enough to be in nonequilibrium. Therefore, a change in the voltage sweep rate only shifts the AH(μx)A_{\mathrm{H}}(\mu_x) curve on the vacancy mobility axis (Figure 5h, f2f_2). As another consequence, the asymmetry and intersection in the left hysteresis branch are volatile features. They depend on vacancy mobility and the cycle number and eventually vanish after enough sweep cycles. In this first I-V curve example, the intrinsic Schottky barriers are so small that SBL is negligible over the entire voltage range. The influence of Schottky barriers and SBL on the I-V curve is explored next in the second example.

For the second case, we consider a similar parameter set S2_2 (Table 4, Methods) as before but introduce intrinsic Schottky barriers ϕ0(x1)=0.144\phi_0(x_1) = 0.144 eV and ϕ0(x2)=0.11\phi_0(x_2) = 0.11 eV with a slightly smaller value at the right contact. Figure 4a demonstrates the excellent match of the simulated and measured I-V curve. Both have the same clockwise direction (right branch) and counterclockwise direction (left branch) as in Figure 4a, but the asymmetry around U=0U = 0 V is more pronounced. In contrast to the previous I-V curve (Figure 4a), the two maximum current magnitudes are notably different, and the right hysteresis area is smaller than the left one. Because the intrinsic Schottky barriers are small but significant, SBL can occur. In equilibrium at t=0t = 0 (Figure 6b), the SBL notably reduces the contact barriers by approximately 25 % and 18 % to values of ϕ(x1)≈0.11\phi(x_1) \approx0.11 eV and ϕ(x2)≈0.09\phi(x_2) \approx0.09 eV. This difference in the reduction of the Schottky barriers is caused by the difference in the intrinsic Schottky barrier ϕ0(x1)\phi_0(x_1) and ϕ0(x2)\phi_0(x_2) at the contacts. At a large intrinsic Schottky barrier, the magnitude of the residual potential gradient ∣∇vψr∣\lvert\nabla_{\mathbf{v}}\psi_{\mathrm{r}}\rvert tends to be larger than at a small barrier. This gradient determines the reduction of the Schottky barrier via Eq. (14), and therefore, SBL tends to reduce the difference in the Schottky barriers compared to the intrinsic barriers.

Comparison of simulations with measurements at the example of an asymmetric hysteresis curve

Figure 6. Comparison of simulations with measurements at the example of an asymmetric hysteresis curve using the parameter set S2_2 (Table 4, Methods). (a) Simulation of the second voltage cycle and measurements. (b) Conduction band edges, and quasi-Fermi levels in equilibrium at t=0t = 0 s with and without SBL. (c) Comparison of simulations of the second voltage cycle with and without SBL and without SBL but with intrinsic Schottky barriers equal to the equilibrium Schottky barriers from the case with SBL (No SBL, reduced ϕ0\phi_0). (d) Magnitude of the current, applied voltage U(t)U(t), and change of the Schottky barrier Δϕ(t)\Delta\phi(t) as functions of the time tt. The measured data are taken from [36].

This affects the band edges EcE_{\mathrm{c}}, EvE_{\mathrm{v}}, and ExE_{\mathrm{x}} in proximity of the contact interfaces, correspondingly (Figure 6b, EvE_{\mathrm{v}} and ExE_{\mathrm{x}} not shown). In Figure 6d, the applied voltage UU, the current magnitude ∣I∣\lvert I\rvert, and the change Δϕ(x1)\Delta\phi(x_1) and Δϕ(x2)\Delta\phi(x_2) of the Schottky barriers are plotted as functions of the time tt over the first two voltage cycles. As the voltage increases (U>0U > 0 V) the band edges at the left contact remain bent upwards (∇vψr<0\nabla_{\mathrm{v}}\psi_{\mathrm{r}} < 0), which results in an increased Schottky barrier Δϕ(x1)>0\Delta\phi(x_1) > 0 (Eq. (14)). At the right contact, the band edges move to lower energies, and the projected electrostatic potential quickly reaches positive values (∇vψr≥0\nabla_{\mathrm{v}}\psi_{\mathrm{r}} \ge0) indicating a downwards bent conduction band. As a result, the change in the right Schottky barrier remains zero, i.e., Δϕ(x2)=0\Delta\phi(x_2) = 0, over most of the right hysteresis cycle (U>0U > 0 V) and does not limit the current. As the voltage approaches negative values, the situation reverses and the change in Schottky barriers is Δϕ(x1)=0\Delta\phi(x_1) = 0 and Δϕ(x2)>0\Delta\phi(x_2) > 0 over most of the left hysteresis cycle (U<0U < 0 V). Note that both Δϕ(t)\Delta\phi(t) are asymmetric around the time at which the respective maximum positive and negative voltage is applied. This asymmetry is also present in the current magnitude and reflects the hysteresis in the time domain.

To identify the influence of SBL on the I-V characteristics, we performed a simulation with the same parameters as for the curve in Figure 6a but without SBL. The two I-V curves with and without SBL are compared in Figure 6c, and qualitative differences are apparent. Without SBL, the hysteresis direction of the right branch is reversed, and a pronounced asymmetry is visible around U=0U = 0 V, i.e., largely different maximum currents and hysteresis areas of the left and right branches. This asymmetry is caused by a larger difference in the equilibrium Schottky barriers if SBL is not considered. The total hysteresis area without SBL is similar to the case with SBL. An additional I-V simulation was performed without SBL using the equilibrium Schottky barriers ϕ(x)\phi(x) obtained from the simulation with SBL as intrinsic Schottky barriers ϕ0\phi_0 (Figure 6c, No SBL, reduced ϕ0\phi_0). The resulting I-V curve matches the one with SBL. This comparison shows that the hysteresis is not dominated by a change in the interfacial current but a limitation in the bulk conductance via the formation of the depletion zone. Consequently, SBL greatly affects the I-V curve, but it does not contribute significantly to the hysteresis area in this example.

The effect of a change in the Schottky barriers is illustrated with additional simulations in Figure 7, including SBL. We used the parameter set S2_2 (Table 4, Methods) as before, but various intrinsic Schottky barriers for the right contact (Figure 7a) and the left contact (Figure 7b) with values from 0.12 eV to 0.18 eV while keeping the respective other Schottky barrier fixed to a value of 0.11 eV. In both cases, we observe a transition from a relatively symmetric I-V curve for the case of similar Schottky barriers to a notably asymmetric curve for significantly different Schottky barriers. This transition is accompanied by a change of the hysteresis direction in the left hysteresis branch if ϕ0(x1)<ϕ0(x2)\phi_0(x_1) < \phi_0(x_2) and in the right hysteresis branch if ϕ0(x2)<ϕ0(x1)\phi_0(x_2) < \phi_0(x_1). A comparison of the I-V curve in Figure 7b (right, ϕ0(x2)=0.18\phi_0(x_2) = 0.18 eV) with the hysteresis curve in Figure 6c demonstrates that the behavior observed without SBL is mainly caused by the difference in the two Schottky barriers of the left and right contact. Other features, such as the formation of a maximum at U<Umax⁡U < U_{\max} and the reduction of the maximum current magnitude with increasing barrier difference are also observed (Figure 7). These features are superficially similar to those caused by the increasing vacancy mobility of the device with symmetric contacts, but their origin differs. The asymmetric Schottky barriers result in different restoring forces compared to the symmetric configuration, which alters the formation and annihilation dynamics of the depletion zone.

