betaFeedback

WorksPhysics and Astronomy

Blocking transition of interface traps in MoS2_2-on-SiO2_2 FETs

Abstract

Electrical conductivity with gate-sweep in a few layer MoS2_2-on-SiO2_2 field-effect-transistor shows an abrupt reduction in hysteresis when cooled. The hysteresis and time dependent conductivity of the MoS2_2 channel are modeled using the dynamics of interface traps' occupancy. The reduction in hysteresis is found to be steepest at a blocking temperature near 225 K. This is attributed to the interplay between thermal and barrier energies and fitted using a distribution of the latter. Further, the charge stored in the blocked traps is programmed at low temperatures by cooling under suitable gate voltage. Thus the threshold gate-voltage in nearly non-hysteretic devices at 80 K temperature is reversibly controlled over a wide range.

Introduction

Single and few layer transition metal chalcogenides [1, 2] offer much potential for device applications including transistors [3, 4] with high frequency capability [5], logic gates [6, 7] for integrated circuits [8] and optoelectronic [9, 10, 11, 12] devices. The MoS2_2 single layer devices with direct band gap [1, 3] in optical range have been of particular interest. The field effect transistors (FETs) based on MoS2_2 show a very promising behavior with scalability, non-ideal behavior and degradation with time as the main hurdles. A pertinent culprit with interesting physics that contributes to non-ideal behavior is interface traps. Such traps lead to reduced mobility and response time as well as increased noise and hysteresis in transfer characteristics. Thus a more comprehensive understanding of the traps is necessary.

The phenomenon of blocking is common in magnetic systems. In ferromagnetic nano-particles, exhibiting superparamagnetism, [13] the blocking arises from the interplay between an anisotropy energy barrier, the thermal energy and the Zeeman energy. As a result, the thermally activated switching rate (τs(T)−1\tau_s(T)^{-1}) between two magnetic states is a sharply rising function of temperature. Thus, their response to an applied magnetic field, measured over certain time (τm\tau_m), shows hysteresis at low temperatures and non-hysteretic paramagnetic behavior at high temperatures. This crossover transition occurs at a blocking temperature TBT_B at which τs(TB)≈τm\tau_s(T_B) \approx\tau_m. In contrast, the blocking of traps in MoS2_2 FETs leads to hysteresis reduction with cooling. A similar behavior is also observed in graphene FETs [14]; though, the much sharper transfer characteristics in MoS2_2 devices with a threshold gate-voltage help carry out a more quantitative analysis.

The threshold gate-voltage at which an FET shows a steep rise in conductance is controlled by both the traps’ charge and the capacitive displacement charge across gate dielectric. A positive hysteresis in the transfer characteristics of MoS2_2 FETs has been studied as a function of various parameters [15, 16, 17, 18] and attributed mainly to charge-traps. This arises from the traps that have a timescale comparable to the gate-voltage sweep-time. This also amounts to a relaxation in channel’s conductance at varying time-scales. The fast traps do not lead to hysteresis but they shield the gate electric field restricting the density of mobile carriers in the channel. This broadens the threshold region and forbids the access to the ambipolar behavior in MoS2_2 FETs even for gate voltages far exceeding the voltage equivalent to the energy gap. In addition, the electrostatic potential of the trap ions leads to reduced mobility of the channel carriers while the variation in the charge-state of traps gives rise to carrier density and mobility fluctuations.

In this paper, the transfer characteristics and its’ time dependence in few layer MoS2_2-on-SiO2_2 FETs as a function of temperature is presented together with a model on the effect of traps on gate-dependent channel conductance. A trap’s charge state determines the channel’s chemical potential which in turn dictates the traps’ occupancy. This makes it a complex non-linear system with coupling between traps’ occupancy mediated by the channel. Thus, even the traps at a single energy and with the same barrier lead to non-exponential relaxation. The hysteresis and its temperature dependence is modeled using some simplifications and analogy with superparamagnets. Finally, the traps’ blocking is used to reversibly control the threshold voltage at 80 K temperature.

Experimental details

Few layer MoS2_2 was transferred on SiO2_2 by a dry method [19] from a natural MoS2_2 single crystal (from SPI, USA) using commercial PDMS film based viscoelastic stamp. The latter is first fixed on a glass slide and an MoS2_2 flake is transferred on it using a scotch tape. The mechanism of this transfer process uses the viscoelastic response of the PDMS film, which behaves as an elastic solid for a short time scale. So pulling the PDMS film from the scotch tape is done at high speed leading to strong adhesion of MoS2_2 on PDMS as viscoelastic solid makes a strong conformal contact with the flake [20]. The PDMS with MoS2_2 flake is aligned with a SiO2_2/Si substrate fixed by carbon tape on a XYZ micro-manipulator and under an optical microscope. The stamp is removed with sufficiently low speed so that the adhesion of the flake to stamp is week and the flake gets transferred to the SiO2\mathrm{SiO_2} surface easily. Raman spectra, see Figure 1(c), were used to confirm few-layer nature of MoS2\mathrm{MoS_2}.

Optical image, electrical schematic, Raman spectra, and current-voltage characteristics of a few-layer MoS2 device

Figure 1. (a) Optical image of few-layer MoS2\mathrm{MoS_2} with gold contacts. (b) The electrical schematic drawing of MoS2\mathrm{MoS_2} FET. (c) Raman spectra measured on exfoliated single layer, few-layer and bulk MoS2\mathrm{MoS_2}. (d) IdsI_{\mathrm{ds}} Vs VdsV_{\mathrm{ds}} for a few-layer device at different gate voltage values.

The number of MoS2\mathrm{MoS_2} layers is determined by optical microscope contrast and verified by Raman Spectroscopy with 532 nm wavelength laser excitation. As seen in Figure 1(c) the separation between the E2g1E^{1}_{2g} and A1gA_{1g} Raman peaks is 18.47, 20.20 and 24.1 cm−1^{-1}, which correspond to the single-layer, few-layer and bulk MoS2\mathrm{MoS_2}, respectively [21, 22].

We make 50 nm thick gold film source-drain contacts using mechanical masking with a 15 μ\mum diameter tungsten wire. Use of Au without Cr/Ti adhesion layer promotes Ohmic contacts due to a very small difference in contact potentials of Au and MoS2\mathrm{MoS_2} [23, 24]. Mechanical masking avoids use of organic lithography resist which can leave residue on MoS2\mathrm{MoS_2}. The wire is carefully aligned under an optical microscope with few-layer MoS2\mathrm{MoS_2} on SiO2\mathrm{SiO_2} substrate. Figure 1(a) shows an optical micrograph of a MoS2\mathrm{MoS_2} device with source-drain contacts. Two probe conductance down to 80 K temperature was measured, with the configuration shown in Figure 1(b) in a homemade vacuum cryostat with a heater for temperature control. A 10 kΩ\Omega series resistance was connected with the gate voltage supply, which was controlled by a data acquisition card using a LabView program. The Ohmic contacts were confirmed by two probe current-voltage characteristics as shown in Figure 1(d). The cryostat was pumped

by a turbo-molecular pump to less than 10−410^{-4} mbar pressure. When the cryostat is dipped into liquid nitrogen for cooling, the vacuum is expected to be much better than this. The device was annealed at 400 K in vacuum to minimize adsorbates on Mo2\mathrm{Mo_2} surface.

Modeling of channel transport with time dependent interface traps

In this section the temperature and time dependent channel transport in presence of interface traps is modeled. We first discuss the conduction in intrinsic 2D channel in presence of back-gate voltage. This is followed by details of the traps, their dynamics and their collective influence on the channel to model its time dependent transport and hysteresis. Finally we make some simplifications to model the temperature dependence of channel hysteresis and blocking transition of the traps.

Intrinsic 2D semiconductor channel

Consider an intrinsic 2D semiconductor with dispersion E(kx,ky)=ℏ2(kx2+ky2)/2m∗E(k_x,k_y)=\hbar^2(k_x^2+k_y^2)/2m^* on a gate-oxide with m∗m^* as the effective mass. This leads to an energy independent density of states (DOS) g(E)=g2D=gsgvm∗/2πℏ2g(E)=g_{2D}=g_sg_vm^*/2\pi\hbar^2 with gsg_s as spin- and gvg_v as valley-degeneracy. The electron and hole densities, nn and pp, respectively, will be given by

n=∫Ec∞g(E)f(T,E−μch)dEandp=∫−∞Evg(E)[1−f(T,E−μch)]dE.n=\int_{E_c}^{\infty}g(E)f(T,E-\mu_{\mathrm{ch}})\mathrm{d}E \quad\text{and}\quad p=\int_{-\infty}^{E_v}g(E)[1-f(T,E-\mu_{\mathrm{ch}})]\mathrm{d}E.

Here μch\mu_{\mathrm{ch}} is the chemical potential of the channel and f(T,E)=[1+exp⁡(E/kBT)]−1f(T,E)=[1+\exp(E/k_BT)]^{-1} is the Fermi function. Eventually, the constant DOS in 2D leads to the expressions for nn and pp as

n(T,μch)=g2DkBTln⁡[1+exp⁡{β(μch−Ec)}]p(T,μch)=g2DkBTln⁡[1+exp⁡{β(Ev−μch)}].(1)\begin{aligned} n(T,\mu_{\mathrm{ch}})&=g_{2D}k_BT\ln[1+\exp\{\beta(\mu_{\mathrm{ch}}-E_c)\}]\\ p(T,\mu_{\mathrm{ch}})&=g_{2D}k_BT\ln[1+\exp\{\beta(E_v-\mu_{\mathrm{ch}})\}]. \tag*{(1)} \end{aligned}

The net charge density in the channel σch=e(p−n)\sigma_{\mathrm{ch}}=e(p-n), in general, arises from dopants, gate electric field and interface trap charges. For an intrinsic 2D MoS2\mathrm{MoS_2} channel with zero gate electric field and no traps σch=0\sigma_{\mathrm{ch}}=0 and thus n=pn=p and so the chemical potential μch0=(Ec+Ev)/2\mu_{\mathrm{ch}}^0=(E_c+E_v)/2. This also assumes identical DOS for the valence and conduction bands. When a gate voltage VgV_g is applied the channel potential changes to VchV_{\mathrm{ch}} and μch=μch0+eVch\mu_{\mathrm{ch}}=\mu_{\mathrm{ch}}^0+eV_{\mathrm{ch}}, see Figure 2. This leads to a non-zero σch\sigma_{\mathrm{ch}} given by

Channel bands and electrostatic potential profiles for different gate voltages

Figure 2. (a) and (b) show the channel bands while (c) and (d) show the electrostatic potential profile between the channel and the gate. (a), (c) are for Vg=0V_{\mathrm{g}} = 0 and (b), (d) for Vg>0V_{\mathrm{g}} > 0. The channel potential (VchV_{\mathrm{ch}}) rises with VgV_{\mathrm{g}} and leads to a downward shift of the bands by eVcheV_{\mathrm{ch}} relative to the channel’s chemical potential μch\mu_{\mathrm{ch}} which is fixed to the ground reference. The jump by VchV_{\mathrm{ch}} between drain (or source) and the channel, see (d), arises due to channel’s quantum capacitance when the channel acquires a charge density.

