|Home | About | Journals | Submit | Contact Us | Français|
Transistors based on MoS2 and other TMDs have been widely studied. The dangling-bond free surface of MoS2 has made the deposition of high-quality high-k dielectrics on MoS2 a challenge. The resulted transistors often suffer from the threshold voltage instability induced by the high density traps near MoS2/dielectric interface or inside the gate dielectric, which is detrimental for the practical applications of MoS2 metal-oxide-semiconductor field-effect transistor (MOSFET). In this work, by using AlN deposited by plasma enhanced atomic layer deposition (PEALD) as an interfacial layer, top-gate dielectrics as thin as 6nm for single-layer MoS2 transistors are demonstrated. The AlN interfacial layer not only promotes the conformal deposition of high-quality Al2O3 on the dangling-bond free MoS2, but also greatly enhances the electrical stability of the MoS2 transistors. Very small hysteresis (ΔVth) is observed even at large gate biases and high temperatures. The transistor also exhibits a low level of flicker noise, which clearly originates from the Hooge mobility fluctuation instead of the carrier number fluctuation. The observed superior electrical stability of MoS2 transistor is attributed to the low border trap density of the AlN interfacial layer, as well as the small gate leakage and high dielectric strength of AlN/Al2O3 dielectric stack.
Molybdenum disulfide (MoS2), as a layered material from the transition metal dichalcogenide (TMD) family, has been widely studied in recent years1,2,3, for its intriguing properties such as atomic-layer thickness, tunable bandgap4, high mobility3 and good thermal stability1. MoS2 MOSFET has been shown to exhibit suppressed short channel effect and has the potential to be used in the next generation nanoelectronics2,5. Many other applications based on single- or multi-layer MoS2, such as flexible electronics6,7,8, photon detectors9,10 and gas sensors11, have been demonstrated. Single-layer MoS2 also offers new opportunities in novel piezoelectronics12,13 and valleytronics14,15. Moreover, the recent advance in wafer-scale deposition of high-quality MoS2 films16 has made these MoS2 based applications even more promising.
However, the performance of MoS2 MOSFET is very sensitive to the ambient conditions and electrical stress17,18,19. As reported previously, when exposed to air, the MoS2 MOSFETs could exhibit large performance variation in terms of large shift in threshold voltage (i.e. hysteresis), due to the water or oxygen adsorbates18,19. This threshold voltage instability (ΔVth) is detrimental to both logic and analog circuit applications. Moreover, these ambient adsorbates will further degrade the carrier mobility and cause severe current fluctuation noise20. The MoS2 MOSFET has to be passivated in order to alleviate the ambient influence and achieve reliable performance. Because of the dangling-bond-free nature of MoS2 surface, the deposition of high quality dielectric on MoS2 can be challenging21,22,23,24,25. Even with passivation, MoS2 MOSFETs still exhibit significant hysteresis problem7,26, due to high-density traps at MoS2/dielectric interface or in the gate dielectric.
Both hexagonal boron nitride (hBN) and aluminum nitride (AlN) are suggested to be ideal gate dielectrics for MoS2 FETs27. It has already been demonstrated that the hysteresis of MoS2 transistor was greatly reduced by using the exfoliated hBN as gate dielectric, benefiting from the good MoS2/dielectric interface8. But as a layered material, hBN is usually obtained by mechanical exfoliation8,28 or synthesized at high temperatures29,30, which makes it difficult to be deposited reliably on large scale as top-gate dielectric. In contrast to the layered hBN, AlN is a bulk material with a larger bandgap of 6.3eV and higher dielectric constant of 9.1427,31, and can be deposited uniformly on large samples by PEALD32 or thermal ALD33. Moreover, our recent result has shown that by using AlN and Al2O3 stack as gate dielectric, AlN is capable of isolating the channel from the bulk traps of Al2O3 and achieving a low border trap density34,35.
In this paper, we report the experimental demonstration of single-layer MoS2 MOSFETs with AlN/Al2O3 as top-gate dielectric. By the insertion of AlN interfacial layer, the gate dielectrics scaling down to as thin as 6nm are realized. The electrical stabilities of the fabricated devices are systematically characterized under different conditions. The transistor shows very small hysteresis even under large gate biases and high temperatures. Low-frequency noise characterization is conducted and the MoS2 transistor exhibits suppressed current fluctuation. The observed excellent threshold voltage stability is contributed by the low border trap density of AlN near the MoS2 interface, as well as the small gate leakage and high dielectric strength of AlN/Al2O3.
