The Integration of Sub-10 nm Gate Oxide on MoS2 with Ultra Low Leakage and Enhanced Mobility

The integration of ultra-thin gate oxide, especially at sub-10 nm region, is one of the principle problems in MoS2 based transistors. In this work, we demonstrate sub-10 nm uniform deposition of Al2O3 on MoS2 basal plane by applying ultra-low energy remote oxygen plasma pretreatment prior to atomic layer deposition. It is demonstrated that oxygen species in ultra-low energy plasma are physically adsorbed on MoS2 surfaces without making the flakes oxidized, and is capable of benefiting the mobility of MoS2 flake. Based on this method, top-gated MoS2 transistor with ultrathin Al2O3 dielectric is fabricated. With 6.6 nm Al2O3 as gate dielectric, the device shows gate leakage about 0.1 pA/μm2 at 4.5 MV/cm which is much lower than previous reports. Besides, the top-gated device shows great on/off ratio of over 108, subthreshold swing (SS) of 101 mV/dec and a mobility of 28 cm2/Vs. With further investigations and careful optimizations, this method can play an important role in future nanoelectronics.

MoS 2 transistors with sub-10 nm top gate dielectrics are seldom reported 1,8,[12][13][14][15][16][17][18] . For example, in recent reports, the top gate dielectrics of MoS 2 transistors are 50 nm Al 2 O 3 16 and 30 nm HfO 2 18 for the work by Pezeshki A. et al. and Krasnozhon D. et al., respectively. Only by realizing high-quality pinhole-free and thin dielectrics over large area on MoS 2 can the continual scaling down of MoS 2 FETs be possible. With the shrink of dielectric thickness, especially at sub-10 nm region, the gate capacitance would be greatly improved, leading to better control of the channel and larger drive current. Some methods have been proposed to achieve uniform growth of high-κ materials on MoS 2 , such as an ultrathin metal oxide buffer layer, organic functionalization of MoS 2 and ultraviolet-ozone exposure [20][21][22] . But most of the work just stopped at the early stage of realizing uniform growth without exploring the impacts of surface functionalization on devices performance, especially on gate leakage.
In this work, a CMOS process compatible method to achieve uniform Al 2 O 3 growth on MoS 2 basal plane by applying a remote O 2 plasma treatment prior to Al 2 O 3 growth is proposed, and top-gated MoS 2 MOSFET with ultrathin Al 2 O 3 dielectric deposited using this method is also studied. Notably, the Al 2 O 3 dielectric layer is about 6.6 nm, which is the thinnest top gate dielectric ever reported, but exhibits the impressive leakage current about 0.1 pA/μ m 2 at 4.5 MV/cm. This leakage is even much smaller than that of MoS 2 transistors capped with much thicker top gate dielectrics 1,8,13,16,17 . At the same time, the top-gated device also shows great on/off ratio of over 10 8 , subthreshold swing (SS) of 101 mV/dec and a mobility of 28 cm 2 /Vs. In addition, mechanism investigations show that after the pretreatment, oxygen atoms are physically adsorbed on the MoS 2 surface without oxidizing it. This non-destructive physical adsorption mechanism is revealed by the advanced ultra-high-vacuum (UHV) in-situ analysis system. We believe it will benefit the two-dimensional electronic devices research a lot.

