Semiconductor channel mediated photodoping in h-BN encapsulated monolayer MoSe2 phototransistors

In optically excited two-dimensional phototransistors, charge transport is often affected by photodoping effects. Recently, it was shown that such effects are especially strong and persistent for graphene/h-BN heterostructures, and that they can be used to controllably tune the charge neutrality point of graphene. In this work we investigate how this technique can be extended to h BN encapsulated monolayer MoSe_2 phototransistors at room temperature. By exposing the sample to 785 nm laser excitation we can controllably increase the charge carrier density of the MoSe_2 channel by {\Delta}n {\approx} 4.45 {\times} 10^{12} cm^{-2}, equivalent to applying a back gate voltage of 60 V. We also evaluate the efficiency of photodoping at different illumination wavelengths, finding that it is strongly correlated with the light absorption by the MoSe_2 layer, and maximizes for excitation on-resonance with the A exciton absorption. This indicates that the photodoping process involves optical absorption by the MoSe_2 channel, in contrast with the mechanism earlier described for graphene/h-BN heterostroctures.


Introduction
Two-dimensional (2D) transition metal dichalcogenides (TMDs) are very attractive materials for the design of optoelectronic devices at the nanoscale [1][2][3][4][5] due to their optical bandgap spanning the visible spectrum, large photoresponse, and high carrier mobility. The most simple and popular device geometry for 2D TMD phototransistors consists of a monolayer crystal transferred onto a SiO2/Si substrate, with metallic contacts built directly on top of the TMD crystal surface. However, recent works showed that encapsulation of the 2D semiconductor channel between hexagonal boron nitride (h-BN) layers largely improves the electrical performance, optoelectronic response, and device stability [6,7] , as it allows to prevent channel degradation due to electrostatic interactions with the metallic electrodes and the SiO2 substrate. [8][9][10] Thus, h-BN encapsulation is rapidly settling as a new standard for high-quality 2D optoelectronics.
In 2D TMDs the optoelectronic response is often caused by two dominant coexisting effects: [11][12][13][14][15] photoconductivity, where light-induced formation of electron-hole pairs (or charged excitons) leads to an increased charge carrier density and electrical conductivity without changing the Fermi energy EF of the 2D channel; and photodoping, where the light-induced filling or depleting of charge traps and gap states (present in the surrounding materials and interfaces) causes a shift in EF. [16][17][18][19] Since part of the charge-states of traps have very long lifetimes, photodoping typically occurs at longer time scales than photoconductivity.
A recent work showed that, for graphene transistors, using an h-BN substrate instead of the usual SiO2 leads to a large enhancement of photodoping effects [20] due to an exchange of charge carriers between graphene and boron nitride, allowing to optically tune the charge carrier density of graphene. [20,21] Further, in a recent experiment, the photodoping effect was used to tune the Fermi energy in encapsulated MoS2 nanoconstrictions at cryogenic temperatures [19] , showing that this encapsulated monolayer MoSe2 phototransistors" (2D Materials, 2019 with photon energies above the absorption edge of MoSe2, a large, long-lasting, photodoping effect appears, allowing to increase the MoSe2 charge carrier density by Δn ≈ 4.5 × 10 12 cm -2 (observed here as a -60 V shift of the threshold gate voltage). This effect is especially strong when the device is exposed to light while a negative gate voltage is applied. After turning off the excitation, the device remains photodoped for several days. By testing the dependence of photodoping on the excitation energy, we find that this effect only occurs for excitation wavelengths above the absorption edge of 1L-MoSe2, indicating that the photodoping effect is mediated by optical excitation of this material, in contrast with earlier theoretical descriptions for graphene/h-BN structures [20] , where photodoping was attributed to the optical excitation of h-BN impurity states [22][23][24][25][26][27][28] . Our results show that long-lasting controllable photodoping can be achieved, even at room  temperature, for 2D TMD phototransistors with h-BN substrates, and shed light on the mechanism responsible for this effect. Figure 1a shows a sketch of the studied monolayer MoSe2 phototransistor, where the semiconductor channel is fully encapsulated between bilayer h-BN and bulk h-BN. We used a dry, adhesive-free pick up technique [29] to fabricate the h-BN/MoSe2/h-BN heterostructure on a SiO2 (285 nm)/p-doped Si substrate. Then, we fabricated Ti (5nm)/Au (75 nm) electrodes on top of the structure by e-beam evaporation (see Methods for details). The 5 nm Ti layer allows to achieve a close match between the metal work-function (4.33 eV) and the electron affinity of a 1L-MoSe2 [30] . The thickness of the different layers was characterized by AFM. Figure 1b shows the twoterminal I-V characteristic of the 1L-MoSe2 channel at four different gate voltages Vg, applied at the bottom Si layer (see Figure 1a), ranging from +30 V to +60 V. The I-Vs show a non-Ohmic behavior due to the presence of tunnel barriers at the contacts. For a detailed study of the electrical behavior of channel and contacts for this device geometry we address the reader to ref. [7] .