MoS2 samples used here were synthesized on sapphire substrate by CVD method using high purity MoO3 and S powder as precursors. The triangular MoS2 flakes have a size of about 100μm (See Supplementary Figure S1). The Raman characterization shows peak distance of 19cm−1 for and A1g, suggesting the MoS2 flakes are single layered36. The MoS2 flakes were then transferred to a silicon substrate capped with 300-nm thermally grown SiO2. Depositing dielectric on the dangling-bond-free MoS2 is a challenging task21,22,23,24,25. At first, the process for depositing AlN/Al2O3 dielectric stack on MoS2 was carefully tested. Both AlN and Al2O3 were deposited using the Oxford Instruments OpAL ALD system. AlN was deposited at 170°C by using trimethylaluminum (TMA) and remote N2 plasma (20 sccm with a coil power of 25 W) as Al and N sources. The growth temperature and the RF power were optimized to minimize the plasma damage to the MoS2, resulting in a relatively slow growth rate of 0.18Å/cycle. The detailed optimization processes are provided in Supplementary Figure S2. After the deposition of 2-nm AlN, 18-nm Al2O3 was grown in-situ under thermal ALD mode at 200°C by using TMA and water vapor as Al and O sources.
The schematic structure of the single-layer MoS2 sample after the deposition of 2-nm AlN and 18-nm Al2O3 is shown in Fig. 1a. The surface morphology was characterized by AFM and shown in Fig. 1b. Continuous and smooth surface was obtained. Within a 1μm2 area, the root mean square (RMS) surface roughness is 0.54nm. In our experiment, the bare SiO2/Si wafer usually has a surface roughness of 0.15nm. However, since the MoS2 samples are transferred to the SiO2/Si wafer, the surface roughness increases to 0.35–0.5nm and dominates the above measured results. In contrast, for the MoS2 sample covered by 20-nm Al2O3 deposited at 200°C, as shown by the AFM image in Fig. 1c, the dielectric film is full of broken areas due to weak adhesion of the precursors (TMA and water) on the dangling-bond-free MoS2 surface, which is consistent with previous reports23,25. Since it has been suggested that organic or solvent residue can assist the successful deposition of dielectrics on MoS222, the failure of direct deposition of Al2O3 on the MoS2 sample also reflects that the MoS2 has maintained a clean surface during the fabrication process. The successful deposition of AlN/Al2O3 on MoS2 could be the result of the relatively low growth temperature of AlN. However, we found that even when the ALD growth temperature of Al2O3 was reduced to 170°C, similar poor quality was still observed when Al2O3 was directly deposited on MoS2 (See Supplementary Figure S1). Thus, we conclude that the improved surface morphology and quality of AlN/Al2O3 dielectric stack on the dangle-bond-free MoS2 surface is mainly the benefits of the low-power remote nitrogen plasma during the PEALD growth of AlN, which is similar to the O2 plasma functionalization of the multilayer MoS2 that was used to promote the ALD deposition of Al2O325. Even though the remote pure N2 plasma is very mild to single-layer MoS2 during several hours’ treatment (See Supplementary Figure S4), defects are still observed due to the possible Ar/H plasma damage during dose and purge of TMA (See Supplementary Figure S2(c)). At present, it cannot be determined to what extent the plasma damage facilitates the uniform dielectric deposition. Further experiments are needed to verify the possibility of AlN deposition on the dangling-bond free MoS2 only by the physical absorption of N-ion on the MoS2 surface, even for single-layer MoS2.
The successful deposition of AlN/Al2O3 stack on MoS2 was further verified by Time-of-Flight Second Ion Mass Spectrometry (ToF-SIMS), as shown in Fig. 1d. In this plot each relevant element’s intensity has been normalized by its peak value and offset vertically for clearer view. We can clearly identify the peak for AlN−, which appears ahead of Mo− and S− during the surface sputtering. Figure 1e shows the cross-sectional high resolution transmission electron microscopic (HRTEM) image of the SiO2/MoS2/AlN/Al2O3 structure. The single-layer MoS2 can be clearly identified. Both AlN and Al2O3 are in amorphous state without a visible distinct junction boundary between them. The N atom concentration is more likely to be higher near the MoS2 interface, due to the possible oxidation of AlN surface during the subsequent Al2O3 growth. The relatively bright area near MoS2/AlN interface might be the result of smaller density of AlN compared to that of Al2O3, which causes less electron scattering in the bright field TEM and thus becomes brighter. The AlN/Al2O3 layers are uniformly grown on MoS2 surface with no gaps or agglomerates, implying a reliable deposition process of AlN/Al2O3 as gate dielectric for MoS2 transistors.