Simulations of the second cycle using different Schottky barriers and vacancy mobilities

Figure 7. Simulations of the second cycle using parameter set S2_2 (Table 4, Methods) changing only the intrinsic Schottky barriers. (a) Current-voltage (I-V) characteristics for three different values of the left barrier ϕ0(x1)\phi_0(x_1) and a fixed value of ϕ0(x2)=0.11\phi_0(x_2) = 0.11 eV for the right barrier. (b) I-V characteristics for three different values of the right barrier ϕ0(x2)\phi_0(x_2) and a fixed value of the left barrier ϕ0(x1)=0.11\phi_0(x_1) = 0.11 eV. (c) I-V characteristics of the first 10 voltage cycles and the 100th cycle, with indicated Schottky barriers and vacancy mobility. (d) Total hysteresis area as a function of the cycle number normalized to its respective value at cycle 100 and for three different mobilities. The parameter set and the Schottky barriers are identical to those used for the calculations in (c).

To analyze the I-V dynamics caused by asymmetric Schottky contacts, we simulate the I-V curve of an example configuration over 100 consecutive voltage cycles (Figure 7c). An asymmetry is apparent in cycle one, which reduces slightly until cycle ten is reached. All following voltage cycles are approximately identical to cycle ten and retain a significant asymmetry demonstrated in the example of cycle 100. As a quantitative measure, we extract the total hysteresis area AH,nA_{\mathrm{H,n}} normalized to its value at cycle 100 and plot it as a function of the number of cycles for simulations with three different vacancy mobilities (Figure 7d). All three functions reach AH,n=1A_{\mathrm{H,n}} = 1 and remain constant after enough cycles. A larger vacancy mobility reduces the number of cycles required to reach the dynamically stable configuration. Consequently, the asymmetry in the I-V characteristics of different Schottky barriers is not a mere volatile feature, as observed in the case of symmetric contacts. Instead, it remains present after a sufficiently large number of cycles reflecting the asymmetric restoring force in equilibrium.

For a more detailed validation, we fitted the model to additional sets of measured I-V curves with various maximum voltages and different extents of asymmetry. The results in Figure 8 show an overall excellent agreement with the measured data. Only minor deviations are apparent, e.g., in the right hysteresis branch of the asymmetric curves in Figure 8a and Figure 8c, with a slightly larger current in the simulation compared to the measurement. All parameters are within the expected range estimated in § 3 (parameter sets S3_3-S6_6, Methods).

Comparison of simulations with measured data of I-V curves of four devices

Figure 8. Comparison of simulations with measured data of I-V curves of four devices (a-d) with various maximum voltages and extends of asymmetry. The measured data are taken from [36], and the simulation parameters are provided in the Methods.

Discussion and Conclusion

We presented a semi-classical charge transport model, which self-consistently couples the drift-diffusion equations for electrons, holes, and charged mobile point defects, to Poisson’s equation for the electrostatic potential. We developed and included a formulation for image-charge-induced Schottky barrier lowering (SBL) and solved the system of equations via a finite-volume discretization to analyze the resistive switching mechanism in two-dimensional memristive devices.

The simulations indicate that the hysteresis in all types of I-V curves with and without significant Schottky barriers is caused by the dynamics of mobile charged vacancies, namely the formation and annihilation of a vacancy depletion region, which reduces the conductivity locally and eventually limits the conductivity of the device. The formation and annihilation dynamics of the depletion region are governed by the operating frequency, vacancy mobility, and asymmetry of the two contact potentials. By changing the time and voltage required for the formation of the depletion zone, these parameters crucially determine the hysteresis direction and symmetry of the I-V curve around U=0U = 0 V. This switching process explains features of the I-V curve also observed experimentally [34], such as the formation of a maximum at U<Umax⁡U < U_{\max} and the reduction of the maximum current magnitude with increasing mobility. For symmetric contacts, the asymmetry of the I-V curve is volatile and reduces with increasing cycle number before it remains constant.

Such an initial cycle-to-cycle instability can be avoided by operating the device in the high-vacancy-mobility range by choosing a corresponding operating frequency (Figure 5f). Large vacancy mobilities reduce the retention time, which is typically desired for transient memristive devices [87, 88, 89, 90] but for most applications of memristive devices as nonvolatile memories, it is undesired [91, 92]. Optimizing the cycle-to-cycle stability and retention time implies selecting the narrow range of the sweep rate where the vacancies are sufficiently mobile to occupy their dynamically stable configuration during the first cycle but are still delayed causing a large hysteresis. This is the case at the local maximum in the hysteresis area plot (Figure 5h), close to the point where the two hysteresis area curves split up.

A permanent asymmetry in the I-V curve can be obtained by introducing Schottky barriers. We show how minuscule changes in the Schottky barriers (<0.04< 0.04 eV) can qualitatively change the I-V characteristic, i.e., reverse the direction and change the hysteresis area and symmetry. Hence, the simulations quantify how deliberate changes can be introduced in the I-V curve for optimizing the device characteristics. Precisely controlling the interface potentials is paramount for fabricating reliable and reproducible devices. Further, we confirmed that SBL represents a crucial contribution to the overall Schottky barrier heights and the slope of the I-V characteristics. SBL reduces the differences between the intrinsic barriers and contributes significantly to their small values in equilibrium. However, while SBL significantly affects the I-V curves, no system configuration was found where SBL determines the hysteresis area. This is because SBL is essentially determined by the projected gradient of the electrostatic potential at the contact interface, and this gradient is by far dominated by the applied voltage and not by the change of the vacancy density at the contacts. While the different hysteresis directions have been previously interpreted as caused by different switching mechanisms [36], our simulations clarify that the exact bulk-limiting mechanism can explain both curves and results in qualitatively different I-V characteristics depending on minor changes in the parameters. While the simulations were performed with MoS2\mathrm{MoS}_2 as a representative TMDC material, the model is not limited to the drift and diffusion of positively charged sulfur vacancies in MoS2\mathrm{MoS}_2, but it can be applied to other materials and kinds of mobile ionic defects with different charge states.