σch=−γCox(kBTe)ln⁡[f(T,Ev−μch0−eVch)f(T,μch0+eVch−Ec)].(2)\sigma_{\mathrm{ch}}=-\gamma C_{\mathrm{ox}}\left(\frac{k_BT}{e}\right)\ln\left[\frac{f(T,E_v-\mu_{\mathrm{ch}}^0-eV_{\mathrm{ch}})}{f(T,\mu_{\mathrm{ch}}^0+eV_{\mathrm{ch}}-E_c)}\right]. \tag*{(2)}

Here, the dimensionless γ=e2g2D/Cox\gamma=e^2g_{2D}/C_{\mathrm{ox}} is the ratio of channel’s quantum capacitance in the degenerate limit and the per-unit-area gate-oxide capacitance. The latter is given by Cox=κϵ0/dC_{\mathrm{ox}}=\kappa\epsilon_0/d with d=300d=300 nm as SiO2\mathrm{SiO_2} thickness, κ=4\kappa=4 as its dielectric constant and ϵ0\epsilon_0 as the permittivity of free space. The non-linear relation between σch\sigma_{\mathrm{ch}} and VchV_{\mathrm{ch}} in Eq. (2) amounts to a non-linear quantum capacitance [25] of the channel. Note that a positive VchV_{\mathrm{ch}} leads to a negative σch\sigma_{\mathrm{ch}} and thus an increase in electron density. A positive VgV_{\mathrm{g}} leads to a positive charge density at the gate electrode which will be equal and opposite to the combined charge densities of the channel carriers, dopant ions and trap ions.

As shown in Figure 2(d), the overall applied VgV_{\mathrm{g}}, between drain/source, kept at ground potential, and the gate electrode, is shared between the channel potential VchV_{\mathrm{ch}} and the voltage drop across the gate dielectric. For positive VgV_{\mathrm{g}} and with μch\mu_{\mathrm{ch}} as the fixed zero energy reference, the electron (negative charge) energy-bands of the channel shift downward by eVcheV_{\mathrm{ch}} leading to an increase in electron density in the channel. Thus Vg−Vch=−σch/CoxV_{\mathrm{g}} - V_{\mathrm{ch}} = -\sigma_{\mathrm{ch}}/C_{\mathrm{ox}}, in the absence of dopants and traps. This and Eq. (2) lead to

Vg=Vch+γ(kBTe)ln⁡[f(T,Ev−μch0−eVch)f(T,μch0+eVch−Ec)].(3)V_{\mathrm{g}} = V_{\mathrm{ch}} + \gamma\left(\frac{k_{\mathrm{B}}T}{e}\right) \ln\left[\frac{f(T,E_{\mathrm{v}}-\mu_{\mathrm{ch}}^{0}-eV_{\mathrm{ch}})}{f(T,\mu_{\mathrm{ch}}^{0}+eV_{\mathrm{ch}}-E_{\mathrm{c}})}\right]. \tag*{(3)}

Assuming same mobility μ\mu and DOS for e and h, the channel’s conductivity G=(n+p)e2μG = (n+p)e^{2}\mu is given by,

G=−e2μg2DkBT×ln⁡[f(T,Ev−μch0−eVch)f(T,μch0+eVch−Ec)].(4)G = -e^{2}\mu g_{2\mathrm{D}}k_{\mathrm{B}}T \times\ln\left[f(T,E_{\mathrm{v}}-\mu_{\mathrm{ch}}^{0}-eV_{\mathrm{ch}})f(T,\mu_{\mathrm{ch}}^{0}+eV_{\mathrm{ch}}-E_{\mathrm{c}})\right]. \tag*{(4)}

For single-layer MoS2_2 we use gs=2g_{\mathrm{s}} = 2, gv=1g_{\mathrm{v}} = 1 and m∗=0.57mem^{*} = 0.57m_{\mathrm{e}} to get g2D=2.64×1014 eV−1cm−2g_{2\mathrm{D}} = 2.64 \times10^{14}\ \mathrm{eV}^{-1}\mathrm{cm}^{-2} and with d=300 nmd = 300\ \mathrm{nm} we get γ=3570\gamma= 3570. Further, for intrinsic MoS2_2 channel we take μch=0\mu_{\mathrm{ch}} = 0, Ec=1 eVE_{\mathrm{c}} = 1\ \mathrm{eV} and Ev=−1 eVE_{\mathrm{v}} = -1\ \mathrm{eV}. The σch\sigma_{\mathrm{ch}}, VchV_{\mathrm{ch}} and ln⁡(G)\ln(G), thus obtained, are plotted as a function of VgV_{\mathrm{g}} in Figure 3 using Eqs. (2), (3) and (4), respectively, at T=300 KT = 300\ \mathrm{K}. We see that the channel conductivity and carrier density stay close to zero for ∣eVg∣<Eg/2|eV_{\mathrm{g}}| < E_{\mathrm{g}}/2 and rise abruptly beyond this range. Thus we expect the threshold voltages eVth∼±Eg/2eV_{\mathrm{th}} \sim\pm E_{\mathrm{g}}/2. This is far from what is seen in actual experiments. We next discuss traps’ energies, relative to channel chemical potential, and the barrier for charge exchange with the channel.

Variation with gate voltage of channel potential, channel charge density, and conductance

Figure 3. Variation with VgV_{\mathrm{g}} of VchV_{\mathrm{ch}} in Volts, channel charge density σ\sigma in γCox(kBT/e)\gamma C_{\mathrm{ox}}(k_{\mathrm{B}}T/e) units and ln⁡(G)\ln(G) for no traps case. One can see that for this case the eVtheV_{\mathrm{th}} values for channel conduction are close to the conduction and valence band energies.

Charge traps near 2D channel

Traps have some similarities to dopants. Near room temperature a donor dopant exists in a semiconductor as a positively charged ion and a band-electron binds to it with an energy EdE_{\mathrm{d}}, just below EcE_{\mathrm{c}}. A negatively charged acceptor dopant, on the other hand, has a bound-hole state with an energy EaE_{\mathrm{a}}, just above EvE_{\mathrm{v}}. An ionized donor dopant leads to a mobile electron in the conduction band while keeping the overall system charge neutral. As dictated by the Fermi distribution, a bound state at EdE_{\mathrm{d}} is unfavorable as compared to an electron at the chemical potential μch\mu_{\mathrm{ch}} of the band-system as μch\mu_{\mathrm{ch}} is close to the middle of the gap for low doping. Similarly, an acceptor dopant leads to a hole in the valence band keeping the overall system neutral and as compared to a hole at the chemical potential of the band system a hole-bound state at EaE_{\mathrm{a}} is unfavorable.

The interface traps differ from dopants in three major ways. First, they are weakly coupled to the channel with an energy barrier between the trap- and channel-bound electron states. The barrier heights for different traps can be different. Small barrier heights will lead to a very small electron exchange time. Second, the energy for trap bound electron state need not be close to EcE_c and it can be anywhere relative to the bands. Third, a band-hole or band-electron will not be able to bind to a trap, and particularly so for large barriers or small coupling. The channel carriers will feel a weak electrostatic potential of the trap ion affecting its mobility. Further, a trap, assumed to be isolated in the sense of not interacting with other traps, will only form localized states and not extended band-like states.

The relative energies of an electron bound to a trap versus that in the channel will be dictated by the electrochemical reduction potential of the two. A more positive reduction potential indicates an increased affinity for electron. For a donor trap DD if the reduction potential for the reaction D++e−→DD^{+} + e^{-} \to D exceeds the channel’s reduction potential, i.e. Ch+e−→Ch−Ch + e^{-} \to Ch^{-} then it is energetically favorable for donor trap to stay in neutral state. In the schematic band-energy diagram, see Figure 4(a), this can be depicted as an electron-occupied neutral donor trap-level being lower than μch\mu_{\mathrm{ch}} by its excess reduction-potential relative to the channel. This also means that if μch\mu_{\mathrm{ch}} is decreased, see Figure 4(b), it will become energetically favorable for the donor trap to transfer an electron to the channel and exist as D+D^{+} [26]. The actual electron transfer can be slow depending on the barrier height between the channel and trap-bound states. Similar argument can be put forward for an acceptor trap. If the reduction potential for A+e−→A−A + e^{-} \to A^{-} exceeds that of Ch++e−→ChCh^{+} + e^{-} \to Ch, then it is energetically favorable for the acceptor trap to be in A−A^{-} state. Figure 4(c),(Figure 4d) depict

Schematics of donor and acceptor trap charge states for different gate voltages

Figure 4. Schematics to illustrate the correlation between the equilibrium charge state of the donor and acceptor traps for different chemical potential values of the channel. The energies of the two traps relative to μch\mu_{\mathrm{ch}} are dictated by the electrochemical reduction potentials as discussed in the text. When VgV_g is reduced [from (a) to (b) or (c) to (d)] leading to raising of bands (relative to μch\mu_{\mathrm{ch}}) and trap level, both the donor trap and acceptor trap, at the energies shown, lose an electron for the new equilibrium state.

the two energetically favorable scenarios for μch\mu_{\mathrm{ch}} relative to trap energy. Over the limited accessible gate-voltage range the other higher ionization states of the traps may not be accessible.

In equilibrium, a donor trap with energy several kBTk_{\mathrm{B}}T lower than μch\mu_{\mathrm{ch}} will remain unionized and will not contribute carriers to the channel. Thus, the traps that are far away in energy from the range of interest of μch\mu_{\mathrm{ch}} will not change their charge state. A net charge density, due to such far-energy traps, will give rise to an equal and opposite charge density in the channel and it can be incorporated in the model through a fixed σ0\sigma_0.

Further, the acceptor traps that can change their state in the relevant range of μch\mu_{\mathrm{ch}} can be, for the modeling purpose, considered as donor traps by incorporating an appropriate change in σ0\sigma_0. Suppose the areal density of acceptor and donor traps, respectively, at a given energy EE is NAN_A and NDN_D. When a fraction 1−x1-x and xx, respectively, of these traps are ionized the total charge density in the traps will be σ0−e(1−x)NA+exND\sigma_0 - e(1-x)N_A + exN_D, i.e. σ0−eNA+ex(NA+ND)\sigma_0 - eN_A + ex(N_A + N_D). This can be interpreted as a fixed charge density σ0−eNA\sigma_0 - eN_A together with NA+NDN_A + N_D donor traps of which xx fraction is ionized. This will be indistinguishable, for modeling the channel conduction, from the actual state. Alternatively, we can simplify by turning the donors into acceptors with appropriate change in fixed charge; however, we adopt the former as a convention.