Single-layer MoS2 transistors with 2-nm AlN/18-nm Al2O3 as top-gate dielectric were fabricated. Source/drain contacts were defined by electron-beam photolithography (EBL), followed by e-beam evaporation of 10-nm Ti and 50-nm Au and lift-off. The MoS2 flakes are further patterned into the designed channel size by EBL and O2 plasma etch. Then an AlN/ Al2O3 (2nm/18nm) stack was deposited on top of MoS2 as gate dielectric following the afore-described PEALD/ALD procedure. Another EBL was conducted to define the top gate, followed by e-beam evaporation of Ti/Au (10nm/50nm) and lift-off. Finally, the contact holes were formed by etching AlN/Al2O3 with developer FHD-5. The SEM image of fabricated devices is shown in Fig. 2a, together with the schematic cross-sectional view of the device. The channel width and gate length are 3μm. There is 100-nm spacing between the gate and source/drain contacts. The MoS2 transistors were measured at room temperature in atmosphere by Agilent B1505A device analyzer/curve tracer. Figure 2b shows result of the gate leakage and the hard breakdown test for the gate dielectric. The current of about 1pA when biased at small gate voltage is mainly limited by the equipment resolution. The AlN/Al2O3 dielectric shows a small leakage of about 0.1pA/μm2 for gate electric field as high as 4MV/cm. At the same time a high breakdown electric field of 8.8MV/cm is measured.
Figure 2c shows the transfer curves measured under different drain voltage biases ranging from 0.1V to 5V for transistor with 2-nm AlN and 18-nm Al2O3 as gate dielectric. During the measurements the gate voltage is swept from −5V to 5V then back to −5V again. The transfer curves for sweeping VG up and VG down are plotted by the solid and dashed lines respectively. Remarkably the transfer curves exhibit very small hysteresis for all the drain voltage biases. A large on/off ratio of about 106 is achieved, and the off-current is still limited by the equipment current resolution. The field effect mobility is extracted to be 3.3cm2/V·s by using equation μ=dID/dV×L/(WCoxVD) in four-probe measurement configuration, during which ε=9 is adopted as the relative dielectric constant for both AlN and Al2O3 to calculate Cox. This observed mobility is comparable to previous results for single-layer MoS2 transistors with MoS2 channel exposed to air8,37, but is smaller than that of 13–16cm2/V·s for similar devices passivated by ALD Al2O338,39. The relatively smaller mobility might be the results of varied CVD growth and fabrication conditions40 or specifically the remote plasma damage introduced by the PEALD growth of AlN in our case (See Supplementary Figure S4). Because single-layer MoS2 has only one atomic layer thickness and lacks the Thomas-Fermi screening effect to mitigate the impacts from the MoS2/dielectric interfaces3, the mobility of single-layer MoS2 becomes more sensitive to the dielectric environments and the plasma damage, and is reported to be smaller than that of multilayer ones8,37,41. By using high-quality exfoliated multilayer MoS2 in the future, which is less vulnerable to potential variations37 and more resistant to surface plasma treatment25,42, higher mobility can be achieved. Figure 2d shows the output curves, the black and red curves represent the results for stepping VG up and VG down respectively. Very good current saturation is observed, and the ID-VD near VD=0V exhibits linear relationship. There is almost no discrepancy between the output curves for sweeping VG up and VG down, suggesting reliable performance.
