Influence of water vapor on the electronic property of MoS₂ field effect transistors

Molybdenum disulfide (MoS₂) has attracted much attention as a novel two-dimensional (2D) material for electronics [1–4], optoelectronics [5–8], valleytronics [9–14], transparent and flexible devices [15–18], and sensors [19, 20]. One limitation of MoS₂-based devices is their high sensitivity to extrinsic effects, such as environmental gases, which cause the instability of the device properties [19–22]. The influence of water vapor on MoS₂ is not yet clear. Some works suggest water vapor is an electron acceptor [23], which increases the resistance of the MoS₂ device. While some other evidence show water vapor causes hysteresis behavior in field effect transistors (FETs) [24], but the depletion of the electron is not so obvious [21, 24]. On the other hand, the desorption of water molecules from MoS₂ is difficult compared with other gases [21, 25]. Furthermore, the mechanics of the above phenomena has not been studied well. Besides the channel materials, substrate is an important factor for the electronic performance of 2D materials devices. It has been confirmed that water adsorption on SiO₂ substrate has a negative influence on the performance of graphene devices [26, 27]. However, similar study on MoS₂ devices is rare.

In the present study, we investigate the effect of water on the electronic properties of MoS₂ FETs and the way to remove the water adsorption and recover the device property.

The exfoliated MoS₂ sheets were transferred onto 300 nm-thick SiO₂ covering a highly doped p⁺⁺ Si wafer as a back gate. Two FETs, supported and suspended, were fabricated on the same piece of MoS₂ with the same channel length and width. The source and drain electrodes were patterned using electron beam lithography (EBL), followed by the deposition of Cr/Au 5 nm/70 nm using electron beam evaporation (EBE). Then, a window was defined by EBL on the channel of one of the two neighboring devices on the same piece of MoS₂, and the SiO₂ beneath the channel was etched away with buffered HF solution. After the cleaning process, the sample was moved out from hot isopropanol with low surface tension to make the MoS₂ channel suspended. A schematic diagram and a scanning electron microscope (SEM) image of a couple of FETs are shown in figure 1. The optical images and Raman spectrum of the MoS₂ are given in online supplementary information. The electronic characteristics of the devices were measured using a semiconductor characterization system (Keithley, 4200) in a probe station (Janis, ST500) that allows for controlling the environment in the sample chamber and in situ annealing. We fabricated several devices through the same process and obtained similar results. The water we used is ultrapure water with its resistivity being 18 MΩ · cm, and the water vapor pressure is 2.33 KPa after water vapor is introduced into the sample chamber.

Zoom In Zoom Out Reset image size

First, the transfer curves and output curves of the FETs based on a 3 layer MoS₂ (The optical images and Raman spectrum of the MoS₂ are given in online supplementary information) were measured in the vacuum with the pressure of 3.3 × 10⁻³ Pa, the results are shown in figure 1 as black lines. The threshold voltage of the suspended FET is negative, while that of the supported FET is near 0 V, indicating that the SiO₂ substrate can deplete electrons from the MoS₂ channel and decrease the carrier concentration. The subthreshold swing (SS) of the supported device is worse than that of the suspended device due to the impurities between the substrate and MoS₂ sheets [28, 29].

After introducing water vapor into the sample chamber, the transfer curves and output curves of the FETs were measured again, the results are shown in figure 2 as red lines. The suspended and supported devices display a similar response to water vapor, the on-state current drops to about 40% of the original value, but the threshold voltage and SS do not change obviously. Water adsorption on SiO₂ has been reported to have a strong effect on nanomaterials devices, such as graphene [26, 27]. However, here, the suspended and supported MoS₂ devices fabricated on the same piece of MoS₂ have the same response to water. Therefore, it should be the water adsorption on the MoS₂ itself that affects the electronic properties of the FETs, the effect of SiO₂ is not important. The fact that SS does not change before and after water introduction indicates water does not introduce interface trapping which can change the gate efficiency. The threshold voltage does not change obviously, together with the unchanged SS, indicates the carrier density does not change obviously before and after introducing water vapor. Therefore, the decrease of the on-state current is mainly due to the decrease of the transconductance and mobility.

Zoom In Zoom Out Reset image size

In order to remove the water adsorption, we pumped the sample chamber for an hour. When the pressure in the chamber was around 3.3 × 10⁻³ Pa, we measured the electronic characteristics again. As shown by the blue lines in figure 2, the electronic characteristics after pumping are about the same as that in water vapor for both the supported and suspended FETs. This result indicates pumping in the vacuum cannot remove the water adsorption.