The excellent match of the simulations with various experimental I-V curves of lateral MoS2\mathrm{MoS}_2-based devices supports the validity of our results. Also, the model parameters obtained from the fit match very well within the expected range (§ 3). Equilibrium vacancy densities in the simulations are around nx≈1026 m−3n_x \approx10^{26}\ \mathrm{m}^{-3} in good agreement with Auger-electron spectroscopy measurements on the modeled devices [36]. For the electron and hole mobilities, we obtain values of μn,p≈1−20 cm2/(Vs)\mu_{\mathrm{n,p}} \approx1 - 20\ \mathrm{cm}^2/(\mathrm{V}s), which is well in the expected range. reported range of μn,p≈0.1−200 cm2/(Vs)\mu_{\mathrm{n,p}} \approx0.1 - 200\ \mathrm{cm}^2/(\mathrm{Vs}) [37, 69, 81, 82, 83]. A smaller range of values is obtained for the vacancy mobilities of μx≈1⋅10−14−1⋅10−13 m2/(Vs)\mu_{\mathrm{x}} \approx1 \cdot10^{-14} - 1 \cdot10^{-13}\ \mathrm{m}^2/(\mathrm{Vs}) in the simulations. By using an analytical model [93], these values translate to activation energies of diffusion of Ea≈0.50−0.53 eVE_{\mathrm{a}} \approx0.50 - 0.53\ \mathrm{eV} (Methods), which matches particularly well with previous estimations for the considered devices of Ea=0.6 eVE_{\mathrm{a}} = 0.6\ \mathrm{eV} [36]. Ab-initio studies result in larger activation energies of Ea≈0.8−2.5 eVE_{\mathrm{a}} \approx0.8 - 2.5\ \mathrm{eV}, depending on the microscopic environment, such as the configuration of vacancy clusters and lines [57]. However, the activation energy of diffusion is expected to be further reduced along grain boundaries or interfaces [34]. Hence, our results could support the relevance of such defects for the device performance, consistent with experimental work on MoS2_2-based lateral devices [34].

The equilibrium Schottky barriers obtained from the simulations vary between approximately 0 eV and 0.17 eV, close to experimental values of ≈0.05 eV\approx0.05\ \mathrm{eV} [85] for Ti-MoS2_2 contacts and 0.3 eV from the Schottky-Mott rule [36]. The device-to-device variations of the Schottky barriers, mobilities, and vacancy densities can be caused by microstructure and interface quality variations, reflecting the fabrication tolerances [94, 95]. Such variations can also occur in the geometry of the individual devices and explain minor deviations between measurements and simulations. Additionally, the charging and discharging dynamics of trap-states could contribute to the I-V characteristics depending on the defect and interface structure of the individual sample [33, 96]. Also, slight deviations from the equilibrium configuration of the system prior to the measurements can introduce an asymmetry in the I-V curves. The overall excellent agreement of the fitted values with experimental data shows that our model can be used beyond a qualitative explanation for the quantitative characterization of the device parameters.

In conclusion, our results strongly corroborate the experimental indications for the relevance of sulfur vacancies in TMDC devices theoretically, and we identified the switching mechanism. Deep insights into the physics of the switching mechanism and the origin of hysteresis are gained as a vital step towards tailoring the device functionality for applications. The results bear general implications for lateral memristive systems and show that mobile ionic defects could dominate their hysteresis. The model and simulation software are made available as powerful tools for the future analysis and design of other semiconductor devices.

Methods

Image-charge dielectric constant. For estimating the image-charge dielectric constant εi\varepsilon_{\mathrm{i}}, the frequency dependency of the in-plane electric permittivity must be known. Measurements on CVD grown monolayers [97] and calculations [98, 99, 100] show a strong dispersive behavior at energies of approximately E>2E > 2 eV [101] and wavelengths of λ<900\lambda< 900 nm with real parts of Re⁡{εr∗}=8−20\operatorname{Re}\{\varepsilon_{\mathrm{r}}^{*}\} = 8 - 20 (at λ=900\lambda= 900 nm) depending on the study [97, 98, 99, 100] but generally larger than the corresponding quasi-static values. Further, ab initio calculations of the complex permittivity demonstrate that the resonance peaks shift to even higher EE and smaller λ\lambda with increasing number of MoS2_2 layers until the bulk properties are reached [101]. Therefore, we assume an approximately constant electric permittivity εr∗\varepsilon_{\mathrm{r}}^{*} for energies E≪2E \ll2 eV, corresponding to photon frequencies of f≪E/h≈5⋅1014f \ll E/h \approx5 \cdot10^{14} Hz with values possibly slightly larger than the quasi-static value of εr≈10\varepsilon_{\mathrm{r}} \approx10 for a layer thickness D=10−15D = 10 - 15 nm. Based on the definition of the image-charge dielectric constant εi≔Re⁡{ε∗(fi)}\varepsilon_{\mathrm{i}} \coloneqq\operatorname{Re}\{\varepsilon^{*}(f_{\mathrm{i}})\} in § 2.3, we estimate the frequency fi=1/Ti=∣vd∣/Δx=μn∣E∣/Δxf_{\mathrm{i}} = 1/T_{\mathrm{i}} = |\mathbf{v}_{\mathrm{d}}|/\Delta x = \mu_{\mathrm{n}}|\mathbf{E}|/\Delta x (vd\mathbf{v}_{\mathrm{d}}: electron drift velocity) by assuming typical values for Δx=1−5\Delta x = 1 - 5 nm, an electron mobility μn=1−200\mu_{\mathrm{n}} = 1 - 200 cm2^{2}/Vs and an electric field magnitude ∣E∣=(0.1−1)107|\mathbf{E}| = (0.1 - 1)10^{7} A/m. The resulting frequencies fi=2⋅1010−2⋅1014f_{\mathrm{i}} = 2 \cdot10^{10} - 2 \cdot10^{14} Hz are at maximum still below the highly dispersive range of εr∗\varepsilon_{\mathrm{r}}^{*}, and therefore, we assume εi≈εr\varepsilon_{\mathrm{i}} \approx\varepsilon_{\mathrm{r}}.

Maximum defect density. We estimate NxN_{\mathrm{x}} considering hexagonal 2H-MoS2_2 with an in-plane lattice constant a=0.316a = 0.316 nm and c=1.229c = 1.229 nm [102], which results in a unit cell volume of V=a2csin⁡(60∘)=0.1063V = a^{2}c\sin(60^\circ) = 0.1063 nm3≈10−28^{3} \approx10^{-28} m3^{3}. Each unit cell comprises two formula units of MoS2_2, i.e., two Mo atoms and four S atoms, and is part of two stacked monolayers with a respective distance of ≈c/2\approx c/2. With VV and the number of S-sites, we estimate a maximum sulfur vacancy density Nx≈4/V≈4⋅1028N_{\mathrm{x}} \approx4/V \approx4 \cdot10^{28} m−3^{-3}.

Activation energy of diffusion and vacancy mobility. The activation energy of diffusion EaE_{\mathrm{a}} corresponding to the fitted values of the mobility μx\mu_{\mathrm{x}} is calculated using the mobility model of Genreith-Schriever [93]. The hopping distance axa_{\mathrm{x}} is approximated by the in-plane lattice constant and set to ax=0.316a_{\mathrm{x}} = 0.316 nm [102], for the attempt frequency we use a value of ν0=1012\nu_{0} = 10^{12} Hz [103]. We omit the activation entropy of migration Sa=0S_{\mathrm{a}} = 0, and consider the minimum electric field magnitude, i.e., Emin⁡=0E_{\min} = 0, as well as the maximum electric field magnitude of each time step Emax⁡E_{\max} averaged over time. With this range of the electric field, we estimate the error in the activation energy that occurs from assuming a constant vacancy mobility μx\mu_{\mathrm{x}}. The side fraction of vacancies is in good approximation ni≔nx/Nx≈0n_{\mathrm{i}} \coloneqq n_{\mathrm{x}}/N_{\mathrm{x}} \approx0. The values for Emax⁡E_{\max} and the resulting EaE_{\mathrm{a}} are summarized in Table 1 for the two fits from Figure 4a (data set S1_1) and Figure 6a (data set S2_2).