To fully exploit the potential of the AlN interfacial layer in promoting the uniform deposition of high-quality dielectric on the dangling-bond free MoS2 surface, gate dielectric consisting of 1-nm AlN and 5-nm Al2O3 is deposited, which is the thinnest dielectric ever reported for MoS2 transistor25. The MoS2 surface after the dielectric deposition is characterized by AFM and shown in Fig. 2e. Even with greatly reduced thickness, the AlN/Al2O3 on the dangling-bond free MoS2 still maintains a smooth surface with no pinholes, implying uniform nucleation sites provided by the 1-nm AlN interfacial layer. Figure 2f shows the transfer curves for MoS2 transistor with this ultra-thin stack as gate dielectric. Effective gate modulation and small hysteresis are observed, suggesting that by using PEALD AlN as an interfacial layer, high-quality dielectric even with sub-10nm thickness can be successfully deposited on the dangling-bond free MoS2, which is very important for the continuous scaling down of MoS2 FET for higher performance.
The transfer curves for transistor with 2-nm AlN/18-nm Al2O3 gate dielectric shown in Fig. 2c were measured with a total sweep time of 7s. However there are reports suggesting that the threshold voltage instability can be strongly affected by the sweep rate18,19. So we also measured the transfer curves with different sweep time ranging from 7s to 107s as shown in Fig. 3a. Still no obvious hysteresis is observed. By zooming in the transfer curves as shown in the two insets of Fig. 3a, we can identify the trends of shifting in the transfer curves. With increased sweep time, the transfer curve continuously shifts left during the up-sweep, and shifts right during the down-sweep. The above observation suggests that more trapped electrons are emitted with a slower up-sweep, and more electrons are trapped with a slower down-sweep. These threshold voltage shift trends result in an increased hysteresis as plotted in Fig. 3b, in which the hysteresis is calculated by the difference of VG corresponding to the same ID during the up- and down-sweep. Different ID criteria have been used, and the resultant hysteresis differs from each other due to the complex trapping and stabilization processes involved during a complete sweep. We find that smaller ID criterion yields larger hysteresis. Interestingly, a negative hysteresis is observed when the sweep time is short, which might be the result of gate-side electron injection43 and is consistent with the pulsed I-V measurement results (See Supplementary Figure S5). In general, with increased sweep time, the hysteresis increases but tends to be stabilized and is limited within 0.1V for all the ID criteria. Such a small hysteresis is the best result ever reported to date for MoS2 FET.
To better understand the observed hysteresis, transfer curves with different VG sweep amplitudes for transistor with 2-nm AlN/18-nm Al2O3 gate dielectric were measured and plotted in Fig. 4a. The gate is swept from negative to positive then back to negative voltage with a total sweep time of 7s. The hysteresis becomes more pronounced with increased VG amplitude. The hysteresis is extracted quantitatively as shown in Fig. 4b. It is observed that the hysteresis has little change when VG amplitude is smaller than 7V, but increases more significantly when VG amplitude is larger. It is also noticed that there is almost no Vth shift until the VG amplitudes is larger than 8V during the down-sweep. Recalling that the gate leakage in Fig. 2b also becomes prominent only when VG is larger than 8V, this Vth shift can then be correlated with the injection and trapping of electrons inside the gate dielectric44,45 as schematically shown in Fig. 4c. On the other hand, when the gate is negatively biased, the high electric field will enhance the emission of electrons trapped inside the gate dielectric, accounting for the negative Vth shift at larger VG amplitude in the down-sweep, as schematically shown in Fig. 4d. The above observation reveals that the MoS2 transistor with AlN/Al2O3 as top-gate dielectric presents small hysteresis even under large gate voltage bias, and the observed weak threshold voltage instability is mainly caused by the gate leakage at high gate electric field and the resultant trapping/detrapping of electrons in the border traps (within the bulk Al2O3 but close to the dielectric/MoS2 interface).
It has been demonstrated that MoS2 transistor is capable of working at high temperature, although significant mobility degradation and threshold voltage shift have been observed46. The MoS2 transistors fabricated in this work with 2-nm AlN/18-nm Al2O3 gate dielectric was tested at temperatures ranging from 25°C to 100°C, all conducted in ambient environment. The transfer curves are plotted in Fig. 5a, with a sweep time of 70s. The two insets of Fig. 5a show the zoomed-in transfer curves, from which we can clearly see the shifting trends of the transfer curves with increased temperature. A negative shift in Vth is observed at higher temperature, which is caused by enhanced emission of electrons trapped in the gate dielectric. The hysteresis increases at higher temperature, as quantitatively shown in Fig. 5b. This is because the gate leakage and the consequent electron injection/trapping is enhanced at higher temperature. Nevertheless, the hysteresis still remains smaller than 0.5V.