Results
Leakage current of gate oxide results in high power consumption and performance degradation of the two dimensional layered transistors 20 . To achieve uniform ALD Al 2 O 3 growth on pristine MoS 2 , functionalization of the MoS 2 surface is required to introduce uniform surface groups that serve as active nucleation sites for the ALD process 21,23 . Initially, an investigation on few-layer MoS 2 flakes was carried out with a plasma enhanced ALD system. For the sample in Fig. 1a, 120 cycles Al 2 O 3 was directly deposited on MoS 2 surface at 200 °C using TMA (Trimethyl Aluminum) and H 2 O as precursors, which were kept at 18 °C in stainless bottles. By comparison, the sample in Fig. 1b was exposed to a low energy remote O 2 plasma treatment before ALD. The pretreatment contained two steps. Each step consisted of 30 s remote O 2 plasma exposure followed by purging with Ar for 5 s. Afterwards, 120 cycles Al 2 O 3 was deposited in the same chamber. The different growth topography of Al 2 O 3 on MoS 2 basal planes are shown in Fig. 1. From Fig. 1a, it can be seen that due to the absence of dangling bonds on MoS 2 basal plane, direct deposition of Al 2 O 3 films are in forms of island-like clusters with large pinholes, and the lateral size of most pinholes are over 100 nm. In this situation, it is easy to imagine that the gate dielectric of top gate MoS 2 transistors have to be thick to form uniform film which is necessary to keep the leakage current at a sufficient lower level. By contrast, with remote oxygen plasma treatment prior to ALD, the grown Al 2 O 3 film is completely uniform on MoS 2 surface as shown in Fig. 1b. The difference is even more evident by the comparison between Fig. 1c,d, which are the corresponding AFM 3D images of Fig. 1a,b. Root mean square (RMS) of the ~12 nm directly deposited film is 5.35 nm, and it decreases to only 0.58 nm with remote oxygen plasma pretreatment, which is about 10% of the value in Fig. 1c. It is obvious that the remote O 2 plasma pretreatment served as an effective method to supply sufficient nucleation sites to achieve a uniform ALD process. More details about the direct deposition of Al 2 O 3 on pristine MoS 2 basal planes are available in the Supplementary Information. According to previous reports with graphene 24 and MoS 2 25,26 , heavy exposure to an oxygen plasma (typically a direct plasma) completely etches the flakes and results in the substitution of sulfur with oxygen and re-deposition of the surface materials during etching of the MoS 2 flakes. Therefore, further analysis is needed to investigate the impact of the low energy remote oxygen plasma.
In-situ investigations were performed to gain an insight into the mechanisms of uniform growth after a remote O 2 plasma pretreatment, looking at whether the MoS 2 flakes were oxidized during the treatment or oxygen atoms were adsorbed on the MoS 2 surface and acted as nucleation sites in the following ALD process. In the in-situ cluster system, the ALD system was connected to an X-ray photoelectron spectroscopy (XPS) system through a high-vacuum transfer line (the pressure was about 10 −10 mbar). The sample was first transferred from the load-lock chamber to the XPS chamber for characterization. It was then transferred to the ALD chamber for 5 s remote O 2 plasma treatment. Afterwards, the sample was transferred back into the XPS chamber for further measurements. This procedure was repeated twice with treatment time of 15 and 30 s. The Mo 3d, S 2s and S 2p regions of the XPS spectra are shown in Fig. 2a,b, respectively. The Mo 3d spectra consists of peaks around 229 and 232 eV, corresponding to the Mo 4+ 3d 5/2 and Mo 4+ 3d 3/2 components, respectively. Similar peaks appear around 161.8 and 163 eV, referring to the S 2p 3/2 and S 2p 1/2 components of the S 2p region, respectively. In Fig. 2, all these peaks have nearly no shift after the remote O 2 plasma treatment, implying that the chemical bonds were not damaged during the process. Also, no peaks appear around 236 eV, demonstrating that the molybdenum atoms were not oxidized after the plasma treatment 27,28 . This is different to the previous results where a direct RF-oxygen plasma was applied 25,26,28 . Based on the discussion above, it seems that remote oxygen plasma treatment is a surface-based process. When a remote oxygen plasma was applied, oxygen atoms were adsorbed onto the MoS 2 surfaces and acted as nucleation sites for the ALD process. The remote plasma was gentle enough such that the flakes were not oxidized during the treatments. When the flakes were transferred into the XPS chamber through the transfer line, the adsorbed oxygen atoms desorbed, caused by the high vacuum in the transfer line and the reductive environment due to the working principle of the molecular pump. In addition, it should be noticed that the intensity of both Mo 3d and S 2p peaks varied with the measurement position actually. As there were a lot of MoS 2 flakes on the tested SiO 2 substrate, and tiny position shifts between adjacent measurements were inevitable, results obtained from different XPS measurements may contain information from different MoS 2 flakes. Meanwhile, since the thickness and density of MoS 2 flakes varied according to their locations, there were some intensity differences in both Mo 3d and S 2p components with treatment time. In this case, the intensity of Mo 3d and S 2p components with 5 + 15 s remote oxygen plasma treatment happened to be the maximum.