Summary of model parameters. In the following, all model parameters and references are summarized. We use a background doping concentration of C=1021C = 10^{21} m−3^{-3} with zc=1z_{\mathrm{c}} = 1 for all simulations. This concentration value improves numerical stability and is sufficiently small to have no significant influence on the results. Further, a temperature of T=300T = 300 K is used. The mobile ionic species x is assumed to behave as n-type dopants [58] with zx=+1z_{\mathrm{x}} = +1 in their ionized state. All material parameters of MoS2_2 obtained from the literature are summarized in Table 2 together with the respective references. If references are provided for a range of values, the value used from within this range is denoted separately. The geometry parameters are listed in Table 3, and the sample-specific parameters, i.e., the parameter sets S1_1 and S2_2, in Table 4.

Table 3: Summary of geometry parameters used for the simulations of the device in [36]. The two values provided for the channel length LL are used together with the parameter sets S1_1-S6_6 (Table 4, Table 5).

Parameter NameExpressionValues usedUnit
Channel lengthLL11 (S1_1, S3_3-S5_5), 22 (S2_2, S6_6)
Channel widthWW1010μ\mum
Channel thicknessDD0.0150.015

Table 3.

Table 5: Parameters obtained from the fit of the model to the experimental data in Figure 8a-Figure 8d. All fits were performed with a sweep rate of f=5f = 5 V/s.

Parameter NameSymbolS3_3 (Fig. 8a)S4_4 (Fig. 8b)S5_5 (Fig. 8c)S6_6 (Fig. 8d)Units
Left Schottky barrierϕ0(x1)\phi_0(x_1)0.1670.1671⋅10−31 \cdot10^{-3}0.1520.1520.1550.155eV
Right Schottky barrierϕ0(x2)\phi_0(x_2)0.1480.1480.0010.0010.1150.115eV
Electron/hole mobilityμn,μp\mu_n,\mu_p1.3⋅10−31.3 \cdot10^{-3}3.6⋅10−43.6 \cdot10^{-4}1.5⋅10−31.5 \cdot10^{-3}6⋅10−46 \cdot10^{-4}m2^2/(Vs)
Vacancy mobilityμx\mu_x8.5⋅10−148.5 \cdot10^{-14}6⋅10−146 \cdot10^{-14}7⋅10−147 \cdot10^{-14}3.4⋅10−143.4 \cdot10^{-14}m2^2/(Vs)
Intrinsic defect energyEx,0E_{x,0}−4.3-4.3−4.32-4.32−4.32-4.32−4.28-4.28eV
Voltage amplitudeUmax⁡U_{\max}10101313101014.514.5V

Table 5.

Funding

The research is funded by the Carl-Zeiss Foundation via the Project Memwerk, the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)– Project-ID 434434223 – SFB 1461, and the Leibniz Competition 2020.

Competing interests

The authors declare that they have no competing interests.

Availability of data and materials

The datasets used and analyzed during the current study are available from the corresponding author upon reasonable request. The simulated data sets can be generated with the open-source software tool ChargeTransport.jl (https://github.com/PatricioFarrell/ChargeTransport.jl).