Nanoscale devices can suffer from the flicker noise (1/f noise) which increases with smaller channel size47. Thus, maintaining a low level of flicker noise is necessary to obtain high performance MoS2 transistors. On the other hand, the flicker noise characterization is a useful tool to diagnose the channel/gate-dielectric interface20,47,48, since the conventional C-V measurement has become difficult due to the limited channel area. The flicker noise spectra for the fabricated MoS2 transistor with 2-nm AlN/18-nm Al2O3 gate dielectric at various temperatures were measured using SR570 low-noise current preamplifier. The results at 25°C are shown in Fig. 6a. The spectra correspond to different gate biases with a fixed drain bias of 0.1V. The measurement of each spectrum takes a total sampling time of 120s. The inset of Fig. 6a shows the average sampled ID (red), together with the quasi-static ID (black), showing no obvious discrepancy. This agreement suggests again very good electrical stability of the MoS2 transistors even after long-term electrical stress. The normalized flicker noise can be expressed as , in which A is the noise amplitude and f is the frequency. Ideally the frequency exponent γ should be close to 120. Figure 6b shows the noise amplitude A ( at 1Hz) and also the fitted frequency exponent γ, for different gate biases at various temperatures. All the values of γ are close to 1, suggesting that the noise spectra follow the 1/f relationship well. Meanwhile, we find that higher flicker noise occur at higher temperatures. The normalized noise at 1Hz decrease monotonously with the gate voltage, suggesting that there is less relative fluctuation when more carriers are present in the channel.
The generation of flicker noise can be explained by the carrier number fluctuation or the mobility fluctuation49. For noise generated by carrier number fluctuation, the flicker noise is proportional to (gm/ID)2, where gm is the gate transconductance, while the flicker noise generated by the mobility fluctuation would be simply proportional to 1/ID49. To find the exact origin of the flicker noise, (at 1Hz) and (gm/ID)2 are plotted together in Fig. 6c. We find that deviates from (gm/ID)2 significantly by several orders. On the other hand, follows 1/ID relationship well as indicated by the dashed line. Therefore, we conclude that the flicker noise observed in our MoS2 transistor is caused by the mobility fluctuation instead of the carrier number fluctuation, which is consistent with previous report for single-layer MoS2 transistor20.
According to the Hooge empirical relationship, the flicker noise can be expressed by 47. The Hooge parameter αH is extracted and plotted in Fig. 6d. It can be seen that the Hooge parameter increases with temperature, indicating severer mobility fluctuation at higher temperature. The measured Hooge parameter at 25°C is 0.011, which is two orders of magnitude smaller than the published result for the same mobility range20, suggesting excellent suppression of flicker noise. The flicker noise caused by carrier number fluctuation is related to the border trap density by equation , in which is the equivalent gate voltage spectral density47. Since in our case the flicker noise is generated by the mobility fluctuation, it is difficult to extract the accurate border trap density directly, but the observed flicker noise has placed an upper limit on the possible border trap density. By assuming the flicker noise is all generated by the carrier number fluctuation when VG is biased above the threshold voltage at 25°C, we calculate the upper limit of the border trap density to be λNt1.6×1012cm−2eV−1 (See Supplementary Figure S6 for the results of SVG).
Single-layer MoS2 MOSFETs with AlN/Al2O3 gate dielectric are demonstrated. Gate dielectrics scaling down to 6nm on the dangling-bond free MoS2 are realized by the insertion of AlN interfacial layer. The MoS2 transistor with 20-nm top-gate dielectric exhibits hysteresis smaller than 0.1V at room temperature in ambient environment. The hysteresis increases when biased at higher gate voltage or measured at higher temperature, but remains below 0.8V even for gate-dielectric electric-field as high as 4.5MV/cm or temperature up to 100°C. This remarkable electrical stability mainly benefits from the low border trap density, enabled by the inserted AlN interfacial layer, and consequently small gate leakage and high dielectric strength of the AlN/Al2O3 stack. The MoS2 MOSFET also presents a low level of flicker noise, which is generated by the mobility fluctuation instead of the carrier number fluctuation. AlN/Al2O3 with AlN as interfacial layer is shown to be a promising candidate as both excellent gate dielectric and effective passivation for implementing reliable MoS2 MOSFETs, and probably also has the potential to be used in other nanodevices, such as transistors based on graphene, carbon nanotubes and other TMDs, which also face the problems of large hysteresis and reliable dielectric deposition due to the lack of surface bond.