The interface traps, all donors, can thus be assumed to have certain energy dependent density of states. The other characteristic feature is the barrier for electron exchange between the trap and the channel. This determines the time scale over which the electron can transfer, between a trap state and the mobile channel states, by thermal activation or perhaps by tunneling.

A simple model on the dynamics of trap-charge is presented in the next section while here we illustrate the equilibrium channel properties due to fast traps having a constant density of states gftrg_{\mathrm{ftr}}. These traps act faster than the gate-voltage sweep time scale. Thus channel and traps exist in equilibrium for all VgV_g values where the (donor) traps with energy sufficiently below μch\mu_{\mathrm{ch}} will be in neutral state while those above will be in the positively charged state. There is a fixed charge density σ0\sigma_0 in the traps as discussed earlier. In such equilibrium when VgV_g is increased from zero, μch\mu_{\mathrm{ch}} will change to μch0+eVch\mu_{\mathrm{ch}}^0 + eV_{\mathrm{ch}} such that the interface traps in the energy range eVcheV_{\mathrm{ch}} will get neutralized from their positively ionized state, see Figure 4(a) and (Figure 4b). Therefore, the interface-traps’ charge density will change from σ0\sigma_0 to σ0−e2Vchgftr\sigma_0 - e^2V_{\mathrm{ch}}g_{\mathrm{ftr}}. This will contribute an equal and opposite charge at the gate electrode. Thus Eq. (3) will change to

Vg=(1+γftr)Vch−σ0Cox+γ(kBTe)ln⁡[f(T,Ev−μch0−eVch)f(T,μch0+eVch−Ec)],(5)V_g = (1 + \gamma_{\mathrm{ftr}})V_{\mathrm{ch}} - \frac{\sigma_0}{C_{\mathrm{ox}}} + \gamma\left(\frac{k_{\mathrm{B}}T}{e}\right)\ln\left[\frac{f(T,E_v-\mu_{\mathrm{ch}}^0-eV_{\mathrm{ch}})}{f(T,\mu_{\mathrm{ch}}^0+eV_{\mathrm{ch}}-E_c)}\right], \tag*{(5)}

with γftr=e2gftr/Cox\gamma_{\rm ftr}=e^2g_{\rm ftr}/C_{\rm ox}, i.e. the ratio of traps’ quantum capacitance to the gate capacitance. The other two equations, i.e. Eq. (2) and 4, remain the same. Figure 5 shows the calculated variation of VchV_{\rm ch}, σch\sigma_{\rm ch} and ln⁡(G)\ln(G) for an MoS2_2 channel at T=300T=300 K as a function of VgV_g. This is for γftr=50\gamma_{\rm ftr}=50, corresponding to gftr=3.7×1012 eV−1−cm−2g_{\rm ftr}=3.7 \times10^{12}\ {\rm eV}^{-1}{\rm-cm}^{-2}, and μch0=0.7\mu_{\rm ch}^{0}=0.7 eV. Using Eq. (5) with Vch=0=VgV_{\rm ch}=0=V_g, the latter corresponds to static charge, σ0/e=g2DkBTln⁡[f(T,Ev−μch0)/f(T,μch0−Ec)]=6.3×107 cm−2\sigma_0/e=g_{2D}k_BT\ln[f(T,E_v-\mu_{\rm ch}^{0})/f(T,\mu_{\rm ch}^{0}-E_c)]=6.3 \times10^{7}\ {\rm cm}^{-2}.

Variation with $V_g$ of $V_{\rm ch}$, channel charge density $\sigma$ and $\ln(G)$

Figure 5. Variation with VgV_g of VchV_{\rm ch} in Volts, channel charge density σ\sigma in γCox(kBT/e)\gamma C_{\rm ox}(k_BT/e) units and ln⁡(G)\ln(G) for γftr=50\gamma_{\rm ftr}=50, corresponding to gftr=3.7×1012 eV−1−cm−2g_{\rm ftr}=3.7 \times10^{12}\ {\rm eV}^{-1}{\rm-cm}^{-2}, and μch0=0.7\mu_{\rm ch}^{0}=0.7 eV. One can see that for this case the eVtheV_{\rm th} values for channel conduction are much larger than the channel energy gap.

A non-zero gftrg_{\rm ftr}, thus, slows down the change in VchV_{\rm ch} with VgV_g and also leads to an increase in the VthV_{\rm th} values. This could be a reason for not seeing the hole doped transport regime over a large VgV_g range accessible in MoS2_2 FETs. In the non-degenerate limit, i.e. (μch−Ev),(Ec−μch)≫kBT(\mu_{\rm ch}-E_v),(E_c-\mu_{\rm ch})\gg k_BT, we use Eq. (4) to get dln⁡(G)/dVch=e/kBTd\ln(G)/dV_{\rm ch}=e/k_BT for electron doping and Eq. (5) to get dVg/dVch=1+γftrdV_g/dV_{\rm ch}=1+\gamma_{\rm ftr}. This leads to the expression for the subthreshold swing (SS) as an experimental method to find γftr\gamma_{\rm ftr}, i.e

SS=(dlog⁡(G)dVg)−1=kBTln⁡10e(1+γftr).(6){\rm SS}=\left(\frac{d\log(G)}{dV_g}\right)^{-1}=\frac{k_BT\ln10}{e}(1+\gamma_{\rm ftr}). \tag*{(6)}

Trap dynamics

The fast traps primarily increase the magnitude of VthV_{\rm th} and SS. The slow traps, particularly the ones having a time scale comparable to VgV_g sweep time, are responsible for the positive hysteresis. The traps that are extremely slow to respond over VgV_g sweep time will only lead to a shift in σ0\sigma_0. Figure 6(a) depicts a schematic where an electron can either be in the channel at μch\mu_{\rm ch} or in the trap at energy EE. There is a barrier between these two

Schematic of the barrier between the trap and the channel, and trap levels relative to the bands

Figure 6. (a) Schematic of the barrier between the trap at energy EE and the channel filled with electrons up to its chemical potential μch\mu_{\rm ch}. (b) The trap levels relative to the bands. In equilibrium, the filled (orange) and empty (blue) states of the traps and the band are dictated by the Fermi function with chemical potential μch\mu_{\rm ch}.

states whose height Δ1,2\Delta_{1,2} will appear different from the two sides. This results into different transition rates τ1,2−1\tau_{1,2}^{-1} from the two sides. The time evolution of occupancy pp, of an electron being in the trap at energy EE, will be dictated by,

dpdt=−τ2−1p+τ1−1(1−p)=τ1−1−(τ1−1+τ2−1)p.(7)\frac{dp}{dt}=-\tau_2^{-1}p+\tau_1^{-1}(1-p)=\tau_1^{-1}-(\tau_1^{-1}+\tau_2^{-1})p. \tag*{(7)}

This leads to the solution

p(t,E)=τ2τ1+τ2+[p(0,E)−τ2τ1+τ2]e−(τ1−1+τ2−1)t.(8)p(t,E)=\frac{\tau_2}{\tau_1+\tau_2}+\left[p(0,E)-\frac{\tau_2}{\tau_1+\tau_2}\right]e^{-(\tau_1^{-1}+\tau_2^{-1})t}. \tag*{(8)}

Thus, at equilibrium, i.e. t→∞t\rightarrow\infty, peq=τ2/(τ1+τ2)p_{\rm eq}=\tau_2/(\tau_1+\tau_2). Assuming identical attempt rates τa\tau_a from the two sides, we get τ1,2=τaexp⁡(Δ1,2/kBT)\tau_{1,2}=\tau_a\exp(\Delta_{1,2}/k_BT). With the barrier height difference Δ2−Δ1=μch−E\Delta_2-\Delta_1=\mu_{\rm ch}-E we get τ2/τ1=exp⁡[(μch−E)/kBT]\tau_2/\tau_1=\exp[(\mu_{\rm ch}-E)/k_BT]. Thus peqp_{\rm eq} can be written as,

peq(E)=τ2/(τ1+τ2)=[1+e(E−μch)/kBT]−1,(9)p_{\rm eq}(E)=\tau_2/(\tau_1+\tau_2)=[1+e^{(E-\mu_{\rm ch})/k_BT}]^{-1}, \tag*{(9)}

which is the Fermi distribution f(T,E−μch)f(T,E-\mu_{\rm ch}). Note that this is not an exact result and it will depend on the attempt rates from the two sides and on the degeneracy. In general a dopant-state occupancy is also not given by the exact F-D distribution [27] due to the spin degeneracy of the dopant level.

The interface traps’ areal charge density, dictated by traps’ occupancy, will in turn determine the filling of the channel bands or μch\mu_{\rm ch} value. We work in an independent electron approximation where the electron filling only affects μch\mu_{\rm ch} and not the individual energies including channel-bands, traps, or the barrier Δ2\Delta_2. With change in μch\mu_{\rm ch} the barrier Δ1\Delta_1 seen by the band electrons at μch\mu_{\rm ch} will change but Δ2\Delta_2 will remain the same, see Figure 6(a). Thus eliminating τ1\tau_1 in favor of τ2\tau_2, μch\mu_{\rm ch} and EE, one can write Eq. (7) and Eq. (8), respectively, as,

p˙=f(T,E−μch)−pτ2[1−f(T,E−μch)](10)\dot{p}=\frac{f(T,E-\mu_{\rm ch})-p}{\tau_2[1-f(T,E-\mu_{\rm ch})]} \tag*{(10)}

and

p(t,E)=p(0,E)e−t/τ2f(T,μch−E)+f(T,E−μch)[1−e−t/τ2f(T,μch−E)].(11)\begin{aligned} p(t,E)={}&p(0,E)e^{-t/\tau_2f(T,\mu_{\rm ch}-E)}\\ &+f(T,E-\mu_{\rm ch})[1-e^{-t/\tau_2f(T,\mu_{\rm ch}-E)}]. \tag*{(11)} \end{aligned}

Here we assumed a time independent μch\mu_{\rm ch} to solve for p(t,E)p(t,E). However, μch\mu_{\rm ch} will actually be determined by the time-dependent occupancy pp of various traps. We discuss this coupling between pp and μch\mu_{\rm ch} next.