To further verify the results obtained by XPS, Raman spectra of the sample before and after the oxygen plasma treatments mentioned above were obtained in air using a 514-nm laser (Fig. 2c). The inset of Fig. 2c displays the spectra enlarged, showing the E g 2 1 and A 1g modes for MoS 2 at ~380 and 405 cm −1 . According to the work of Bertrand P. A. 29 , the in-plane E g 2 1 mode is brought about by the opposite vibration of two S atoms with respect to a Mo atom and the A 1g mode is generated from the out-of-plane vibration of S atoms in opposite directions. From the inset of Fig. 2c, consistency of peak positions between these two spectra at the E g 2 1 and A 1g modes can be observed before and after the sample undergoing the remote oxygen plasma treatments, implying that bonding situations of Mo and S atoms didn't change. In addition, the peak that centers near 820 cm −1 could be used to estimate the extent of oxidation that occurred 30 . In case that the MoS 2 flakes were oxidized, this peak would be more defined and intense after the treatments. As expected, this peak showed no intensity difference before and after the remote O 2 plasma pretreatments, indicating that the MoS 2 flakes remained un-oxidized during the pretreatments. From these results, it is clear that during the treatments, the remote O 2 plasma is gentle enough to avoid damaging the MoS 2 flakes. Instead, the produced oxygen species are adsorbed onto the MoS 2 surface and serve as nucleation sites for the initial TMA pulses during the ALD process.
As the mobility is of great significance when evaluating the performance of electronic devices, back-gated MoS 2 -based field effect transistors were fabricated to estimate the impact of the remote O 2 plasma pretreatment on the device mobility. A cross-sectional schematic of the MoS 2 transistor with the remote O 2 plasma treatment is shown in Fig. 3a, and thickness of the MoS 2 flake is about 8.4 nm as shown in Fig. 3b, which correspond to ~12 monolayers. The corresponding photograph of the device structure is shown in Fig. 3c. For the electrical characterization, one of the electrodes acts as a drain and the other one is grounded, acting as a source. Initially, Cr/Au electrodes are used with a MoS 2 channel by applying a source-drain bias (V ds ) to the pair of electrodes as shown in Fig. 3c and a gate bias (V bg ) to the heavily doped silicon substrate. As shown in the insets in Fig. 4a,c, the I ds -V ds curves are all linear in the range from − 40 to 40 mV with or without the remote O 2 plasma pretreatment, indicating that the Cr/Au contacts are ohmic contacts. The transfer and output characteristics for the MoS 2 transistor before and after 60 s remote O 2 plasma pretreatment are obtained for comparison. The data presented in Fig. 4 show typical n-type transistor behavior with an on/off ratio (I on /I off ) over 10 7 . This high on/off ratio compared to graphene transistors is attributed to the large band gap of MoS 2 . It is also observed in Fig. 4a,c that both the shape of the transfer curves and the values of the ON current are improved after 60 s remote O 2 plasma pretreatment. For example, the transfer current at V ds = 500 mV increases from 1.56 × 10 −5 to 3.38 × 10 −5 A after a 60 s remote O 2 plasma pretreatment. A low field-effect mobility is extracted using the Equation: [1] where L = 1 μ m is the channel length, W = 4.2 μ m is the channel width, and C i = 1.15 × 10 −8 F/cm 2 is the capacitance density between the channel and the back gate (details of the mobility extraction can be found in Supplementary Information). Results show that the mobility increases from the original value of 22.15 to 33.57 cm 2 /Vs after 60 s remote O 2 plasma pretreatment. Moreover, from the comparison between Fig. 4b (without pretreatment) and Fig. 4d (with pretreatment), the output current increase greatly as well after 60 s remote O 2 plasma pretreatment. Taking the I ds -V ds curve under V bg = 40 V for example, the output current increases from 2.81 × 10 −5 to 9.07 × 10 −5 A after 60 s remote O 2 plasma pretreatment. Furthermore, for both the transfer and the output characteristics, excellent field-effect behavior is observed. The evolution of the device mobility is tested with different pretreatment time. As shown in Fig. 5, the mobility of the device reaches its peak value with 60 s pretreatment, and decreases slightly with a prolonged pretreatment time, but still higher than the original value. Error bars of Fig. 5 are contributed from repeated measurements each time. To verify the experimental phenomenon, many other back-gate devices were fabricated and tested in the same manner. As expected, a similar phenomenon was observed, proving these results were not a coincidence. This indicates that there is a compromise between the mobility and pretreatment time. It should be noticed that the optimized pretreatment time should be different with different instruments. Figure 6 shows the top gate transfer characteristics and leakage current of a few layer MoS 2 transistor with top gate dielectric deposited using remote oxygen plasma pretreatment. Top gate dielectric of this device is 60 cycles Al 2 O 3 (about 6.6 nm) which was deposited at 300 °C with 60 s remote oxygen plasma pretreatment. L = 1 μ m and W = 5 μ m are the channel length and channel width, respectively. For all   the measurements in Fig. 6, back gate of the device is grounded as shown in Fig. 6a. From the inset of Fig. 6b, the linear relationship between I ds and V ds within − 40 mV − 40 mV indicates that Cr/Au electrodes form perfect ohmic contacts with the MoS 2 channel. In addition, for all the transfer curves presented in Fig. 6b, great on/off ratio of the current over 10 8 can be observed within the ± 3 V range of the top gate voltage. Top gate leakage current is also measured in the same device. Compared to previously reported top-gate leakage of 2 pA/μ m 2 within 2 MV/cm 1,8 , as shown in Fig. 6c, the leakage current is less than 5 × 10 −13 A (about 0.1 pA/μ m 2 ) in the measurement range of −3 V to 3 V (4.5 MV/cm). This leakage is much smaller and at same time with an ultrathin gate oxide. The field effect mobility of this top gate device is extracted using Equation (1) discussed above, which was 28 cm 2 /Vs under V ds = 0.5 V with the SS to be 101 mV/dec.