MoS2 flakes were synthesized on sapphire substrate by CVD method using high purity MoO3 and S powder as precursors. The synthesized MoS2 has a flake size of about 100μm, and the layer thickness was checked by Raman measurement (514nm, inVia Renishaw). The Raman characterization shows peak distance of 19cm−1 for and A1g, suggesting the MoS2 flakes are single layered. To transfer the MoS2 flakes to Si substrate covered by 300nm thermal SiO2, the sample was first spin coated with PMMA A4 at 3000rpm and baked at 130°C for 2min, then soaked in 10% KOH at 80°C for hours until the PMMA membrane was separated from the sapphire substrate and floated on the water surface. The PMMA membrane was fished to DIW for several times and finally fished to the Si target substrate.
Source/drain contacts were first formed by EBL (Raith e-line), followed by e-beam evaporation of 10-nm Ti and 50-nm Au and lift-off in acetone. MoS2 flakes were then patterned to the desired channel size by EBL and O2 plasma etch. Both AlN and Al2O3 were deposited using the Oxford Instruments OpAL ALD system. AlN was deposited at 170°C by using trimethylaluminum (TMA) and remote N2 plasma (20 sccm with a coil power of 25 W) as Al and N sources. The growth temperature and the RF power were optimized to minimize the plasma damage to the MoS2, resulting in a relatively slow growth rate of 0.18Å/cycle. After the deposition of AlN, Al2O3 was grown in-situ under thermal ALD mode at 200°C by using TMA and water vapor as Al and O sources. Two kinds of AlN/Al2O3 dielectric stacks consisting of 2-nm AlN/18-nm Al2O3 or 1-nm AlN/5-nm Al2O3 are deposited. The thicknesses of AlN and Al2O3 were verified by ellipsometer on the Si dummy wafer (J.A. Woollam M-2000V). During the measurement, 1.5nm native SiO2 layer was taken into consideration. A Cauchy dielectric model for AlN was developed by fitting the Ellipsometer data measured from 18.5nm AlN. Another EBL was conducted to define the 3-μm-long gate electrode made of Ti/Au (10nm/50nm). Finally the source and drain contact holes were formed by EBL and wet etch in FHD-5 for 10min. After the fabrication, no further annealing was conducted.
The successful deposition of AlN and Al2O3 thin films was verified by TOF-SIMS V (ION-TOF GmbH, Münster, Germany), with an analysis spot size of 41μm×41μm. The surface of MoS2 after deposition was checked by AFM (XE150S, Park system). The cross-sectional view of MoS2/AlN/Al2O3 was done by TEM (JEOL 2010F). The electric device performances were measured by Agilent B1505A device analyzer/curve tracer, inside a probe station equipped with a thermal chuck. To measure the low frequency noise, the transistor was first biased at a fixed gate voltage and drain voltage (), then the drain current was amplified by the low noise current preamplifier (SR570, Stanford Research System) and sampled for a total time of 120s. Later on Fourier transformation was performed to the recorded current for every 4s, and the final single-sided current power spectrum was obtained by averaging the 30 independent results.
How to cite this article: Qian, Q. et al. Improved Gate Dielectric Deposition and Enhanced Electrical Stability for Single-Layer MoS2 MOSFET with an AlN Interfacial Layer. Sci. Rep. 6, 27676; doi: 10.1038/srep27676 (2016).
This work was financially supported by N_HKUST636/13. The authors would like to thank all the supports from the technical staffs in Nanoelectronics Fabrication Facility (NFF) and Material Characterization and Preparation Facility (MCPF) of The Hong Kong University of Science and Technology.
Author Contributions K.J.C. and Q.Q. conceived the experiments. F.L., Y.X. and R.Y. synthesized the MoS2 flakes. Q.Q., M.H. and Z.Z. fabricated the devices. B.L. helped with the EBL procedure. Q.Q. conducted the ALD deposition and test of devices. Q.Q. and M.H. performed the SIMS/TEM characterizations. K.J.C., Q.Q. and B.L. wrote the manuscript. All authors reviewed the manuscript.