Time dependence of channel properties

When the charge stored in interface traps changes with time, the density of mobile carriers, and thus μch\mu_{\rm ch} will also change. The carriers in the channel are assumed to equilibrate over a time much smaller than τ1,2\tau_{1,2}. We denote the areal density of states of slow traps with a transition time τ2\tau_2 in range τ\tau to τ+dτ\tau+d\tau as gstr(τ,E)dτg_{\rm str}(\tau,E)d\tau. When VgV_g is changed from zero, the displacement charge and the fast traps will react immediately leading to a new μch\mu_{\rm ch} and then the slow traps will start changing their occupancy. This will lead to a slow change in μch\mu_{\rm ch}. At an instant tt if the occupancy of the traps is p(t,τ,E)p(t,\tau,E) then the change in charge density of slow traps from the t=0t=0 equilibrium state will be given by, Δσstr(t)=e∬[1−p(t,τ,E)−1+p(0,τ,E)]gstr(τ,E)dτdE\Delta\sigma_{\rm str}(t)=e\iint[1-p(t,\tau,E)-1+p(0,\tau,E)]g_{\rm str}(\tau,E)d\tau dE. By differentiating we get,

σ˙str=−e∬p˙(t,τ,E)gstr(τ,E)dτdE.(12)\dot{\sigma}_{\rm str}=-e\iint\dot{p}(t,\tau,E)g_{\rm str}(\tau,E)d\tau dE. \tag*{(12)}

With slow traps, the σ0\sigma_0 in Eq. (5), relating instantaneous VgV_g and VchV_{\rm ch}, gets replaced by σ0+σstr\sigma_0+\sigma_{\rm str}. Differentiating this modified relation with respect to time, one gets,

V˙g=(1+γftr)V˙ch−σ˙str/Cox+γV˙ch×[2−f(T,Ev−μch0−eVch)−f(T,μch0+eVch−Ec)].(13)\dot{V}_g=(1+\gamma f_{\rm tr})\dot{V}_{\rm ch}-\dot{\sigma}_{\rm str}/C_{\rm ox}+\gamma\dot{V}_{\rm ch}\times\left[2-f(T,E_v-\mu_{\rm ch}^{0}-eV_{\rm ch})-f(T,\mu_{\rm ch}^{0}+eV_{\rm ch}-E_c)\right]. \tag*{(13)}

A simple case is the trap states existing only at certain energy E0E_0 and with characteristic time τ0\tau_0, i.e. gstr(τ,E)=nstrδ(τ−τ0)δ(E−E0)g_{\rm str}(\tau,E)=n_{\rm str}\delta(\tau-\tau_0)\delta(E-E_0) with nstrn_{\rm str} as the areal density. This leads to σ˙str=−ep˙nstr\dot{\sigma}_{\rm str}=-e\dot{p}n_{\rm str} with p˙\dot{p} dictated by Eq. (10) with E=E0−μch0E=E_0-\mu_{\rm ch}^{0} and τ2=τ0\tau_2=\tau_0. Using this in Eq. (13), we get

V˙ch[1+γftr+γ{1−f(T,Ev−μch0−eVch)+f(T,Ec−μch0−eVch)}]=V˙g−(enstrτ0Cox)f(T,E0−μch0−eVch)−p1−f(T,E0−μch0−eVch).(14)\dot{V}_{\rm ch}\left[1+\gamma f_{\rm tr}+\gamma\{1-f(T,E_v-\mu_{\rm ch}^{0}-eV_{\rm ch})+f(T,E_c-\mu_{\rm ch}^{0}-eV_{\rm ch})\}\right]=\dot{V}_g-\left(\frac{en_{\rm str}}{\tau_0C_{\rm ox}}\right)\frac{f(T,E_0-\mu_{\rm ch}^{0}-eV_{\rm ch})-p}{1-f(T,E_0-\mu_{\rm ch}^{0}-eV_{\rm ch})}. \tag*{(14)}

Let’s consider a step change in VgV_g from the Vg=0V_g=0 equilibrium state to Vg1V_{g1} at t=0t=0 and then V˙g=0\dot{V}_g=0 for t>0t>0. VchV_{\rm ch}, pp and channel conductance GG will evolve with time with the latter being directly measurable. This evolution is dictated by two coupled first order ordinary non-linear differential equations, i.e. Eqs. (10) and (14), that can be solved numerically. Figure 7 shows the time evolution at T=300T=300 K of VchV_{\rm ch}, pp and GG for traps at single energy E0=0.82E_0=0.82 eV (from the middle of the gap) and when VgV_g is changed from zero to 21.4 V. We assume μch0=0.7\mu_{\rm ch}^{0}=0.7 eV as arising from σ0/e=6.3×107 cm−2\sigma_0/e=6.3\times10^{7}\ {\rm cm}^{-2}. The other used parameters are: nstr=1.48×1011 cm−2n_{\rm str}=1.48\times10^{11}\ {\rm cm}^{-2} giving (enstr/Cox)=2(en_{\rm str}/C_{\rm ox})=2 eV and γftr=50\gamma f_{\rm tr}=50. The jump in VgV_g leads to a jump in VchV_{\rm ch} from zero to 0.2 V and then it decreases continuously to about 0.15 V as the traps’ charge increases. The channel Fermi energy μch=μch0+eVch\mu_{\rm ch}=\mu_{\rm ch}^{0}+eV_{\rm ch} thus jumps to 0.9 eV and then decreases to about 0.85 eV. This is just above the trap energy E0=0.82E_0=0.82 eV and thus leads to more than 75% filling of the traps from

Time evolution of channel voltage and trap occupancy, and conductance

Figure 7. (a) shows the time evolution of channel voltage VchV_{\mathrm{ch}} and trap occupancy pp. The solid line in (b) shows the time evolution of conductance GG for a step change in VgV_{\mathrm{g}} from zero to 21.4 V at t=0t=0 (see text for details). The dashed line in (b) depicts an exponential decay function with characteristic time of 0.04τ00.04\tau_0.

nearly zero, see the discontinuous line in Figure 7(a). As seen in this figure none of the VchV_{\rm ch}, pp or GG time-evolution can actually be described by an exponential. This is illustrated in Figure 7(b) for GG where the dashed line shows an exponential relaxation with a characteristic rate 25τ0−125\tau_0^{-1}. This rate closely matches the initial relaxation rate, i.e. τ0−1f(T,μch−E0)\tau_0^{-1}f(T,\mu_{\rm ch}-E_0) in Eq. (11), which, with μch0−E0=80\mu_{\rm ch}^{0}-E_0=80 meV, works out as 23τ0−123\tau_0^{-1} at room temperature.

Conductance hysteresis and blocking transition

On SiO2_2 the few layer MoS2_2 is experimentally observed to be n-doped with its μch\mu_{\rm ch} close to EcE_c. Thus VthV_{\rm th} for n-type conduction is usually found within Vg=±50V_g=\pm50 V and p-type conduction is not observed. In the absence of slow traps one does not expect significant hysteresis in transfer characteristics. In such a case when VgV_g is ramped forward from an extreme negative to extreme positive value the n-type conduction will start at a threshold VgV_{\mathrm{g}} value, say Vth0V_{\mathrm{th0}} and it would stop at the same value when VgV_{\mathrm{g}} is ramped back.

Figure 8(a) shows a measured conductance hysteresis loop at room temperature when VgV_{\mathrm{g}} is changed from 0 to -80 V then to +80 V and finally back to zero, all at the same rate. Figure 8(b) is the schematic of how the slow traps charge density σstr\sigma_{\mathrm{str}} and μch\mu_{\mathrm{ch}} change during this VgV_{\mathrm{g}} cycle. When VgV_{\mathrm{g}} is ramped to −V0-V_0 from zero over certain time τm\tau_{\mathrm{m}}, both the channel and slow traps accumulate positive charge leading to a rise in σstr\sigma_{\mathrm{str}} and lowering of μch\mu_{\mathrm{ch}} relative to EcE_{\mathrm{c}}. Now when VgV_{\mathrm{g}} is ramped forward towards +V0+V_0, the displacement charge in the channel changes fast while σstr\sigma_{\mathrm{str}} turns around slowly. This leads to a sharp rise in μch\mu_{\mathrm{ch}} which crosses μth\mu_{\mathrm{th}} at some threshold Vg=Vthf<Vth0V_{\mathrm{g}}=V_{\mathrm{thf}}<V_{\mathrm{th0}} when σstr=σf\sigma_{\mathrm{str}}=\sigma_{\mathrm{f}}. By this point only some traps change back their charge state to negative and the remaining still contribute mobile electrons in the channel together with those due to rising VgV_{\mathrm{g}}. At Vg=+V0V_{\mathrm{g}}=+V_0 the traps accumulate a negative charge density and some of it, say σb\sigma_{\mathrm{b}}, will still remain when VgV_{\mathrm{g}} turns around to reach Vthb>Vth0V_{\mathrm{thb}}>V_{\mathrm{th0}} at which the conduction stops.

Measured conductance hysteresis loop and schematic of slow-trap charge density, chemical potential, and gate voltage

Figure 8. (a) shows the measured gate dependent drain current for a few layer MoS2_2 at Vds=1V_{\mathrm{ds}}=1 V as a function of VgV_{\mathrm{g}} and over a VgV_{\mathrm{g}} cycle from 0 to -80 V, then to +80 and back to zero. (b) shows the schematic changes in slow-trap charge density σstr\sigma_{\mathrm{str}} (solid red line), μch\mu_{\mathrm{ch}} (solid blue line) over a cyclic change in VgV_{\mathrm{g}} (solid black line) between ±V0\pm V_0. The discontinuous red line depicts σstr\sigma_{\mathrm{str}} at initial and final zero VgV_{\mathrm{g}} and the discontinuous blue line shows the EcE_{\mathrm{c}} relative to which μch\mu_{\mathrm{ch}} changes. The discontinuous horizontal black line just below EcE_{\mathrm{c}} shows the μth\mu_{\mathrm{th}} at which the channel starts conducting.

At the conduction threshold, the channel carrier density, the chemical potential, and thus the quantum capacitance as well as the charge stored in fast traps will

be same and independent of the VgV_{\mathrm{g}} history. Therefore, the difference in charge density at the gate electrode for the two threshold VgV_{\mathrm{g}} values will be equal to the difference in the charge density of the traps, i.e. σf−σb=Cox(Vthb−Vthf)\sigma_{\mathrm{f}}-\sigma_{\mathrm{b}}=C_{\mathrm{ox}}(V_{\mathrm{thb}}-V_{\mathrm{thf}}) or Δσstr=CoxΔVth\Delta\sigma_{\mathrm{str}}=C_{\mathrm{ox}}\Delta V_{\mathrm{th}}. Our objective here is to understand and model the temperature dependence of ΔVth\Delta V_{\mathrm{th}} which is proportional to Δσstr\Delta\sigma_{\mathrm{str}}. The physics of this is similar to the super-paramagnetic hysteresis, which is briefly discussed in the Appendix with the parameters that are relevant for traps. In high electron mobility transistors based on semiconductor heterojunction the hysteresis, similar to superparamagnets, is found only at low temperatures [28], presumably due to a better coupling of the traps to the channel.

When one ramps VgV_{\mathrm{g}} to +V0+V_0, a given trap’s occupancy will change according to Eq. (7) with a time dependent μch=μch0+eVch\mu_{\mathrm{ch}}=\mu_{\mathrm{ch}}^{0}+eV_{\mathrm{ch}} dictated by Eqs. (12) and (13). This makes the occupancy of different traps coupled and rather complex. ΔVth\Delta V_{\mathrm{th}} is dictated by the difference in occupancy of the traps at the two threshold voltages. In the absence of the detailed knowledge about the traps’ distribution, i.e. gstr(τ,E)g_{\mathrm{str}}(\tau,E), and associated barriers we make certain simplifying assumptions. The magnitude of hysteresis and large SS value imply that the overall, fast and slow, trap density is large. We assume that this makes the overall change in μch\mu_{\mathrm{ch}} much smaller than the energy gap EgE_{\mathrm{g}} as well as the energy barrier Δ2\Delta_{2}. This also implies that only the traps in a narrow energy range, as compared to EgE_{\mathrm{g}} and Δ2\Delta_{2}, change their state over experimental VgV_{\mathrm{g}} sweep range.