References

  1. [1]Bughin, J.; Seong, J.; Manyika, J.; Chui, M.; Joshi, R. Notes from the AI frontier: Modeling the impact of AI on the world economy. Available online: https://www.mckinsey.com/featured-insights/artificial-intelligence/notes-from-the-ai-frontier-modeling-the-impact-of-ai-on-the-world-economy (accessed on 31 March 2023).
  2. [2]Jones, N. How to stop data centres from gobbling up the world's electricity. Nature 2018, 561, 163–166, doi:10.1038/d41586-018-06610-y.DOI
  3. [3]Dhar, P. The carbon impact of artificial intelligence. Nature Machine Intelligence 2020.DOI
  4. [4]Lin, H.-Y. Colors of artificial intelligence. Computer 2021, 54, 95–99, doi:10.1109/MC.2021.3102359.DOI
  5. [5]Bailey, B. AI power consumption exploding. Available online: https://semiengineering.com/ai-power-consumption-exploding (accessed on 31 March 2023).
  6. [6]Ludvigsen, K.G.A. ChatGPT’s electricity consumption. Available online: https://towardsdatascience.com/chatgpts-electricity-consumption-7873483feac4 (accessed on 31 March 2023).
  7. [7]Furber, S. Large-scale neuromorphic computing systems. J. Neural Eng. 2016, 13, 51001, doi:10.1088/1741-2560/13/5/051001.DOI
  8. [8]Yu, S. Neuro-inspired computing with emerging nonvolatile memories. Proc. IEEE 2018, 106, 260–285, doi:10.1109/JPROC.2018.2790840.DOI
  9. [9]Yang, J.J.; Strukov, D.B.; Stewart, D.R. Memristive devices for computing. Nat. Nanotechnol. 2013, 8, 13–24, doi:10.1038/nnano.2012.240.DOI
  10. [10]Zidan, M.A.; Strachan, J.P.; Lu, W.D. The future of electronics based on memristive systems. Nat. Electron. 2018, 1, 22–29, doi:10.1038/s41928-017-0006-8.DOI
  11. [11]Dittmann, R.; Menzel, S.; Waser, R. Nanomionic memristive phenomena in metal oxides: the valence change mechanism. Advances in Physics 2021, 70, 155–349, doi:10.1080/00018732.2022.2084006.DOI
  12. [12]Huh, W.; Lee, D.; Lee, C.-H. Memristors based on 2D materials as an artificial synapse for neuromorphic electronics. Adv. Mater. 2020, 32, e2002092, doi:10.1002/adma.202002092.DOI
  13. [13]Ko, T.-J.; Li, H.; Mofid, S.A.; Yoo, C.; Okbogue, E.; Han, S.S.; Shawkat, M.S.; Krishnaprasad, A.; Islam, M.M.; Dev, D.; et al. Two-dimensional near-atom-thickness materials for emerging neuromorphic devices and applications. iScience 2020, 23, 101676, doi:10.1016/j.isci.2020.101676.DOI
  14. [14]Cheng, P.; Sun, K.; Hu, Y.-H. Memristive behavior and ideal memristor of 1T phase MoS₂ nanosheets. Nano Lett. 2016, 16, 572–576, doi:10.1021/acs.nanolett.5b04260.DOI
  15. [15]Ge, R.; Wu, X.; Kim, M.; Shi, J.; Sonde, S.; Tao, L.; Zhang, Y.; Lee, J.C.; Akinwande, D. Atomristor: nonvolatile resistance switching in atomic sheets of transition metal dichalcogenides. Nano Lett. 2018, 18, 434–441, doi:10.1021/acs.nanolett.7b04342.DOI
  16. [16]Xu, R.; Jang, H.; Lee, M.-H.; Amanov, D.; Cho, Y.; Kim, H.; Park, S.; Shin, H.-J.; Ham, D. Vertical MoS₂ double-layer memristor with electrochemical metallization as an atomic-scale synapse with switching thresholds approaching 100 mV. Nano Lett. 2019, 19, 2411–2417, doi:10.1021/acs.nanolett.8b05140.DOI
  17. [17]Yan, X.; Zhao, Q.; Chen, A.P.; Zhao, J.; Wang, J.; Wang, H.; Zhang, L.; Li, X.; Xiao, Z.; et al. Vacancy-induced synaptic behavior in 2D WS₂ nanosheet-based memristor for low-power neuromorphic computing. Small 2019, 15, e1901423, doi:10.1002/smll.201901423.DOI
  18. [18]Wan, F.; Wang, Q.; Harumoto, T.; Gao, T.; Ando, K.; Nakamura, Y.; Shi, J. Truly electroforming-free memristor based on TiO₂-CoO phase-separated oxides with extremely high uniformity and low power consumption. Adv. Funct. Mater. 2020, 30, 2007101, doi:10.1002/adfm.202007101.DOI
  19. [19]Khan, S.A.; Kim, S. Comparison of diverse resistive switching characteristics and demonstration of transitions among them in Al-incorporated HfO₂-based resistive switching memory for neuromorphic applications. RSC Adv. 2020, 10, 31342–31347, doi:10.1039/D0RA06389D.DOI
  20. [20]Sangwan, V.K.; Lee, H.-S.; Bergeron, H.; Balla, I.; Beck, M.E.; Chen, K.-S.; Hersam, M.C. Multi-terminal memtransistors from polycrystalline monolayer molybdenum disulfide. Nature 2018, 554, 500–504, doi:10.1038/nature25747.DOI
  21. [21]Lee, H.-S.; Sangwan, V.K.; Rojas, W.A.G.; Bergeron, H.; Jeong, H.Y.; Yuan, J.; Su, K.; Hersam, M.C. Dual-gated MoS₂ memtransistor crossbar array. Adv. Funct. Mater. 2020, 30, 2003683, doi:10.1002/adfm.202003683.DOI
  22. [22]Ding, G.; Yang, B.; Chen, R.-S.; Mo, W.-A.; Zhou, K.; Liu, Y.; Shang, G.; Zhai, Y.; Han, S.-T.; Zhou, Y. Reconfigurable 2D WSe₂-based memtransistor for mimicking homosynaptic and heterosynaptic plasticity. Small 2021, 17, e2103175, doi:10.1002/smll.202103175.DOI
  23. [23]Yan, X.; Qian, J.H.; Sangwan, V.K.; Hersam, M.C. Progress and challenges for memtransistors in neuromorphic circuits and systems. Adv. Mater. 2022, 34, e2108025, doi:10.1002/adma.202108025.DOI
  24. [24]Choi, M.S.; Lee, G.-H.; Yu, Y.-J.; Lee, D.-Y.; Lee, S.H.; Kim, P.; Hone, J.; Yoo, W.J. Controlled charge trapping by molybdenum disulphide and graphene in ultrathin heterostructured memory devices. Nat. Commun. 2013, 4, 1624, doi:10.1038/ncomms2652.DOI
  25. [25]Coleman, E.; Mirabelli, G.; Bolshakov, P.; Zhao, P.; Caruso, E.; Gity, F.; Monaghan, S.; Cherkaoui, K.; Balestra, V.; Wallace, R.M.; et al. Investigating interface states and oxide traps in the MoS₂/oxide/Si system. Solid-State Electron. 2021, 186, 108123, doi:10.1016/j.sse.2021.108123.
  26. [26]Durand, P.C.; Zhang, X.-G.; Fowlkes, J.; Najmaei, S.; Lou, J.; Li, A.-P. Defect-mediated transport and electronic irradiation effect in individual domains of CVD-grown monolayer MoS₂. J. Vac. Sci. Technol. B 2015, 33, doi:10.1116/1.4906331.DOI