Discussion
In summary, uniform Al 2 O 3 growth on the MoS 2 basal plane was successfully achieved by applying a remote O 2 plasma pretreatment before ALD, and the mechanism was investigated systematically. After a remote oxygen plasma pretreatment, the oxygen species are physically adsorbed onto the surfaces of the MoS 2 flakes and act as nucleation sites for the ALD cycles. The transport studies reveal an extra benefit of this method, which is that unlike many other methods that might sacrifice the device mobility to achieve uniform high-κ growth, this method improves the device mobility by 50%. Furthermore, top-gated MoS 2 transistor with ultrathin Al 2 O 3 dielectric was also fabricated. With only 6.6 nm Al 2 O 3 as dielectric, which is the thinnest top gate dielectric ever reported so far, the device shows impressive leakage about 0.1 pA/μ m 2 at 4.5 MV/cm. Besides, the top-gated device shows great on/off ratio of over 10 8 , subthreshold swing (SS) of 101 mV/dec and a mobility of 28 cm 2 /Vs. According to the mechanism, it is believed that this method can also be adopted for high-κ growth on other two dimensional nanostructures and used in other devices. With further investigations and optimizations, this method could play an important role in the future nanoelectronics.

Methods
Preparation of the few-layer MoS 2 flakes. Ultrathin layers of MoS 2 were obtained from bulk crystals (SPI supplies Brand) using the classical tape-based mechanical exfoliation method commonly used for graphene, then transferred onto degenerately doped Si substrates covered with 300 nm SiO 2 . The thicknesses of these flakes were determined with a Bruker Multimode 8 atomic force microscope (AFM).

Atomic layer deposition of Al 2 O 3 on MoS 2 flakes and characterization. Some of the MoS 2 flakes
were loaded into the Picosun R200 ALD chamber for direct Al 2 O 3 deposition. During the deposition, TMA and H 2 O served as the aluminum and oxygen precursors, respectively, and different growth temperatures and pulse time were adopted to observe their impacts. For some of the flakes, the remote O 2 plasma pretreatments were carried out in the same chamber before Al 2 O 3 was deposited. Here, "remote" means that the plasma source is located remotely from the substrate stage, such that the substrate is not involved in the generation of the plasma. It is then carried to the sample surface by the carrier gas 31 . The X-ray photoelectron spectroscopy (XPS) system used was made by SPECS GmbH. The X-ray source for data acquisition during the in-situ characterization was SPECS XR50 X-ray source. Considering that the signal intensity was not so strong due to the low density of MoS 2 flakes on the substrate and the signal intensity might be further weakened by using an X-ray monochromator, we finally carried out the in-situ characterization using a non-monochromatic XPS source. In the in-situ XPS measurements, all the spectra were taken using a Mg Kα X-ray source (hν = 1253.6 eV). The working pressure in the ultra-high-vacuum (UHV) chamber for the data acquisition was maintained at the magnitude of 10 −10 mbar. The element library and the quantification factors used during measurements were provided via the system SpecsLab2 software, and after the measurements, the data analysis was carried out using CasaXPS software. The binding energies in the XPS spectra were calibrated in the conventional way against the adventitious carbon C 1s singlet (E b = 284.6 eV). The Raman spectra of the MoS 2 flakes before the remote O 2 plasma treatment were measured with a Renishaw inVia Raman microscope in air using a 514 nm laser. Then after going through the in-situ XPS characterization mentioned above, Raman spectra of the same sample was measured again with all the measurement settings to be the same.

MoS 2 transistors fabrication.
The degenerately doped Si substrate and the 300-nm SiO 2 layer served as the back gate and the gate dielectric, respectively. The source and drain contacts were formed using electron-beam lithography followed by deposition of 10 nm Cr and 70 nm Au. The electrical properties of the transistors were measured with an Agilent B1500 semiconductor device parameter analyzer.