For the temperature dependence ΔVth\Delta V_{\mathrm{th}}, the exact details of VgV_{\mathrm{g}} cycle will not make a significant difference as long as the overall time scale of the cycle is the same. We thus consider a case where VgV_{\mathrm{g}} is first kept at zero for long enough time to achieve an equilibrium occupancy of traps and then it is abruptly changed to −V0-V_{0} and held at this value for time τm\tau_{\mathrm{m}} and then it is ramped to zero over time τm\tau_{\mathrm{m}}. In this way μch\mu_{\mathrm{ch}} will first decrease to certain lowest value μch−\mu_{\mathrm{ch}-} and then rise passing through μth\mu_{\mathrm{th}} at certain Vg=VthfV_{\mathrm{g}}=V_{\mathrm{thf}} where the channel starts conducting. We consider a similar excursion from Vg=0V_{\mathrm{g}}=0 equilibrium state, where the channel is insulating, to +V0+V_{0} where it is kept for τm\tau_{\mathrm{m}}. The channel will start conducting during this τm\tau_{\mathrm{m}} when μch\mu_{\mathrm{ch}} rises above μth\mu_{\mathrm{th}} and reaches a maximum value μch+\mu_{\mathrm{ch}+}. When VgV_{\mathrm{g}} is ramped back to zero over time τm\tau_{\mathrm{m}}, μch\mu_{\mathrm{ch}} will now go below μth\mu_{\mathrm{th}} at certain Vg=VthbV_{\mathrm{g}}=V_{\mathrm{thb}} where the channel stops conducting.

We write from Eq. (11) for the difference in occupancy of a trap at energy EE, i.e. Δp=p+−p−\Delta p=p_{+}-p_{-}, corresponding to the two extreme gate voltages ±V0\pm V_{0} as,

Δp=f(T,E−μch+)[1−exp⁡(−τmτ2f(T,μch+−E))]−f(T,E−μch−)[1−exp⁡(−τmτ2f(T,μch−−E))].(15)\begin{aligned} \Delta p={}&f(T,E-\mu_{\mathrm{ch}+})\left[1-\exp\left(-\frac{\tau_{\mathrm{m}}}{\tau_{2}f(T,\mu_{\mathrm{ch}+}-E)}\right)\right]\\ &-f(T,E-\mu_{\mathrm{ch}-})\left[1-\exp\left(-\frac{\tau_{\mathrm{m}}}{\tau_{2}f(T,\mu_{\mathrm{ch}-}-E)}\right)\right]. \tag*{(15)} \end{aligned}

Here, we have assumed identical initial occupancy of the traps before VgV_{\mathrm{g}} is brought to the two extreme values. Note also that τ2=τaexp⁡(Δ2/kBT)\tau_{2}=\tau_{\mathrm{a}}\exp(\Delta_{2}/k_{\mathrm{B}}T) is a steep function of temperature TT in the range of interest.

As discussed earlier, the traps that actually change their charge state in response to VgV_{\mathrm{g}} will be in a narrow energy range. We believe that the potential barrier for an interface trap to change its charge state is actually much higher, i.e. Δ1,Δ2≫kBT,∣Δ1−Δ2∣\Delta_{1},\Delta_{2}\gg k_{\mathrm{B}}T,\lvert\Delta_{1}-\Delta_{2}\rvert. So the temperature dependence of τ2f(T,μch−E)\tau_{2}f(T,\mu_{\mathrm{ch}}-E) is primarily dictated by Δ2\Delta_{2}. In fact, the distribution in Δ2\Delta_{2} or τ2\tau_{2} as compared to that in EE of relevant slow traps dominates the behavior. Thus we absorb f(T,μch−E)f(T,\mu_{\mathrm{ch}}-E) in τ2\tau_{2} for the purpose of variation with temperature and write,

Δp(τm)=[f(T,E−μch+)−f(T,E−μch−)][1−exp⁡(−τm/τaexp⁡(Δ2/kBT))].(16)\Delta p(\tau_{\mathrm{m}})=\left[f(T,E-\mu_{\mathrm{ch}+})-f(T,E-\mu_{\mathrm{ch}-})\right]\left[1-\exp\left(-\frac{\tau_{\mathrm{m}}/\tau_{\mathrm{a}}}{\exp(\Delta_{2}/k_{\mathrm{B}}T)}\right)\right]. \tag*{(16)}

The traps occupancy that actually dictates σstr\sigma_{\mathrm{str}} is when VgV_{\mathrm{g}} is ramped back towards zero over time ∼τm\sim\tau_{\mathrm{m}} from the two extremes such that μch=μth\mu_{\mathrm{ch}}=\mu_{\mathrm{th}}, see Figure 8. This will change Δp\Delta p by a factor ∼exp⁡(−τm/τaexp⁡(Δ2/kBT))\sim\exp\left(-\frac{\tau_{\mathrm{m}}/\tau_{\mathrm{a}}}{\exp(\Delta_{2}/k_{\mathrm{B}}T)}\right). This is similar to the super-paramagnets discussed in the Appendix. Different traps may have different energy barriers and with the fact that only traps in a narrow energy range near μch\mu_{\mathrm{ch}} change their charge state, the distribution of Δ2\Delta_{2} will dominate the temperature dependence of VthV_{\mathrm{th}}. Combining the weekly temperature dependent prefactor [f(T,E−μch+)−f(T,E−μch−)]\left[f(T,E-\mu_{\mathrm{ch}+})-f(T,E-\mu_{\mathrm{ch}-})\right] with the slow trap’s energy barrier distribution function nstr(Δ2)n_{\mathrm{str}}(\Delta_{2}), we conclude

ΔVth∝∫nstr(Δ2)[1−exp⁡(−τm/τaexp⁡(Δ2/kBT))]×exp⁡(−τm/τaexp⁡(Δ2/kBT)) dΔ2.(17)\Delta V_{\mathrm{th}}\propto\int n_{\mathrm{str}}(\Delta_{2})\left[1-\exp\left(-\frac{\tau_{\mathrm{m}}/\tau_{\mathrm{a}}}{\exp(\Delta_{2}/k_{\mathrm{B}}T)}\right)\right]\times\exp\left(-\frac{\tau_{\mathrm{m}}/\tau_{\mathrm{a}}}{\exp(\Delta_{2}/k_{\mathrm{B}}T)}\right)\,\mathrm{d}\Delta_{2}. \tag*{(17)}

In case of the same barrier value Δ2\Delta_{2} for all traps we expect a peak in ΔVth\Delta V_{\mathrm{th}} with at a blocking temperature TB=Δ2/[kBln⁡(τm/τa)]T_{\mathrm{B}}=\Delta_{2}/[k_{\mathrm{B}}\ln(\tau_{\mathrm{m}}/\tau_{\mathrm{a}})]. A distribution around a mean Δ2\Delta_{2} will increase the width of this peak. Another unknown parameter here is τm/τa\tau_{\mathrm{m}}/\tau_{\mathrm{a}}, i.e. the ratio of measurement time, or VgV_{\mathrm{g}} sweep time, to attempt rate.

Experiments on hysteresis, blocking and gate-cooling

In this section we discuss experimental measurements focusing on the slow traps in an FET device with a few-layer MoS2_{2} on SiO2_{2}. This helps us understand the energy and barrier distribution associated with these traps. The observed temperature dependence of hysteresis, quantified by ΔVth\Delta V_{\mathrm{th}}, is presented next together with the blocking model discussed earlier. Finally, the reversible handle on VthV_{\mathrm{th}} through blocking of the traps in desired charge state by cooling under different gate voltages is discussed.

Hysteresis and time dependence at room temperature

The transfer characteristics, shown in Figure 8(a), of a few layer MoS2_{2} FET at room temperature exhibit a large hysteresis. The on-state high conductance at Vg=+80V_{\mathrm{g}}=+80 V due to n-doping can be attributed to the electron rich sulfur vacancies and other n-type impurities present in natural MoS2_{2} crystals [29]. This also leads to the pinning of EcE_{\mathrm{c}} of MoS2_2 close to the Fermi energy of contact-metal (gold) and thus negligible electron Schottky barrier at the MoS2_2-metal contacts [30, 31]. The blue line in Figure 8(a) marks the subthreshold region for backward VgV_{\mathrm{g}} sweep. The sub-threshold swing (SS) from this line works out as 3 V/dec as opposed to 0.06 V/dec, i.e. the value expected for no traps, see Eq. (6). This measured SS gives γftr≈50\gamma_{\mathrm{ftr}} \approx50 and gftr=3.7×1012g_{\mathrm{ftr}} = 3.7 \times10^{12} eV−1^{-1}cm−2^{-2}.

Figure 9(a) shows the measured Ids−VgI_{\mathrm{ds}} - V_{\mathrm{g}} curves for different VgV_{\mathrm{g}} sweep ranges varying from ±10\pm10 (ΔVg=20\Delta V_{\mathrm{g}} = 20) V, to ±90\pm90 V (ΔVg=180\Delta V_{\mathrm{g}} = 180 V). There is negligible hysteresis for ±10\pm10 V sweep range as VthV_{\mathrm{th}} values for both the sweep directions are well within this sweep-range and nearly equal. With increasing sweep range, VthfV_{\mathrm{thf}} reduces and VthbV_{\mathrm{thb}} increases leading to a monotonic rise in ΔVth\Delta V_{\mathrm{th}}, see Figure 9(b). Thus the slow traps, responsible for hysteresis, are nearly uniformly distributed over the μch\mu_{\mathrm{ch}} range accessible up to the largest VgV_{\mathrm{g}} sweep range. From ΔVth\Delta V_{\mathrm{th}} we can find the areal density of slow traps responsible for hysteresis for a given ΔVg\Delta V_{\mathrm{g}} by using CoxΔVth/eC_{\mathrm{ox}}\Delta V_{\mathrm{th}}/e with Cox/e=7.6×1010C_{\mathrm{ox}}/e = 7.6 \times10^{10} cm−2^{-2}V−1^{-1}. Typical resulting values of areal density of slow traps ∼1012\sim10^{12} cm−2^{-2} are smaller than the usual three-dimensional (3D) semiconductors and similar to other 2D materials like graphene [32, 33].