  27. [27]Guo, Y.; Wei, X.; Shu, J.; Liu, B.; Yin, J.; Guan, C.; Han, Y.; Gao, S.; Chen, Q. Charge trapping at the MoS₂-SiO₂ interface and its effects on the characteristics of MoS₂ metal-oxide-semiconductor field effect transistors. Appl. Phys. Lett. 2015, 106, 103109, doi:10.1063/1.4914968.DOI
  28. [28]Illarionov, Y.Y.; Knobloch, T.; Wlattl, M.; Zrеpa, G.; Pospischil, A.; Polyushkin, D.K.; Furchi, M.M.; Mueller, T.; Grasser, T. Energetic mapping of oxide traps in MoS₂ field-effect transistors. 2D Mater. 2017, 4, 25108, doi:10.1088/2053-1583/aa734a.DOI
  29. [29]Li, T.; Du, G.; Zhang, B.; Zeng, Z. Scaling behavior of hysteresis in multilayer MoS₂ field effect transistors. Appl. Phys. Lett. 2014, 105, 93107, doi:10.1063/1.4894865.DOI
  30. [30]Shu, J.; Wu, G.; Guo, Y.; Liu, B.; Wei, X.; Chen, Q. The intrinsic origin of hysteresis in MoS₂ field effect transistors. Nanoscale 2016, 8, 3049–3056, doi:10.1039/c5nr07336g.DOI
  31. [31]van der Zande, A.M.; Huang, P.Y.; Chenet, D.A.; Berkelbach, T.C.; You, Y.; Lee, G.-H.; Heinz, T.F.; Reichman, D.R.; Muller, D.A.; Hone, J.C. Grains and grain boundaries in highly crystalline monolayer molybdenum disulphide. Nat. Mater. 2013, 12, 554–561, doi:10.1038/nmat3633.DOI
  32. [32]Chen, M.; Nam, H.; Wi, S.; Pressnitz, G.; Gunawan, I.M.; Liang, X. Multibit data storage states formed in plasma-treated MoS₂ transistors. ACS Nano 2014, 8, 4023–4032, doi:10.1021/nn501181t.DOI
  33. [33]Lynall, D.; Nair, S.V.; Gutstein, D.; Shik, A.; Savelyev, I.G.; Bumar, H.E. Surface state dynamics dictating transport in InAs nanowires. Nano Lett. 2018, 18, 1387–1395, doi:10.1021/acs.nanolett.7b05106.DOI
  34. [34]Sangwan, V.K.; Jariwala, D.; Kim, I.S.; Chen, K.-S.; Marks, T.J.; Lauhon, L.J.; Hersam, M.C. Gate-tunable memristive phenomena mediated by grain boundaries in single-layer MoS₂. Nat. Nanotechnol. 2015, 10, 403–406, doi:10.1038/nnano.2015.56.DOI
  35. [35]Najmaei, S.; Liu, Z.; Zhou, W.; Zou, X.; Shi, G.; Lei, S.; Yakobson, B.I.; Idrobo, J.-C.; Ajayan, P.M.; Lou, J. Vapour phase growth and grain boundary structure of molybdenum disulphide atomic layers. Nat. Mater. 2013, 12, 754–759, doi:10.1038/nmat3673.DOI
  36. [36]Da Li, W.; Zhu, B.; Zhu, X.; Wang, J.; Ryu, B.; Lu, W.D.; Lu, W.; Liang, X. MoS₂ memristors exhibiting variable switching characteristics toward biorealistic synaptic emulation. ACS Nano 2018, 12, 9240–9252, doi:10.1021/acsnano.8b03977.DOI
  37. [37]Schmidt, H.; Giustiniano, F.; Eda, G. Electronic transport properties of transition metal dichalcogenide field-effect devices: surface and interface effects. Chem. Soc. Rev. 2015, 44, 7715–7736, doi:10.1039/C5CS00275C.DOI
  38. [38]Shi, S.; Sun, Z.; Hu, Y.H. Synthesis, stabilization and applications of 2-dimensional 1T metallic MoS₂. J. Mater. Chem. A 2018, 6, 23932–23977, doi:10.1039/C8TA08152B.DOI
  39. [39]Kim, C.; Issarapane, S.; Moon, I.; Lee, K.Y.; Ra, C.; Lee, S.; Yang, Z.; Yoo, W.J. High-electric-field-induced phase transition and electrical breakdown of MoTe₂. Adv. Electron. Mater. 2020, 6, 1900964, doi:10.1002/aelm.201900964.
  40. [40]Zhang, F.; Zhang, H.; Krylyuk, S.; Milligan, C.A.; Zhu, Y.; Zemlyanov, D.Y.; Bendersky, L.A.; Burton, B.P.; Davydov, A.V.; Appenzeller, J. Electric-field induced structural transition in vertical MoTe₂- and Mo₁₋ₓWₓTe₂-based resistive memories. Nat. Mater. 2019, 18, 55–61, doi:10.1038/s41563-018-0234-y.DOI
  41. [41]Zhu, X.; Da Li; Liang, X.; Lu, W.D. Ionic modulation and ionic coupling effects in MoS₂ devices for neuromorphic computing. Nat. Mater. 2019, 18, 141–148, doi:10.1038/s41563-018-0248-5.DOI
  42. [42]Li, H.; Tao, L.; Xu, J.-B. Intrinsic memristive mechanisms in 2D layered materials for high-performance memory. J. Appl. Phys. 2021, 129, 50902, doi:10.1063/5.0035764.DOI
  43. [43]Komsa, H.-P.; Krasheninnikov, A.V. Native defects in bulk and monolayer MoS₂ from first principles. Phys. Rev. B 2015, 91, 125304, doi:10.1103/PhysRevB.91.125304.DOI
  44. [44]Liu, D.; Guo, Y.; Fang, L.; Robertson, J. Sulfur vacancies in monolayer MoS₂ and its electrical contacts. Appl. Phys. Lett. 2013, 103, 183113, doi:10.1063/1.4824893.DOI
  45. [45]Spetzler, B.; Geng, Z.; Rossnagel, K.; Ziegler, M.; Schwierz, F. Lateral 2D TMDC Memristors – Experiment and Modeling. In 2022 IEEE 16th International Conference on Solid-State & Integrated Circuit Technology (ICSICT). 2022 IEEE 16th International Conference on Solid-State & Integrated Circuit Technology (ICSICT), 2022; pp 1–3.DOI
  46. [46]Sze, S.M.; Crowell, C.R.; Kahng, D. Photoelectric determination of the image force dielectric constant for hot electrons in Schottky barriers. J. Appl. Phys. 1964, 35, 2534–2536, doi:10.1063/1.1702894.DOI
  47. [47]Sze, S.M.; Ng, K.K. Physics of semiconductor devices, 3rd ed.; Wiley-Interscience: Hoboken, N.J., 2007, ISBN 9780470068328.
  48. [48]Vaknin, Y.; Dagan, R.; Rosenwanks, Y. Schottky barrier height and image force lowering in monolayer MoS₂ field effect transistors. Nanomaterials (Basel) 2020, 10, doi:10.3390/nano10122346.DOI
  49. [49]Farrell, P.; Rotundo, N.; Doan, D.H.; Kantner, M.; Fuhrmann, J.; Koprucki, T. Drift-Diffusion Models. In Handbook of optoelectronic device modeling and simulation: Lasers, Modulators, Photodetectors, Solar Cells, and Numerical Methods - Volume Two; Piprek, J., Ed.; CRC Press: Portland, 2017; pp 733–772, ISBN 9781315123318.DOI
  50. [50]Abdel, D.; Vágner, P.; Fuhrmann, J.; Farrell, P. Modelling charge transport in perovskite solar cells: potential-based and limiting ion depletion. Electrochimica Acta 2021, 390, 138696, doi:10.1016/j.electacta.2021.138696.DOI
  51. [51]Crowel, C.R.; Sze, S.M. Current transport in metal-semiconductor barriers. Solid-State Electron. 1966, 9, 1035–1048, doi:10.1016/0038-1101(66)90311-3.
  52. [52]Abdel, D.; Farrell, P.; Fuhrmann, J. Assessing the quality of the excess chemical potential flux scheme for degenerate semiconductor device simulation. Opt Quant Electron 2021, 53, 1–10, doi:10.1007/s11082-021-02803-4.DOI