Photodoping and transfer I-V characteristics.
We now investigate the effect of illumination on the transfer characteristic of the 1L-MoSe2 channel. We found that the following procedure is suitable for characterizing the occurrence and persistence of photodoping effect: We ramp the gate voltage from Vg = +70 V to -70 V (trace) and then back to +70 V (retrace) at a ramping speed of 1 V/s while keeping a constant drain-source voltage Vds = 0.5 V and measuring the drain-source current Ids. This measurement is first carried out in dark and then repeated upon illumination, as described below. The black curve in Figure 2a shows a transfer characteristic measured while keeping the 1L-MoSe2 device unexposed to light. While ramping Vg from +70 V to -70 V only a small, constant increase of the drain source current is observed with respect to the transfer characteristic measured in dark, which we attribute to photoconductivity (see inset in Figure 2). Note that for pure photodoping an increase of the off-state current is not expected. When Vg is ramped back from -70 V to 0 V, however, we observe a large shift of the transfer curve towards negative gate voltages due to photodoping. A third transfer curve acquired in dark after exposure to light confirms that the shift persists when the illumination is removed, independently of the Vg ramping direction. As further discussed below, we attribute this shift to a light-induced electron migration from h-BN donor localized states to the MoSe2 valence band.
Next, we characterize the stability of the observed photodoping. Figure 2b shows the time (t) evolution of the drain-source current Ids in our device for Vds = 0.5 V and Vg = 0 V after encapsulated monolayer MoSe2 phototransistors" (2D Materials, 2019). DOI: 10.1088/2053-1583/ab0c2d. 6 photodoping. The sample is first exposed to illumination at λ = 785 nm and Vg = -70 V for 48 hours and then the light source is turned off and Vg is brought back to 0 V immediately before the measurement starts. As shown in the Figure, Ids decreases over time due to the slow increase of the threshold gate voltage Vth (estimated from Ids as discussed below) as photodoping fades away. The time evolution of Ids can be well described by a double exponential function plus an offset:  The parameters p1…p5 are obtained by least square fitting to the experimental data, which yields p1 = 7.48 nA, p2 = 0.02 h -1 , p3 = 3.08 nA, p4 = 0.16 h -1 and p5 = 1.10 nA. The double exponential decay profile of Ids indicates that at least two separate relaxation mechanisms, dominant at different time scales, are involved in this process. We remark that, even at room temperature, the photodoping effect persists for remarkably long times, and even 40 hours after photodoping we get Vth = -3 V, shifted by 28 V below its value prior to light exposure.  Similarly, we find that Vth can also be increased by exposing the device to light at Vg = +70 V. This process allows to recover the original value of Vth, prior to photodoping after few hours of exposure (shown in Suppl. Info. S1) and even allows to shift the Vth slightly above this value. The maximum positive shift (ΔVth = + 8 V), corresponding to the purple curve in Fig. 2c, was obtained after 15 hours of exposure at Vg = +70 V.
It is worth noting that, apart from the shift of Vth, the overall shape of the transfer curves remains  Next, to investigate the spectral response of the observed photodoping we measure transfer characteristics while illuminating the device for a range of excitation wavelengths, from 850 nm to 765 nm (Figure 3a). Before each measurement we keep the system in dark at Vg = +70 V until the same initial threshold voltage Vth ≈ 25 V is reached. Then, we ramp the threshold voltage from +70 V to -70 V at a ramping speed of -1 V/s. When the gate voltage is brought to Vg = -70 V while exposing the device to light, the threshold gate voltage Vth is lowered due to photodoping (as described earlier and shown in Figure 2a). Figure 3b shows the shift of Vth observed between the trace (non-photodoped) and retrace (photodoped) transfer curves as a function of the wavelength, and Figure 3c shows a photocurrent spectrum measured in the same device for comparison (see Methods section and ref. [12] for details). Remarkably, the photodoping-induced shift is strongest for illumination at λ = 795 nm (Figure 3b), closely matching the A exciton resonance of 1L-MoSe2 In Figure 4, we present an alternative mechanism that allows to account for the observed spectral response. For simplicity, we only consider the bottom h-BN layer, but the same description can be   Figure 2b, if the system is kept in dark after photodoping, the h-BN will slowly recover its charge neutrality, as the localized states get filled by charge carriers from the MoSe2 conduction band (indicated by the grey arrow in Figure 4d). However, as experimentally observed, this process is much slower that the electron migration from the localized states to the MoSe2 valence band, yielding a persistent photodoping. The difference in characteristic times for depletion and filling of localized states suggests that these states are more strongly coupled with the MoSe2 valence band than with its conduction band, but the reason for this remains unclear to us. We remark that, as mentioned above, the proposed description is still valid if the h-BN layer is placed on top of the 1L-MoSe2, as in this case a nonzero electric field will still appear in the SiO2 layer as consequence of the light-induced charge accumulation at localized states near the MoSe2/h-BN interface.
It is worth noting that a photodoping mechanism similar to the one here described can also appear for SiO2/TMD structures [31] . However, the reported characteristic times for depletion of impurity states in these structures are typically several orders of magnitude lower than those observed here for h-BN substrates. In consequence, photodoping for 2D phototransistors on SiO2 substrates does not yield a persistent shift of Vth and manifests instead as a sublinear excitation power dependence of photoconductivity [32] . In our device, we do not expect a measurable contribution to photodoping from SiO2, as this would require the photoexcited charge carriers to migrate from the MoSe2 layer to the SiO2, physically separated by the 7.5 nm thick h-BN layer.
Importantly, in light of the model discussed above, a similar electron migration from h-BN to MoSe2 could also appear in absence of light if some states of the valence band are depleted by applying a sufficiently large negative gate voltage. However, in our devices we do not observe conduction through the valence band (i.e. the gate transfer characteristics do not present ambipolar behavior), indicating that even for the largest applied negative gate Vg = -70 V, the density of depleted states in the valence band is negligible.