Ids versus Vg curves and threshold-voltage variation

Figure 9. (a) IdsI_{\mathrm{ds}} Vs VgV_{\mathrm{g}} at Vds=1V_{\mathrm{ds}} = 1 V for different sweep ranges of VgV_{\mathrm{g}} from ±10\pm10 to ±90\pm90 V. All these curves are acquired at the same VgV_{\mathrm{g}} sweep rate. The inset shows the zoomed-in portion for ±10\pm10 and ±20\pm20 V range VgV_{\mathrm{g}} sweeps. (b) Variation of VthfV_{\mathrm{thf}}, VthbV_{\mathrm{thb}} and ΔVth\Delta V_{\mathrm{th}} with sweep range ΔVg\Delta V_{\mathrm{g}} as extracted from (a).

A careful look at Figure 9(b) shows an asymmetry between VthfV_{\mathrm{thf}} and VthbV_{\mathrm{thb}} with the former changing more with ΔVg\Delta V_{\mathrm{g}} than later. This can be expected even for uniform distribution of traps as the magnitude of change in μch\mu_{\mathrm{ch}} for positive VgV_{\mathrm{g}} is less than that for negative VgV_{\mathrm{g}}. This is due to the rapid increase in channel’s quantum capacitance when μch\mu_{\mathrm{ch}} approaches EcE_{\mathrm{c}}. This will amount to activation of traps in narrower energy range for same the magnitude positive VgV_{\mathrm{g}} than negative. A continuous rise in the rate at which VthbV_{\mathrm{thb}} changes with ΔVg\Delta V_{\mathrm{g}} and up to the highest ΔVg\Delta V_{\mathrm{g}} implies an increase in slow traps’ DOS near EcE_{\mathrm{c}}. Also towards large ΔVg\Delta V_{\mathrm{g}} values VthfV_{\mathrm{thf}} seems to saturate indicating a reduction in slow trap’s density of states when μch\mu_{\mathrm{ch}} moves away from EcE_{\mathrm{c}} and into the gap. From the monotonic rise in ΔVth\Delta V_{\mathrm{th}} with ΔVg\Delta V_{\mathrm{g}} we conclude that the slow traps are somewhat uniformly distributed. Although from the details of the VthfV_{\mathrm{thf}} and VthbV_{\mathrm{thb}} variation the traps seem to be concentrated over a limited energy range close to EcE_{\mathrm{c}}.

Figure 10(a) shows a measured time dependent IdsI_{\mathrm{ds}} as a function of time when VgV_{\mathrm{g}} is abruptly changed from −80-80 to +80+80 V. There is a fast initial relaxation followed by a slow stretched exponential tail indicating multiple time scales. This relaxation would be rather complex to fit to a microscopic model, as discussed earlier, in the absence of the knowledge about the distribution of trap energies and activation barriers. A fitting with multiexponential or stretched exponential does work and it has indeed been used [15] to conclude a distribution in barrier energies. However, due to the coupling between the dynamics of different trap’s occupancy and μch\mu_{\mathrm{ch}}, even traps at single energy and with the same barrier can lead to non-exponential relaxation, with a long tail that can resemble a stretched-exponential, see Figure 7(b).

Time dependence and gate-voltage curves at different sweep rates

Figure 10. (a) The measured time dependence of IdsI_{\mathrm{ds}} when VgV_{\mathrm{g}} is changed abruptly from -80 to +80 V. The inset shows the zoomed-in initial part of the relaxation. (b) IdsI_{\mathrm{ds}} vs Gate voltage curves for different sweep rates of back gate voltage at a fixed Vds=1V_{\mathrm{ds}} = 1 V. The solid squares in the inset show ΔVth\Delta V_{\mathrm{th}} as a function of overall sweep time with the red line showing a fit to a sum of two exponential relaxations.

As a consequence of this slow relaxation of traps, the hysteresis has a significant dependence on the VgV_{\mathrm{g}} sweep rate for a fixed sweep range. Figure 10(b) shows the conductance hysteresis loops acquired at different sweep rates from 0.26 to 24.6 V/s. A high sweep rate also gives higher peak conductance as less number of traps acquire negative charge leading to more electrons in the channel. In fact, for some of the very fast sweep rates, a saturation or a downturn in channel conductivity is seen with VgV_{\mathrm{g}} due to a delayed response of the traps which depletes electrons from the channel. As discussed earlier, the rate of filling of an empty trap state at a given energy will increase with VgV_{\mathrm{g}} as μch\mu_{\mathrm{ch}} rises with VgV_{\mathrm{g}}. Figure 9(b) inset shows the variation of ΔVth\Delta V_{\mathrm{th}} as a function of the VgV_{\mathrm{g}} sweep time. It fits well to a double exponential function, ΔVth=α−βe−r1Δt−γe−r2Δt\Delta V_{\mathrm{th}} = \alpha- \beta e^{-r_1\Delta t} - \gamma e^{-r_2\Delta t} with r1−1=35r_1^{-1} = 35 s, r2−1=292.5r_2^{-1} = 292.5 s and α\alpha, β\beta and γ\gamma as constants.

Blocking transition of interface traps

Figure 11(a) shows IdsI_{\mathrm{ds}}–VgV_{\mathrm{g}} curves at several temperatures between 300 and 80 K over ±80\pm80 V VgV_{\mathrm{g}} sweep range and 2.6 V/s sweep-rate. For these measurements, the device was first kept at room temperature at Vg=0V_{\mathrm{g}} = 0 for 2-3 hours in order to equilibrate the traps and then cooled and stabilized at each different temperature keeping Vg=0V_{\mathrm{g}} = 0. The hysteresis can be seen to reduce with cooling though the rate of reduction is not monotonic as seen in Figure 11(b). ΔVth\Delta V_{\mathrm{th}} reduces slowly near room temperature and the rate of reduction, i.e. dΔVth/dTd\Delta V_{\mathrm{th}}/dT, peaks near 225 K and then the rate as well as ΔVth\Delta V_{\mathrm{th}} diminish as 80 K temperature is approached.

Temperature dependence of log($I_{\mathrm{ds}}$) versus $V_{\mathrm{g}}$ curves and $\Delta V_{\mathrm{th}}$ as a function of temperature

Figure 11. (a) Temperature dependence of log⁡(Ids)\log(I_{\mathrm{ds}}) Vs VgV_{\mathrm{g}} curves at Vds=1V_{\mathrm{ds}} = 1 V between 80 and 300 K. (b) The solid circles show ΔVth\Delta V_{\mathrm{th}} as a function of temperature. The solid line shows the calculated variation of ΔVth\Delta V_{\mathrm{th}} using Eq. (17) with the barrier distribution function n(Δ2)n(\Delta_2) depicted by the solid line in the inset. This n(Δ2)n(\Delta_2) is the sum of three Gaussian distributions shown by discontinuous lines in the inset. The traps with Δ2\Delta_2 beyond 8000 K do not contribute to ΔVth\Delta V_{\mathrm{th}} variation over the studied temperature range.

When compared to superparamagnets, as discussed in the Appendix with parameters relevant to traps, one expects to see a peak in ΔVth\Delta V_{\mathrm{th}}. In case of actual superparamagnets, where the attempt rate as well as barrier height are much smaller, the temperature dependence of M-H curves shows an opposite behavior as hysteresis disappears with increasing temperature. In that case, due to small barrier height the barrier can be made to vanish at accessible magnetic fields. However, in case of traps in MoS2_2 devices the barrier is large and the accessible VgV_{\mathrm{g}} range only permits a small μch\mu_{\mathrm{ch}} variation which is insufficient to make the barrier vanish. Thus, at low temperatures, the traps do not change their state and one does not see hysteresis. Further, the hysteresis does not vanish at high temperatures, in case of traps, as there is a distribution in barrier height Δ2\Delta_2 which may continue to very high values for some of the traps.

The continuous line in Figure 11(b) shows the temperature dependence of ΔVth\Delta V_{\mathrm{th}} found using Eq. (17) and a Δ2\Delta_2 distribution depicted in the inset. Here we have used a fixed τm/τ0=1013\tau_m/\tau_0 = 10^{13} though a change in this value by up to even two orders of magnitude only slightly affects the required n(Δ2)n(\Delta_2) for fitting the measured ΔVth(T)\Delta V_{\mathrm{th}}(T). Traps with Δ2\Delta_2 higher than 8000 K do not contribute to the hysteresis at temperatures 300 K or below. One may see a decline in ΔVth\Delta V_{\mathrm{th}} at further higher temperatures, however, we find that the IdsI_{\mathrm{ds}}–VgV_{\mathrm{g}} curves do not exhibit so sharp transitions at VthfV_{\mathrm{thf}} and VthbV_{\mathrm{thb}}. This could be from activation of larger number of traps and some of the slow traps may turn into fast ones at higher temperatures. Other extrinsic effects, such as traps’ diffusion, may also come into play. Eventually, very high VgV_{\mathrm{g}} needed to access VthfV_{\mathrm{thf}} and VthbV_{\mathrm{thb}}, and particularly at high temperatures, also leads to the breakdown of the dielectric oxide and permanent device damage.

Gate cooling and reversible control of VthV_{\mathrm{th}}

Figure 12(a) shows IdsI_{\mathrm{ds}} Vs VgV_{\mathrm{g}} measured at 80 K temperature after cooling the device from 350 K temperature to 80 K in presence of different gate voltages, labeled as VgcV_{\mathrm{gc}}, between -80 and 90 V. The device was first warmed to 350K in vacuum and kept at desired VgcV_{\mathrm{gc}} for an hour before cooling it down to 80K. As expected there is negligible hysteresis at 80 K but more striking is the reversible change in VthV_{\mathrm{th}} over a wide range from -40 to +40 V. At negative VgcV_{\mathrm{gc}} the traps get blocked in a positively charged state. This trap charge electron dopes the channel and thus a negative VgV_{\mathrm{g}} is needed to deplete it. Similarly a positive VgcV_{\mathrm{gc}} leads to traps blocked with negative charge that depletes the electrons from the channel and thus a positive VgV_{\mathrm{g}} is needed to make it conduct. In this way the traps act as a controllable virtual gate.

Effect of cooling the device and variation of threshold voltage

Figure 12. (a) Effect of cooling the device from 350 K to 80 K under different applied gate voltages VgcV_{\mathrm{gc}} from -80V to 90V. All the curves measured at 80 K and for Vds=1V_{\mathrm{ds}} = 1 V show negligible hysteresis. (b) Variation of VthV_{\mathrm{th}} with VgcV_{\mathrm{gc}}. The axis labels on the right in (b) show the corresponding blocked slow trap density.

Figure 12(b) shows the variation of VthV_{\mathrm{th}} with VgcV_{\mathrm{gc}}. The VthV_{\mathrm{th}} value at 80 V can be converted into an appropriate charge density σstr\sigma_{\mathrm{str}} associated with the blocked slow-traps. The axis labels on right shows this σstr/e=CoxVth\sigma_{\mathrm{str}}/e = C_{\mathrm{ox}}V_{\mathrm{th}}. Another fact from this figure is the nearly linear relation between VthV_{\mathrm{th}} and VgcV_{\mathrm{gc}} with a slope close to 1/2. This indicates that about half of the charge induced by VgcV_{\mathrm{gc}} gets stored in the blocked slow-traps while the remaining half is taken up by the fast traps and channel carriers. This is striking as the change in μch\mu_{\mathrm{ch}} (or VchV_{\mathrm{ch}}) with VgV_{\mathrm{g}} near the conduction threshold is quite non-linear, see Figure 5.