  53. [53]Fuhrmann, J. VoronoiFVM.jl: Finite volume solver for coupled nonlinear partial differential equations (https://github.com/j-fu/VoronoiFVM.jl), 2019-2023.DOI
  54. [54]Cho, K.; Park, W.; Park, J.; Jeong, H.; Jang, J.; Kim, T.-Y.; Hong, W.-K.; Hong, S.; Lee, T. Electric stress-induced threshold voltage instability of multilayer MoS₂ field effect transistors. ACS Nano 2013, 7, 7751–7758, doi:10.1021/nn402348r.DOI
  55. [55]Illarionov, Y.Y.; Smithe, K.K.H.; Waltl, M.; Knobloch, T.; Pop, E.; Grasser, T. Improved hysteresis and reliability of MoS₂ transistors with high-quality CVD growth and Al₂O₃ encapsulation. IEEE Electron Device Lett. 2017, 38, 1763–1766, doi:10.1109/LED.2017.2768602.DOI
  56. [56]Jariwala, D.; Sangwan, V.K.; Late, D.J.; Johns, J.E.; David, V.P.; Marks, T.J.; Lauhon, L.J.; Hersam, M.C. Band-like transport in high mobility unencapsulated single-layer MoS₂ transistors. Appl. Phys. Lett. 2013, 102, 173107, doi:10.1063/1.4803920.DOI
  57. [57]Le, D.; Rawl, T.B.; Rahman, T.S. Single-layer MoS₂ with sulfur vacancies: structure and catalytic application. J. Phys. Chem. C 2014, 118, 5346–5351, doi:10.1021/jp411256g.DOI
  58. [58]Yang, J.; Kawai, H.; Wong, C.P.Y.; Goh, K.E.J. Electrical doping effect of vacancies on monolayer MoS₂. J. Phys. Chem. C 2019, 123, 2933–2939, doi:10.1021/acs.jpcc.8b10496.DOI
  59. [59]Komsa, H.-P.; Krasheninnikov, A.V. From point to extended defects in two-dimensional MoS₂: Evolution of atomic structure under electron irradiation. Phys. Rev. B 2013, 88, 35301, doi:10.1103/PhysRevB.88.035301.DOI
  60. [60]Li, S.-L.; Tsukagoshi, K.; Orgiu, E.; Samorì, P. Charge transport and mobility engineering in two-dimensional transition metal chalcogenide semiconductors. Chem. Soc. Rev. 2016, 45, 118–151, doi:10.1039/C5CS00517E.DOI
  61. [61]Rasmussen, F.A.; Thygesen, K.S. Computational 2D materials database: electronic structure of transition-metal dichalcogenides and oxides. J. Phys. Chem. C 2015, 119, 13169–13183, doi:10.1021/acs.jpcc.5b02950.DOI
  62. [62]Yun, W.S.; Han, S.W.; Hong, S.C.; Kim, I.G.; Lee, J.D. Thickness and strain effects on electronic structures of transition metal dichalcogenides: 2H-MX₂ semiconductors (M=Mo, W; X=S, Se, Te). Phys. Rev. B 2012, 85, 33305, doi:10.1103/PhysRevB.85.033305.DOI
  63. [63]Kam, K.K.; Parkinson, B.A. Detailed photocurrent spectroscopy of the semiconducting group VI transition metal dichalcogenides. J. Phys. Chem. 1982, 86, 463–467, doi:10.1021/j100393a010.DOI
  64. [64]Mak, K.F.; Lee, C.; Hone, J.; Shan, J.; Heinz, T.F. Atomically thin MoS₂: a new direct-gap semiconductor. Phys. Rev. Lett. 2010, 105, 136805, doi:10.1103/PhysRevLett.105.136805.DOI
  65. [65]Splendiani, A.; Sun, L.; Zhang, Y.; Li, T.; Kim, J.; Chim, C.-Y.; Galli, G.; Wang, F. Emerging photoluminescence in monolayer MoS₂. Nano Lett. 2010, 10, 1271–1275, doi:10.1021/nl903868w.DOI
  66. [66]Mak, K.F.; Lee, C.; Hone, J.; Shan, J.; Heinz, T.F. Atomically thin MoS₂: a new direct-gap semiconductor. Phys. Rev. Lett. 2010, 105, 136805, doi:10.1103/PhysRevLett.105.136805.DOI
  67. [67]Yu, S.; Xiong, H.D.; Eshun, K.; Yuan, H.; Li, Q. Phase transition, effective mass and carrier mobility of MoS₂ monolayer under tensile strain. Appl. Surf. Sci. 2015, 325, 27–32, doi:10.1016/j.apsusc.2014.11.079.DOI
  68. [68]Phuc, H.V.; Hieu, N.N.; Hoi, B.D.; Hieu, N.V.; Thu, T.V.; Hung, N.M.; Ilyasov, V.V.; Poklonski, N.A.; Nguyen, C.V. Tuning the electronic properties, effective mass and carrier mobility of MoS₂ monolayer by strain engineering: first-principle calculations. J. Electron. Mater. 2018, 47, 730–736, doi:10.1007/s11664-017-5843-8.
  69. [69]Zhao, Y.; Ouyang, G. Thickness-dependent photoelectric properties of MoS₂/Si heterostructure solar cells. Sci. Rep. 2019, 9, 17381, doi:10.1038/s41598-019-53936-2.DOI
  70. [70]Xi, J.; Zhao, T.; Wang, D.; Shuai, Z. Tunable electronic properties of two-dimensional transition metal dichalcogenide alloys: a first-principles prediction. J. Phys. Chem. Lett. 2014, 5, 285–291, doi:10.1021/jz402375s.DOI
  71. [71]Ramasubramaniam, A. Large excitonic effects in monolayers of molybdenum and tungsten dichalcogenides. Phys. Rev. B 2012, 86, 115409, doi:10.1103/PhysRevB.86.115409.DOI
  72. [72]Kadantsev, E.S.; Hawrylak, P. Electronic structure of a single MoS₂ monolayer. Solid State Commun. 2012, 152, 909–913, doi:10.1016/j.ssc.2012.02.005.DOI
  73. [73]Davelou, D.; Kopidakis, G.; Kioseoglou, G.; Remediakis, I.N. MoS₂ nanostructures: semiconductors with metallic edges. Solid State Commun. 2014, 192, 42–46, doi:10.1016/j.ssc.2014.04.023.DOI
  74. [74]Kang, Y.; Jeon, D.; Kim, T. Local mapping of the thickness-dependent dielectric constant of MoS₂. J. Phys. Chem. C 2021, 125, 3611–3615, doi:10.1021/acs.jpcc.0c11198.DOI
  75. [75]Chen, X.; Wu, Z.; Xu, S.; Wang, L.; Huang, R.; Han, Y.; Ye, W.; Xiong, W.; Han, T.; Long, G.; et al. Probing the electron states and metal-insulator transition mechanisms in molybdenum disulphide vertical heterostructures. Nat. Commun. 2015, 6, 6088, doi:10.1038/ncomms7088.DOI
  76. [76]Liang, Y.; Huang, S.; Soklaski, R.; Yang, L. Quasiparticle band-edge energy and band offsets of monolayer of molybdenum and tungsten chalcogenides. Appl. Phys. Lett. 2013, 103, 42106, doi:10.1063/1.4816517.DOI
  77. [77]Han, S.W.; Kwon, H.; Kim, S.K.; Ryu, S.; Yun, W.S.; Kim, D.H.; Hwang, J.H.; Kang, J.-S.; Baik, J.; Shin, H.J.; et al. Band-gap transition induced by interlayer van der Waals interaction in MoS₂. Phys. Rev. B 2011, 84, 45409, doi:10.1103/PhysRevB.84.045409.DOI
  78. [78]Das, S.; Chen, H.-Y.; Penumatcha, A.V.; Appenzeller, J. High performance multilayer MoS₂ transistors with scandium contacts. Nano Lett. 2013, 13, 100–105, doi:10.1021/nl303583v.DOI