Then we heated the sample to 500 K in the vacuum trying to remove water adsorption by heat. After annealing for about 1 h, the sample was cooled down to room temperature in the vacuum, and we measured the electronic property of the devices again. As shown in figure 2 as green lines, the electronic characteristics after annealing are about the same as that in water vapor for both the supported and suspended FETs. This result indicates thermal annealing in vacuum cannot remove the water adsorption on the MoS₂, either.

In order to confirm the above results, we repeated the experiments again on the same devices. This time, we pumped the sample in the vacuum (3.3 × 10⁻³ Pa) for 12 h. After the long pumping time, the transfer curves still have very little change in the linear coordinates, while in the logarithmic coordinates, a threshold shift to the negative side can be observed in both the suspended and supported FETs, as shown in figure 3. This threshold shift can be attributed to the desorption of oxygen induced by long time keeping in a high vacuum, as has been reported previously [30]. Besides the threshold shift, no other difference is observed from the electronic properties after the two pumping processes. These results confirm that pumping in vacuum cannot remove the adsorbed water on MoS₂.

Zoom In Zoom Out Reset image size

After the long pumping time, we introduced water vapor into the sample chamber again and measured the electronic characteristics. As shown in figure 3, the transfer curves obtained in water vapor for the first and the second times are about the same, which means the effect of water gas adsorption has been maintained from the first time to the second time. This can be explained as followed: the adsorption positions have been occupied in the first time water vapor introducing, not much more water molecules can be absorbed to the MoS₂ surface further in the second time, so that the electronic properties remain roughly the same.

The sample chamber was then pumped back to the high vacuum and the devices were annealed at 500 K again. The electronic characteristics measured after the pumping and annealing process are shown in figure 3 as green lines, the results are similar to that shown in figure 2. The above observation proves that the water molecules adsorbed on the MoS₂ sheets are difficult to remove simply by pumping in vacuum and/or thermal annealing in vacuum.

We would like to emphasize here that the response to the water vapor and the different treatments performed here are the same for both the suspended FETs and the supported FETs, so that the SiO₂ substrate and its interface with the MoS₂ channel do not play an important role on the water sensing property of MoS₂ devices. As mentioned above, the on-state current decrease in water vapor is mainly due to the decrease of the transconductance and mobility. Since the same samples are compared before and after water adsorption, the channel length L, the channel width W, the capacity of the insulator C_i and the source-drain voltage V_DS are fixed, the variation of transconductance can be fully explained by the variation of carrier mobility. The carrier mobility can be calculated by $\mu =\displaystyle \frac{{\rm{d}}{I}_{DS}\,}{{\rm{d}}{V}_{g}}\displaystyle \frac{L}{W{C}_{i}{V}_{DS}},$ where W = 2 μm and L = 1 μm in the present devices, V_DS = 1 V and C_i is the capacity per unit area of 300 nm thick SiO₂/vacuum insulate layer. The mobility of the suspended devices is found to drop from 18.2 cm² (V · s)⁻¹ to 10.6 cm² (V · s)⁻¹ after water adsorption and the mobility of the supported devices drops from 17.1 cm² (V · s)⁻¹ to 7.3 cm² (V · s)⁻¹ after water adsorption.