Discussion and conclusions

When the thermally grown SiO2_2 surface is stored in ambient air, the surface siloxanes terminated edge on the substrate react with water and gradually revert to Si-OH, after which the substrate becomes rehydrated which can act as electron trap center [34, 35]. Further, a monolayer or submonolayer of hydrogen-bonded water stays on SiO2_2 and cannot be removed by pumping in a vacuum even over long periods. This is also a possible source of interface traps. There can also be traps or dopants within MoS2_2 channel that can arise from donor-like S mono-vacancies or more complex defects involving S vacancies. The slow traps having barriers of large heights, and presumably large widths, permit electron exchange only through thermal activation rather than through quantum tunneling. Other extrinsic effects where the change

in interface charge happens through electrochemical reactions at the interface involving species of hydrogen and oxygen. This may bring the diffusion barriers for these species into the picture that can also influence the time scale of charge transfer.

Transport measurements are sensitive to the traps’ distribution only over a narrow energy range near the conduction threshold. With a significant electron doping in MoS2_2 on SiO2_2 our results are consistent with a distribution of traps near EcE_{\mathrm{c}}. The inability to access hole-doped conduction even till Vg=−100V_{\mathrm{g}} = -100 V may indicate significant traps’ density, both slow and fast, if one goes only by channel conduction. However, the contacts may also play a role and other investigations are required to conclude on this aspect. In fact, our attempts to access the hole doped regime by combining the gate-cooling at Vgc=90V_{\mathrm{gc}} = 90 V and large negative VgV_{\mathrm{g}} till -100 V at 80 K temperature also did not succeed.

In conclusion, a temperature dependent study of few-layer MoS2_2 FET transfer characteristics shows hysteresis with a large difference ΔVth\Delta V_{\mathrm{th}} between the backward- and forward-sweep threshold gate-voltages. This is modeled using the hysteresis in interface trap charge density. The model also describes the complex coupled dynamics of channel carrier density and traps’ charge density and thus even the traps with single energy and barrier can lead to non-exponential relaxations. The observed temperature dependence of ΔVth\Delta V_{\mathrm{th}} is attributed to the blocking of traps and fitted to a distribution of energy barriers for charge exchange between the traps and the channel. Finally, the blocking helps to get nearly non-hysteretic behavior at 80 K temperature with a voltage threshold programmable by gate-cooling voltage.

Acknowledgements

Authors acknowledge discussions on blocking transition with Ranjit Thapa and funding from SERB-DST of the Government of India.

Blocking in Superparamagnets

A superparamagnet consists of non-interacting nanosized ferromagnetic crystals in single domain limit with their magnetic reversal described by the Stoner-Wohlfarth model [36]. Such crystals with uniaxial anisotropy exhibit two opposite magnetic moment ±ms\pm m_{\mathrm{s}} states of equal energy, in the absence of external field, that are separated by an anisotropy energy-barrier, say Δ0\Delta_{0}. The external magnetic field BB along +z+z-direction makes spin-up state favorable to spin-down state by energy 2msB2m_{\mathrm{s}}B. Thus the barrier seen by spin-up state is increased to Δ0+msB\Delta_{0}+m_{\mathrm{s}}B and for the spin-down state it reduces to Δ0−msB\Delta_{0}-m_{\mathrm{s}}B. We further assume the same time scale τ0\tau_{0} of magnetic dynamics in the two minima and the dominance of thermal activation over quantum tunneling. The rate of transition at temperature TT from spin-up to spin-down state will be given by τ↑↓−1=τ0−1exp⁡[−(Δ0−mB)/kBT]\tau_{\uparrow\downarrow}^{-1}=\tau_{0}^{-1}\exp[-(\Delta_{0}-mB)/k_{\mathrm{B}}T] and the reverse transition rate will be τ↓↑−1=τ0−1exp⁡[−(Δ0+mB)/kBT]\tau_{\downarrow\uparrow}^{-1}=\tau_{0}^{-1}\exp[-(\Delta_{0}+mB)/k_{\mathrm{B}}T]. The average magnetic moment is given by m=⟨m⟩=(p↑−p↓)msm=\langle m\rangle=(p_{\uparrow}-p_{\downarrow})m_{\mathrm{s}} with p↑p_{\uparrow} and p↓p_{\downarrow} as the probability of being in the respective spin-state. With p↑+p↓=1p_{\uparrow}+p_{\downarrow}=1 we get ⟨m⟩=(2p↑−1)ms\langle m\rangle=(2p_{\uparrow}-1)m_{\mathrm{s}}. The time dependence of p↑p_{\uparrow} is dictated by,

dp↑dt=−τ↑↓−1p↑+τ↓↑−1p↓=−(τ↑↓−1+τ↓↑−1)p↑+τ↓↑−1.\frac{dp_{\uparrow}}{dt}=-\tau_{\uparrow\downarrow}^{-1}p_{\uparrow}+\tau_{\downarrow\uparrow}^{-1}p_{\downarrow}=-(\tau_{\uparrow\downarrow}^{-1}+\tau_{\downarrow\uparrow}^{-1})p_{\uparrow}+\tau_{\downarrow\uparrow}^{-1}.

This leads to the equation of motion for mm as,

dmdt=−(τ↑↓−1+τ↓↑−1)m+(τ↓↑−1−τ↑↓−1).\frac{dm}{dt}=-(\tau_{\uparrow\downarrow}^{-1}+\tau_{\downarrow\uparrow}^{-1})m+(\tau_{\downarrow\uparrow}^{-1}-\tau_{\uparrow\downarrow}^{-1}).

Substituting for τ↑↓−1\tau_{\uparrow\downarrow}^{-1} and τ↓↑−1\tau_{\downarrow\uparrow}^{-1}, we get

dmdt=2τ0−1exp⁡(−Δ0kBT)cosh⁡(msBkBT)[tanh⁡(msBkBT)−⟨m⟩].(18)\frac{dm}{dt}=2\tau_{0}^{-1}\exp\left(-\frac{\Delta_{0}}{k_{\mathrm{B}}T}\right)\cosh\left(\frac{m_{\mathrm{s}}B}{k_{\mathrm{B}}T}\right)\left[\tanh\left(\frac{m_{\mathrm{s}}B}{k_{\mathrm{B}}T}\right)-\langle m\rangle\right]. \tag*{(18)}

Thus at equilibrium, i.e. when dm/dt=0dm/dt=0, we get meq=tanh⁡(msB/kBT)m_{\mathrm{eq}}=\tanh(m_{\mathrm{s}}B/k_{\mathrm{B}}T) expected for this two state system. However, the rate at which the equilibrium is attained is dictated by 2τ0−1exp⁡(−Δ0/kBT)cosh⁡(msB/kBT)2\tau_{0}^{-1}\exp(-\Delta_{0}/k_{\mathrm{B}}T)\cosh(m_{\mathrm{s}}B/k_{\mathrm{B}}T), which for ∣msB∣≫kBT\lvert m_{\mathrm{s}}B\rvert\gg k_{\mathrm{B}}T will become τ0−1exp⁡[−(Δ0±msB)/kBT]\tau_{0}^{-1}\exp[-(\Delta_{0}\pm m_{\mathrm{s}}B)/k_{\mathrm{B}}T]. Typical τ0\tau_{0} values for magnetic systems are of ∼10−9\sim10^{-9} s order and thus within the measurement time scale τm∼1\tau_{\mathrm{m}}\sim1 s, the equilibrium is attained either for ms∣B∣≳Δ0m_{\mathrm{s}}\lvert B\rvert\gtrsim\Delta_{0} or for T≳TB=Δ0/[kBln⁡(τm/τ0)]T\gtrsim T_{\mathrm{B}}=\Delta_{0}/[k_{\mathrm{B}}\ln(\tau_{\mathrm{m}}/\tau_{0})]. The former corresponds to the vanishing of the barrier between two states due to applied field and the latter defines the blocking temperature TBT_{\mathrm{B}} with hysteresis for T<TBT<T_{\mathrm{B}} and no hysteresis for T>TBT>T_{\mathrm{B}}.

In order to model the temperature dependence of hysteresis with parameters relevant to the charge traps, we consider a superparamagnetic-like system but with a large τ0−1\tau_{0}^{-1} and a large barrier Δ0\Delta_{0} such that Δ0≫msB≫kBT\Delta_{0}\gg m_{\mathrm{s}}B\gg k_{\mathrm{B}}T. Therefore, the barrier does not vanish at any typical applied BB and in fact the barrier always dominates the energetics. We can solve (18) numerically for a time-dependent BB changed at a constant rate and in a cycle. Starting from t=0t=0, where B=0B=0 and μ=0\mu=0, BB is ramped up to +B0+B_{0} over time τm/2\tau_{\mathrm{m}}/2. It is then ramped down and to −B0-B_{0} and then again to +B0+B_{0}, all at the same rate. We assume Δ0/kB=6500\Delta_{0}/k_{\mathrm{B}}=6500 K, msB0/kB=500m_{\mathrm{s}}B_{0}/k_{\mathrm{B}}=500 K and τm/τ0=1013\tau_m/\tau_0 = 10^{13}. This leads to mm Vs BB as plotted in Figure 13. At low temperatures (say, 185 K) we see that both the response of mm to the magnetic field and the hysteresis are negligible. As temperature rises, the response and hysteresis both increase but at high temperatures (say 250 K), the response is large but hysteresis vanishes. We use the difference Δm\Delta m in mm values at B=0B = 0 during reverse and forward field change of the same cycle as a measure of the hysteresis. As seen in the inset of Figure 13, this Δm\Delta m exhibits a peak at the blocking temperature TB=Δ0/[kBln⁡(τm/τ0)]=217T_B = \Delta_0/[k_B \ln(\tau_m/\tau_0)] = 217 K.

Calculated average magnetic moment versus applied magnetic field at different temperatures

Figure 13. Calculated average magnetic moment (mm) Vs applied field (BB) at different temperatures. The field is swept in a cycle between ±500kB/ms\pm500k_{\mathrm{B}}/m_{\mathrm{s}} at constant rate with total time 2τm2\tau_{\mathrm{m}} (see text for details). The dots in the inset show the variation of the difference in two mm values at zero field, i.e. Δm\Delta m, with temperature illustrating how the hysteresis peaks near 217 K. The continuous line in inset is the plot following from Eq. 19 with a=0.76a=0.76 and b=0.26b=0.26.