Discussion and final remarks
In all, we demonstrated that photodoping can be used for controllably and persistently tuning the Fermi energy in h-BN encapsulated 1L-MoSe2 phototransistors at room temperature, allowing to tune the carrier density by Δn = 4.5 × 10 12 cm -2 . The photoinduced shift of Vth was observed up to a few days after exposure to light. The measured spectral response of this effect revealed that photodoping only appears for wavelengths above the absorption edge of MoSe2, clearly indicating that this effect is mediated by optical excitation of the 1L-MoSe2 channel followed by charge migration from h-BN to the channel. Thus, the efficiency of photodoping maximizes for excitation wavelengths at which the 2D channel is highly absorbing.
For the 1L-MoSe2 region directly below the electrodes the optical absorption is expected to be highly suppressed due to screening of the light electric field by the metallic layer. However, should photodoping still occur at these regions, it would produce an increased built-in voltage [7] for the contact, modifying the band alignment between the metallic electrode and the semiconductor channel. This would be observed as an additional contribution to the shift of Vth similar to that of the photodoping mechanism discussed above. Note that for this case the migration of electrons from h-BN to the MoSe2 channel is still needed.

13
It is worth noting that the mechanism proposed here does not require to make any assumption on the nature of the electron-donor localized states, which could be attributed to interfacial states, structural defects or impurities present at the h-BN [22][23][24][25][26][27][28]33] .
In all, our results demonstrate that the use of h-BN substrates for enhancing the photodoping effect can be easily extended to several two-dimensional semiconductors beyond graphene, and that photodoping is expected to appear for any wavelength at which a significant photoconductivity can occur at the 2D channel.   Figure S1 shows the time evolution of Ids and Vth under illumination at λ = 785 nm, Vg = 70 V and Vds = 0.5 V. This process allows to accelerate the time for recovery of Vth until it reaches its original value, prior to photodoping (Vth = 25 V). However, it only allows to increase Vth photodoping over this value to a small extent.

S2. Role of excitons in photoconductivity of MoSe2 phototransistors
As discussed in the main text, photoconductivity relies on the generation of optically excited charge carriers. However, for exciton absorption, photoexcited electrons and holes combine to form excitons with zero net charge. Therefore, in order to contribute to the measured charge current, neutral excitons need to dissociate into free electrons and holes.
It has been proposed that, in monolayer TMD phototransistors, neutral A 0 excitons can dissociate due to the large electric gradients formed near the electrodes, giving a nonzero contribution to 20 photoconductivity [S1] . Furthermore, it was shown in prior literature that for TMD/h-BN devices the optical absorption spectrum is mainly dominated by trions, rather than excitons [S2] . Since trions carry a nonzero charge, they can contribute to photoconductivity even without dissociation.
Further discussion regarding photoconductivity in 1L-MoSe2 phototransistors can be found in our earlier work [S3].