The carrier mobility obtained above is field effect mobility, which is also affected by the contact resistance. To evaluate the possible influence of the contact, we measure the contact resistance of the devices with the transfer length method. Supported devices with different channel length (0.5 μm, 1 μm, 1.5 μm) were fabricated on the same piece of a 4 layer MoS₂ sheet (The Raman spectrum and the optical images are given in the online supplementary information), as shown in figure 4. The source and drain metal and their fabrication method are the same as the above suspended and supported FETs. As shown in figure 5, the devices with difference channel length have the same threshold voltage before and after water adsorption. The contact resistance is the half of the y intercept of the resistance versus channel length plots at V_g = 20 V (at the on-state). The contact resistance is 8.8 KΩ in the vacuum and 12.0 KΩ in water vapor, while the total resistance is 118 KΩ and 297 KΩ for the 0.5 μm device, as shown in figure 5(c). The total resistance of the devices with longer channels are higher than that in the 0.5 μm device. To take the contact resistance into account, the current should be $I=V/({R}_{c}+{R}_{ch}),$ where R_c is the total contact resistance and R_ch is the resistance of the channel. Therefore, the modified mobility should be multiplied by a coefficient $1+{R}_{c}/{R}_{ch}.$ Due to the small contact resistance relative to the channel resistance, the modified mobility is around 22 cm²/(V · s) before water adsorption. Note, even the modified mobility is still lower than the high value in the literature [31]. The decrease of field effect mobility after water adsorption indicates the decrease of carrier mobility in the channel. As the suspended devices have similar responses as the supported devices to water vapor and the different treatments, the above statement is also true for the suspended devices. The mechanism should be that the adsorbed water molecules form new scattering centers on the MoS₂ surface, which reduces the carrier mobility. The slightly increased mobility of the suspended devices can be explained by the removal of the substrate. We observed many Coulomb impurities in the interface between SiO₂ substrate and MoS₂, which induce relative low mobility [32]. Because the fabrication processes of the present devices include wet etching, impurities and defects could exist after etching, a very clean and perfect surface is difficult to achieve. This could explain why the mobility in the suspended devices before water adsorption is still not the highest compared with that in the literature.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Then, is there any way to remove the water adsorption and recover the electronic property of MoS₂ FETs? We have found a method here. As shown in figure 6, the transfer curve returns to the initial state after a large current passing through the device. We fix the gate voltage at 30 V to set the device to the on-state, and we increase the drain voltage gradually, as shown in figure 6(c). First, we set the drain voltage V_DS to change from 3 V to 6 V, and then back to 3 V, and measure the curve I_DS1 shown in figure 6(c). Next, we set V_DS to change from −4 V to −7 V, and then back to −4 V, and obtain the curve I_DS2. The curves, I_DS3, I_DS4 and I_DS5, are obtained when V_DS changes at higher levels step by step. In this process, the current in the MoS₂ channel increases step by step. The water adsorption can be removed when a large current passes through the MoS₂ channel for both the supported and the suspended devices. As described in detail in our recent work [33], we can estimate the temperature distribution for V_DS = 5 V and I_DS = 100 μA through Comsol by taking into account all the factors, including the Joule heat, the thermal conductivity and the thermal capacity of the MoS₂, SiO₂, Si substrate and the metal electrodes, etc. We find the temperature increased by Joule heating at the present current is less than 447 K for the supported devices, which is less than the temperature in the heat annealing process. Therefore, the increase of the on-state current cannot be attribute to heat annealing.

Zoom In Zoom Out Reset image size

The adsorption energy of water on MoS₂ has been found to be around 0.2 ∼ 0.3 eV according to previous computation results [34, 35]. The phonon energy is around 0.026 eV at room temperature, and around 0.05 eV at 500 K (the present thermal annealing temperature), these energies are much less than the adsorption energy. Therefore, it is understandable that pumping in the vacuum and thermal annealing cannot remove the water adsorption from MoS₂. Why can current annealing remove the water adsorption and recover the property of MoS₂ FETs? We propose the following explanation. The contact between Cr/Au and MoS₂ has a Schottky barrier. At large V_DS, the electrons inject into the conduction band of the MoS₂ channel should experience a strong band bending at the MoS₂-metal contact, which could generate excitons via impact excitation [36]. The energy distribution of those hot carriers no longer obeys Boltzman distribution under a high electric field [37]. The hot carriers have been found to induce gas desorption in semiconductors [38, 39] and metals [40]. As the source-drain bias we applied is similar to that in [36], the hot electron energy in this work should be similar to that in [36], where hot carrier associated excitons have been demonstrated. The energy of the hot carriers (electrons) is sufficient to generate excitons and is larger than the desorption energy of water from MoS₂, so that can cause water desorption.

The effects of water on MoS₂ FETs and their recovery are studied. The suspended and supported MoS₂ devices fabricated on the same piece of MoS₂ show similar response to water and to different treatments (such as pumping in vacuum, thermal annealing and current annealing), indicating the effects of the SiO₂ substrate and its interface with MoS₂ are negligable. Water adsorption is found to decrease the mobility of MoS₂ probably through introducing a scattering center on the surface. Water vapor is found not to introduce doping or traps obviously, demonstrated by the unchanged threshold voltage and SS after water adsorption. Long time pumping in the vacuum and thermal annealing in the vacuum cannot desorb the water molecules. We find current annealing at high source-drain bias can effectively remove the water adsorption and recover the electronic property of the MoS₂ FETs, and we propose the mechanism is through the high energy hot carriers under large bias.

We thank Yuxiang Han, Zhiqiang Tang, Mei Sun and Xing Li for valuable discussions. This work was supported by the NSF of China (No. 11374022, 11528407 and 61621061).