The solid line in Figure 13 inset, given by

Δmms=2[1−exp⁡{−τm2τ0exp⁡(−Δ0−amsB0kBT)}]×exp⁡{−τm2τ0exp⁡(−Δ0+bmsB0kBT)},(19)\frac{\Delta m}{m_s} = 2\left[1-\exp\left\{-\frac{\tau_m}{2\tau_0}\exp\left(-\frac{\Delta_0-a m_s B_0}{k_B T}\right)\right\}\right]\times\exp\left\{-\frac{\tau_m}{2\tau_0}\exp\left(-\frac{\Delta_0+b m_s B_0}{k_B T}\right)\right\}, \tag*{(19)}

follows from an analytical model behavior with aa and bb as constants between zero and one. The forward (higher to lower energy state) relaxation rate at field B0B_0 is τf−1=τ0−1exp⁡[−(Δ0−msB0)/kBT]\tau_f^{-1} = \tau_0^{-1}\exp[-(\Delta_0-m_sB_0)/k_B T] while the reverse rate is τr−1=τ0−1exp⁡[−(Δ0+msB0)/kBT]\tau_r^{-1} = \tau_0^{-1}\exp[-(\Delta_0+m_sB_0)/k_B T]. The zero field rate will be τz=τ0−1exp⁡[−(Δ0)/kBT]\tau_z = \tau_0^{-1}\exp[-(\Delta_0)/k_B T]. As per Eq. (18), when BB is changed abruptly to B0B_0 from zero, mm will change from zero to m1=tanh⁡(msB0/kBT)[1−exp⁡(−τm/2τf)]m_1 = \tanh(m_sB_0/k_B T)[1-\exp(-\tau_m/2\tau_f)] in time τm/2\tau_m/2. We can take tanh⁡(msB0/kBT)=1\tanh(m_sB_0/k_B T) = 1 for large B0B_0. Now if BB is abruptly made zero mm will become m2=m1exp⁡(−τm/2τz)m_2 = m_1\exp(-\tau_m/2\tau_z) after time τm/2\tau_m/2. For continuous ramp the actual relaxation rate will be in between τf−1\tau_f^{-1} and τz−1\tau_z^{-1} for forward ramp and in between τz−1\tau_z^{-1} and τr−1\tau_r^{-1} for return ramp leading to 0<a,b<10 < a, b < 1. A similar excursion in BB from zero to −B0-B_0 will lead to −m2-m_2 and thus Δm=2m2\Delta m = 2m_2.

References

  1. [1]K. F. Mak, C. Lee, J. Hone, J. Shan, and T. F. Heinz, “Atomically thin MoS₂: a new direct-gap semiconductor”, Phys. Rev. Lett. 105, 136805 (2010).DOI
  2. [2]Z. Zhou, Zhixian, and Y. K. Yap, “Two-dimensional electronics and optoelectronics: present and future Electronics”, Electronics 6, 53 (2017).
  3. [3]B. Radisavljevic, A. Radenovic, J. Brivio, V. Giacometti, and A. Kis, “Single-Layer MoS₂ Transistors”, Nat. Nanotechnol. 6, 147 (2011).DOI
  4. [4]Y. Yoon, K. Ganapathi, and S. Salahuddin, “How Good Can Monolayer MoS₂ Transistors Be?”, Nano Lett. 11, 3768 (2011).DOI
  5. [5]D. Krasnozhon, D. Lembke, C. Nyffeler, Y. Leblebici, and A. Kis, “MoS₂ Transistors Operating at Gigahertz Frequencies”, Nano Lett. 14(10), 5905 (2014).DOI
  6. [6]L. M. Martinez, N. J. Pinto, C. H. Nayor, and A. T. C. Johnson, “MoS₂. based dual input logic AND gate”, AIP Adv. 6, 125041 (2016).DOI
  7. [7]S. Wachter, D. K. Polyushkin, O. Bethge, T. Mueller, “A microprocessor based on a two-dimensional semiconductor”, Nat. Commun. 8, 14948 (2017).DOI
  8. [8]B. Radisavljevic, M. B. Whitwick, and A. Kis, “Integrated Circuits and Logic Operations Based on Single-Layer MoS₂”, ACS Nano 5(12), 9934 (2011).DOI
  9. [9]O. Lopez-Sanchez,D. Lembke, M. Kayci, A. Radenovic, and A. Kis, “Ultrasensitive photodetectors based on monolayer MoS₂”, Nat. Nanotechnol. 8, 497 (2013).DOI
  10. [10]K. F. Mak, K. He, J. Shan, T. F. Heinz, “Control of valley polarization in monolayer MoS₂ by optical helicity”, Nat. Nanotechnol. 7, 494 (2012).DOI
  11. [11]O. Lopez-Sanchez, E. A. Lladó, V. Koman, A. F. Morral, A. Radenovic, and A. Kis, “Light generation and harvesting in a van der Waals heterostructure”, ACS Nano 8, 3042 (2014).
  12. [12]M.-L. Tsai, S.-H. Su, J.-K. Chang, D.-S. Tsai, C.-H. Chen, C.-I Wu, L.-J. Li, L.-J. Chen, and J.-H. He, “Mono-layer MoS₂ heterojunction solar cells”, ACS Nano 8(8), 8317 (2014).
  13. [13]J. I. Gittleman, B. Abeles, and S. Bozowski, “Superparamagnetism and relaxation effects in granular Ni-SiO₂ and Ni-Al₂O₃ films”, Phys. Rev. B 9, 3891 (1974).
  14. [14]A. K. Singh and A. K. Gupta, “Reversible control of doping in graphene-on-SiO₂ by cooling under gate-voltage”, J. Appl. Phys. 122, 195305 (2017).DOI
  15. [15]D. J. Late, B. Liu, H. S. S. Ramakrishna Matte, V. P. Dravid, and C. N. R. Rao, “Hysteresis in Single-Layer MoS₂ Field Effect Transistors”, ACS Nano 6(6), 5635 (2012).DOI
  16. [16]T. Li, G. Du, B. Zhang, and Z. Zeng, “Scaling behavior of hysteresis in multilayer MoS₂ field effect transistors”, Appl. Phys. Lett. 105, 093107 (2014).
  17. [17]Y. Guo, X. Wei, J. Shu, B. Liu, J. Yin, C. Guan, Y. Han, S. Gao, and Q. Chen, “Charge trapping at the MoS₂-SiO₂ interface and its effects on the characteristics of MoS₂ metal-oxide semiconductor field effect transistors”, Appl. Phys. Lett. 106, 103109 (2015).DOI
  18. [18]Y. Park, H. W. Baac, J. Heo, and G. Yoo, “Thermally activated trap charges responsible for hysteresis in multilayer MoS₂ field-effect transistors”, Appl. Phys. Lett. 108, 083102 (2016).DOI
  19. [19]A. Castellanos-Gomez, M. Buscema, R. Molenaar, V. Singh, L. Janssen, H. S. J. van der Zant, and G. A. Steele, “Deterministic transfer of two-dimensional materials by all-dry viscoelastic stamping”, 2D Mater.1, 011002 (2014).arxiv.org/abs/1311.4829
  20. [20]M. A. Meitl, Z. T. Zhu, V. Kumar, K. J. Lee, X. Feng, Y. Y. Huang, I. Adesida, R. G. Nuzzo, and J. A. Rogers, “Transfer printing by kinetic control of adhesion to an elastomeric stamp”, Nature Materials 5, 33 (2005).
  21. [21]C. Lee, H. Yan, L. E. Brus, T. F. Heinz, J. Hone, and S. Ryu, “Anomalous Lattice Vibrations of Single and Few-Layer MoS₂”, ACS Nano 4(5), 2695 (2010).DOI
  22. [22]A. Castellanos-Gomez, N. Agraït, and G. Rubio-Bollinger , “Optical identification of atomically thin dichalcogenide crystals”, Appl. Phys. Lett. 96, 213116 (2010).DOI
  23. [23]A. Allain, J. Kang, K. Banerjee, A. Kis, “Electrical contacts to two-dimensional semiconductors”, Nature Mater 14, 1195 (2015).DOI
  24. [24]S. Das, H.-Y. Chen, A. V. Penumatcha, and J. Appenzeller, “High Performance Multilayer MoS₂ Transistors with Scandium Contacts”, Nano Lett. 13(1), 100 (2013).DOI
  25. [25]N. Ma and D. Jena, “Carrier statistics and quantum capacitance effects on mobility extraction in 2D crystal semiconductor FETs”, 2D Mater. 2, 015003 (2015).
  26. [26]In the schematics in Fig. 4, the band-system moves up or down depending on its potential and filling while μch remains fixed at zero similar to Fig.2. The interface traps being in close vicinity of the channel will experience the same potential and thus the trap energy levels will move up or down with the bands. The channel bands will actually shift down (up) with increasing negative (positive) charge in the bands.
  27. [27]See Ch-28 of N. W. Ashcroft and N. D. Mermin, “Solid State Physics”, Saunders College, Philadelphia (1976).
  28. [28]P. Kushwaha, S. Sinha, C. K. Karmakar, M. Sahu, R. K. Kaneriya, P. P. Kumar, and A. Bhattacharya, “Characterization of GaN HEMT at Cryogenic Temperatures”, IEEE MTT-S Int. Microwave and RF Conf. (IMaRC), 1, doi: 10.1109/IMaRC49196.2021.9714643 (2021).DOI
  29. [29]H.-P. Komsa and A. V. Krasheninnikov, “Native defects in bulk and mono-layer MoS₂ from first principles”, Phys. Rev. B 91, 125304 (2015).
  30. [30]H. Liu, A. T. Neal, and P. D. Ye, “Channel length scaling of MoS₂ MOSFETs”, ACS Nano 6(10), 8563 (2012).DOI
  31. [31]D. Liu, Y. Guo, L. Fang, and J. Robertson, “Sulfur vacancies in mono-layer MoS₂ and its electrical contacts”, Appl. Phys. Lett. 103, 183113 (2013).
  32. [32]P. Fiorenza, F. Giannazzo, S. Cascino, M. Saggio, and F. Roccaforte, “Identification of two trapping mechanisms responsible of the threshold voltage variation in SiO₂/4H-SiC MOSFETs”, Appl. Phys. Lett. 117, 103502 (2020).DOI
  33. [33]H. Wang, Y. Wu, C. Cong, J. Shang, and T. Yu, “Hysteresis of electronic transport in graphene transistors”, ACS Nano. 4(12), 7221 (2010).DOI
  34. [34]See page 62 of R. K. Iler, The Chemistry of Silica (Wiley-Interscience, New York, 1979).
  35. [35]K. Nagashio, T. Yamashita, T. Nishimura, K. Kita, and A. Toriumi, “Electrical transport properties of graphene on SiO₂ with specific surface structures”, J. Appl. Phys. 110, 024513 (2011).DOI
  36. [36]E. C. Stoner and E. P. Wohlfarth., “A mechanism of magnetic hysteresis in heterogeneous alloys”, Phil. Trans. Roy. Soc. of London, Series A, Mathematical and Physical Sciences, 240(826), 599 (1948).

Paper details

vlnc:W_xg7j8ptj
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