  79. [79]Nourbakhsh, A.; Zubair, A.; Sajjad, R.N.; Tavakkoli K G, A.; Chen, W.; Fang, S.; Ling, X.; Kong, J.; Dresselhaus, M.S.; Kaaxiras, E.; et al. MoS₂ field-effect transistor with sub-10 nm channel length. Nano Lett. 2016, 16, 7798–7806, doi:10.1021/acs.nanolett.6b03999.DOI
  80. [80]Xiao, J.; Zhang, Y.; Chen, H.; Xu, N.; Deng, S. Enhanced performance of a monolayer MoS₂/WSe₂ heterojunction as a photoelectrochemical cathode. Nano-Micro Lett. 2018, 10, 60, doi:10.1007/s40820-018-0212-6.DOI
  81. [81]Yu, Z.; Ong, Z.-Y.; Li, S.; Xu, J.-B.; Zhang, G.; Zhang, Y.-W.; Shi, Y.; Wang, Y. Analyzing the carrier mobility in transition-metal dichalcogenide MoS₂ field-effect transistors. Adv. Funct. Mater. 2017, 27, 1604093, doi:10.1002/adfm.201604093.DOI
  82. [82]Li, S.-L.; Wakabayashi, K.; Xu, Y.; Nakaharai, S.; Komatsu, K.; Li, W.-W.; Lin, Y.-F.; Aparéício-Ferreira, A.; Tsukagoshi, K. Thickness-dependent interfacial coulomb scattering in atomically thin field-effect transistors. Nano Lett. 2013, 13, 3546–3552, doi:10.1021/nl4010783.DOI
  83. [83]Radisavljevic, B.; Radenovic, A.; Brivio, J.; Giacometti, V.; Kis, A. Single-layer MoS₂ transistors. Nat. Nanotechnol. 2011, 6, 147–150, doi:10.1038/nnano.2010.279.DOI
  84. [84]Kim, C.; Moon, H.; Lee, D.; Choi, M.S.; Ahmed, F.; Nam, S.; Cho, Y.; Shin, H.-J.; Park, S.; Yoo, W.J. Fermi level pinning at electrical metal contacts of monolayer molybdenum dichalcogenides. ACS Nano 2017, 11, 1588–1596, doi:10.1021/acsnano.6b07159.DOI
  85. [85]Allain, A.; Kang, J.; Banerjee, K.; Kis, A. Electrical contacts to two-dimensional semiconductors. Nat. Mater. 2015, 14, 1195–1205, doi:10.1038/nmat4452.DOI
  86. [86]Wang, T.; Shi, Y.; Puglisi, F.M.; Chen, S.; Zhu, K.; Zuo, Y.; Li, X.; Jing, X.; Han, T.; Guo, B.; et al. Electroforming in metal-oxide memristive synapses. ACS Appl. Mater. Interfaces 2020, 12, 11806–11814, doi:10.1021/acsami.9b19362.DOI
  87. [87]Yoo, J.; Woo, J.; Song, J.; Hwang, H. Threshold switching behavior of Ag-Si based selector device and hydrogen doping effect on its characteristics. AIP Advances 2015, 5, 127221, doi:10.1063/1.4938548.DOI
  88. [88]Jiang, H.; Belkin, D.; Savel'ev, S.E.; Lin, S.; Wang, Z.; Li, Y.; Joshi, S.; Midya, R.; Li, C.; Rao, M.; et al. A novel true random number generator based on a stochastic diffusive memristor. Nat. Commun. 2017, 8, 882, doi:10.1038/s41467-017-00869-x.DOI
  89. [89]Vahl, A.; Carstens, N.; Strunskus, T.; Faupel, F.; Hassanien, A. Diffusive Memristive Switching on the Nanoscale, from Individual Nanoparticles towards Scalable Nanocomposite Devices. Sci. Rep. 2019, 9, 17367, doi:10.1038/s41598-019-53720-2.DOI
  90. [90]Wang, Z.; Wu, H.; Burr, G.W.; Hwang, C.S.; Wang, K.L.; Xia, Q.; Yang, J.J. Resistive switching materials for information processing. Nat. Rev. Mater. 2020, 5, 173–195, doi:10.1038/s41578-019-0159-3.DOI
  91. [91]Wang, L.; Shen, X.; Gao, Z.; Fu, J.; Yao, S.; Cheng, L.; Lian, M. Review of recent advances of 2D materials in memristive neuromorphic circuits. J. Mater. Sci. 2022, 57, 4915–4940, doi:10.1007/s10853-022-06954-x.DOI
  92. [92]Ielmini, D.; Waser, R. Resistive Switching; Wiley-VCH Verlag GmbH & Co. KGaA: Weinheim, Germany, 2016, ISBN 3527680934.
  93. [93]Genreith-Schriever, A.R.; Souza, R.A. de. Field-enhanced ion transport in solids: reexamination with molecular dynamics simulations. Phys. Rev. B 2016, 94, 224304, doi:10.1103/PhysRevB.94.224304.
  94. [94]Chen, M.; Nam, H.; Rokni, H.; Wi, S.; Yoon, J.S.; Chen, P.; Kurabayashi, K.; Lu, W.; Liang, X. Nanoimprint-assisted shear exfoliation (NASE) for producing multilayer MoS₂ structures as field-effect transistor channel arrays. ACS Nano 2015, 9, 8773–8785, doi:10.1021/acsnano.5b01715.DOI
  95. [95]Chen, M.; Wi, S.; Nam, H.; Priessnitz, G.; Liang, X. Effects of MoS₂ thickness and air humidity on transport characteristics of plasma-doped MoS₂ field-effect transistors. J. Vac. Sci. Technol. B: Nanotechnol. Microelectron 2014, 32, 06FF02, doi:10.1116/1.4897133.DOI
  96. [96]Lu, Z.; Lee, O.; Wong, J.C.; Salahuddin, S. Surface states in a monolayer MoS₂ transistor. J. Mater. Res. 2016, 31, 911–916, doi:10.1557/jmr.2015.405.DOI
  97. [97]Mukherjee, B.; Tseng, F.; Gunlycke, D.; Amara, K.K.; Eda, G.; Simsek, E. Complex electrical permittivity of the monolayer molybdenum disulfide (MoS₂) in near UV and visible. Opt. Mater. Express 2015, 5, 447, doi:10.1364/OME.5.000447.DOI
  98. [98]Li, S.-L.; Miyazaki, H.; Song, H.; Kuramochi, H.; Nakaharai, S.; Tsukagoshi, K. Quantitative Raman spectrum and reliable thickness identification for atomic layers on insulating substrates. ACS Nano 2012, 6, 7381–7388, doi:10.1021/nn205173.DOI
  99. [99]Shen, C.-C.; Hsu, Y.-T.; Li, L.-J.; Liu, H.-L. Charge dynamics and electronic structures of monolayer MoS₂ films grown by chemical vapor deposition. Appl. Phys. Express 2013, 6, 125801, doi:10.7567/APEX.6.125801.DOI
  100. [100]Liu, J.-T.; Wang, T.-B.; Li, X.-J.; Liu, N.-H. Enhanced absorption of monolayer MoS₂ with resonant back reflector. J. Appl. Phys. 2014, 115, 193511, doi:10.1063/1.4878700.DOI
  101. [101]Kumar, A.; Ahluwalia, P.K. Tunable dielectric response of transition metals dichalcogenides MX₂ (M=Mo, W; X=S, Se, Te): Effect of quantum confinement. Phys. B: Condens. Matter 2012, 407, 4627–4634, doi:10.1016/j.physb.2012.08.034.DOI
  102. [102]Stefan Seeger. Herstellung und Charakterisierung der Schichtgittersulfide MoS₂ und WS₂ für photovoltaische Anwendungen. PhD Thesis; Technische Universität Berlin, Berlin, 2006.DOI
  103. [103]Dirkmann, S.; Kaiser, J.; Wenger, C.; Mussenbrock, T. Filament growth and resistive switching in hafnium oxide memristive devices. ACS Appl. Mater. Interfaces 2018, 10, 14857–14868, doi:10.1021/acsami.7b19836.DOI

Paper details

vlnc:W_ahx36xnx
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents