Recent studies have increased the cut off frequencies achievable by exfoliated MoS2 by employing a combination of channel length scaling and geometry modification. However, for industrial scale applications, the mechanical cleavage process is not scalable but, thus far, the same device improvements have not been realized on chemical vapor deposited MoS2. Here we use a gate-first process flow with an embedded gate geometry to fabricate short channel chemical vapor deposited MoS2 radio frequency transistors with a notable fT of 20 GHz and fmax of 11.4 GHz, and the largest high-field saturation velocity, vsat = 1.88 × 106 cm/s, in MoS2 reported so far. The gate-first approach, facilitated by cm-scale chemical vapor deposited MoS2, offers enhancement mode operation, ION/IOFF ratio of 108, and a transconductance (gm) of 70 μS/μm. The intrinsic fT (fmax) obtained here is 3X (2X) greater than previously reported top-gated chemical vapor deposited MoS2 radio frequency field-effect transistors. With as-measured S-parameters, we demonstrate the design of a GHz MoS2-based radio frequency amplifier. This amplifier has gain greater then 15 dB at 1.2 GHz, input return loss > 10 dB, bandwidth > 200 MHz, and DC power consumption of ~10 mW.
Molybdenum disulfide (MoS2) is a two-dimensional (2D) semiconductor in the family of transition metal dichalcogenides. Its single layer direct bandgap of ~ 1.8 eV1 allows for high ION/IOFF metal-oxide semiconducting field-effect transistors (FETs).2 More relevant for radio frequency (RF) applications, theoretical studies predict MoS2 can afford saturation velocities, vsat, greater than 3 × 106 cm/s.3 While the RF applications of MoS2 are still in their infancy, recent studies have pushed exfoliated MoS2 cut off frequencies significantly higher.4,5,6 The study by Krasnozhon et al. employed edge contacts to achieve an intrinsic fT of 25 GHz in trilayer exfoliated MoS2.5 The work by Cheng et al. used transferred gate stacks on multilayer exfoliated MoS2 to obtain an intrinsic fT of 42 GHz.6 However, for industrial scale applications, the mechanical cleavage process is not scalable and, thus far, there have been few studies on chemical vapor deposited (CVD) MoS2 RF FETs.7, 8 Our prior results demonstrated monolayer CVD MoS2 top-gated FETs with an fT of 6.2 GHz.7 Furthermore, the work of Chang et al. used transferred monolayer CVD MoS2 on a flexible substrate to achieve an fT of 5.6 GHz.8 As has been done with exfoliated MoS2, optimized device configurations must be applied to CVD MoS2 to push cut off frequencies higher. In this study, we take a step in this direction by taking an embedded gate approach to CVD MoS2 FETs.
The embedded gate transistor has been well established in literature, particularly for the most prototypical 2D material, graphene.9 Graphene RF FETs using embedded gate transistors have achieved sub-THz cut off frequencies.10 However, the Dirac cone band structure of graphene results in a zero bandgap, which limits current saturation in graphene devices.11 The embedded gate structure has been shown to significantly improve current saturation in graphene RF FETs, resulting in improved voltage and power gain.12 Embedded gate FETs using CVD MoS2 for digital circuits were recently studied, which found increased scalability, with higher yield and uniformity vs. their top-gated counterparts.13 Additionally, embedded gate CVD MoS2 FETs show enhancement mode operation, which is essential for complex multistage integrated circuits. The common theme allowing improved performance among all 2D materials using embedded gates is a resist-free channel-dielectric interface with a reduced number of fabrication process steps after the transfer of the active material.
Motivated by the advantages described above, in this work, we employ an embedded gate structure to advance the RF characteristics of CVD MoS2. Our devices exhibit enhancement mode operation, ION/IOFF ratio of 108, and the highest CVD MoS2gm of 70 μS/μm. Using the measured S-parameters, we demonstrate a MoS2-based cascode amplifier showing gain greater than 15 dB at 1.2 GHz.
Results and discussion
Figure 1a shows the Raman spectra of the as-grown and transferred MoS2. The slight shift in the in-plane vibrational mode, E2g, corresponds to a release of strain that exists in as-grown samples due to the mismatch of thermal coefficients between MoS2 and the SiO2 substrate.14 The result of the same effect is also observed in the blue shift in the peak position of the photoluminescence spectra (Fig. 1a inset). Figure 1b shows an optical image of a fabricated CVD MoS2 device array. Figure 1b inset (bottom-right) shows the zoomed-in image of the gate fingers while Fig. 1b inset (top-left) is a schematic of the cross-sectional view of the embedded gate device structure.
Figure 2a shows the Ids−Vgs transfer characteristics of an embedded gate MoS2 FET with a physical gate length, Lg = 150 nm. The embedded gate structure preserves a clean MoS2-dielectric interface with less fixed-charge and organic impurities.13 As a result, the threshold voltage, Vth, is positive and close to 0 V, in contrast to many top-gated structures, which suffer from uncompensated charge at the surface.7 As shown in Fig. 2b, we measure a transconductance, gm, of 70 µS/µm at a Vds = 2.5 V. This is among the highest reported gm for MoS2 FETs on exfoliated or CVD grown films, at equivalent electric fields.5, 7 The maximum field-effect mobility, µFE, is 21.2 cm2/Vs, and after contact resistance correction is 81.6 cm2/Vs.15 In addition to the improved MoS2-dielectric interface, the embedded gate structure provides improved gate control over the channel. After transfer, the MoS2 wraps conformally to the rectangular gate fingers. The electrostatic control is improved over top-gated devices because the electric field flux reaching the MoS2 channel is increased at an equivalent gate voltage. As an enhancement mode device, we can sufficiently deplete the channel to achieve an ION/IOFF ratio of 108.
Figure 2c shows the Ids−Vds output characteristics for the same device. The device achieves current densities of 120 µA/µm at Vds = 3.5 V. There is clear current saturation beyond Vds of 3 V, resulting in a drain conductance, gds, within 5 µS/µm in the current saturation region (3 to 3.5 V of Vds). The sub-linear curves at low Vds indicate a significant Schottky barrier. We attribute this to undoped contacts and access regions. Our previous top-gated CVD MoS2 RF FETs  employed sub-stoichiometric HfO1.56 which extrinsically doped the MoS2 contacts and access regions. This n-doping helped reduce the contact/access resistance due to thinning of the Schottky barrier width. The difference in surface charge transfer doping between oxygen-deficient HfO2−x and stoichiometric Al2O3+x has been observed and well quantified in literature.16,17,18 The elemental composition and chemical stoichiometry of atomic layer deposition (ALD) alumina were investigated by x-ray photoelectron spectroscopy (XPS), as shown in Figure S6. The chemical composition of the Al2O x was quantified using the Casa XPS software, in which x was found to be about 3.1. The Al2O3.1 used in this study is oxygen-rich and closer to stoichiometric composition, providing minimal extrinsic doping. As a result, the extracted contact resistance for the devices in this study is Rc = 3.5 kΩ∙µm, which is larger than that of our sub-stoichiometric HfO1.56 devices.7 However, using a higher bandgap dielectric, alumina, and a carefully deposited uniform gate finger surface allows for the scaling of the dielectric thickness down to 10 nm. We note that in our devices the top MoS2 is exposed to ambient, allowing further dielectric/passivation layers or sensor measurements. The choice of top dielectric can provide the needed charge transfer to further improve the DC and RF characteristics.
Microwave performance was characterized from 0.1–15 GHz using an Agilent two-port vector network analyzer (VNA-E8361C). To accurately subtract the effects of the parasitic capacitances and resistances in the ground-signal-ground layout, we employed the standard de-embedding method using OPEN and SHORT measurements on the same device. Figure 3a shows the short circuit current gain |h21| vs. frequency. Operating at Vds = 3.5 V with Lg = 150 nm, we achieve an extrinsic fT of 3.3 GHz and, after de-embedding, an intrinsic fT of 20 GHz. The intrinsic fT is corroborated with the Gummel approach, as shown in Figure S3 of the Supplementary Information (SI). The lateral electric field used for the fT measurement is in the high-field limit where carrier transport is determined by vsat = 2π∙fT,int∙Lg, and in our device is 1.88 × 106 cm/s. This is the highest extracted vsat for room temperature MoS2 RF FETs, either exfoliated or CVD (Table 1). Following Equation 1 of the SI, a higher fT can be achieved by improving the gm (by shortening the Lg), optimizing the layout to reduce parasitic capacitances, and reducing the contact resistance.
Another figure of merit for high-frequency transistors is the maximum frequency of oscillation, fmax. This is the frequency limit at which there is power gain, given matched input and output impedances, and can be expressed by Equation 2 of the SI. Figure 3b shows the maximum available power gain vs. frequency. Operating at the same DC bias, we measure an extrinsic fmax of 9.8 GHz and an intrinsic fmax of 11.4 GHz. This is the highest extrinsic fmax for MoS2 reported so far, and the highest intrinsic value for CVD MoS2. We attribute the high fmax to good current saturation, leading to a small gds and a large output resistance, ro. Mason’s unilateral gain, Ug, is shown in Figure S4 of the SI, which is 10.3 GHz. Figure 3c shows the scaling of fT and fmax with gate length. In the high-field limit, the fT generally scales inversely with length while fmax does not follow any trend due to varying individual device gate resistance and output conductance. In RF circuit design, it is important to know the intrinsic voltage gain Av. The general expression for Av is gm/gds, but it can also be measured vs. frequency with impedance Z-parameters. Figure 3d shows the extrinsic and intrinsic unity voltage gain cut off frequency as 4.6 and 19.5 GHz, respectively. The outstanding RF device performance achieved here can be attributed to (i) a resist-free MoS2-dielectric interface, (ii) a reduced number of post-transfer fabrication process steps, compared with a top gate flow, and (iii) current saturation leading to a large ro.
Variability is an important factor in obtaining high yield with device-to-device uniformity on a large-scale RF transistor integration. The sources of variability in the fabrication process presented here include intrinsic material non-uniformity, e-beam lithography, and local differences in the various device capacitances. Switching to a gas-based precursor system such as metalorganic chemical vapor deposition (MOCVD) can improve the material coverage and uniformity,19 however, further work is needed to achieve the same device performance. A fully photolithographic process can help mitigate device-to-device feature size variations, which will also allow predictable device parasitics. To give some insight into the variability and scalability of our CVD MoS2 RF devices, we have shown the data from a single wafer device array in Table S1 of the SI. We note that the non-functioning devices are not necessarily due to poor CVD MoS2 sites, but were also caused by fabrication failures and electrical failures during biasing. An example of a fabrication failure is poor lift-off of the metal while an example of an electrical failure is gate breakdown during DC biasing.
Using the as-measured S-parameters, we designed a RF amplifier to demonstrate CVD MoS2 as a viable material for a front-end low noise amplifier (LNA). We chose a cascode topology to minimize the Miller effect, provide input-output isolation, and improve the overall gain-bandwidth product. The amplifier schematic is shown in Fig. 4a. Figure 4b shows the amplifier gain (S21 in dB) and the input match (S11 in dB). We see a gain greater than 15 dB operating at 1.2 GHz, attributed to good current saturation. Figure 4c shows the input-output isolation greater than 20 dB at 1.2 GHz as afforded by the cascode topology. The high gain of our front-end amplifier helps with setting the noise floor of the RF receiver. An input match network (IMN) is designed for matching the input impedance (Zin) to the source impedance, Rs = 50 Ω. With the IMN we see input return loss greater than 10 dB. An output match network is employed to improve the gain over the frequency bandwidth of 1.1–1.3 GHz. A standard Chebyshev Nth-order band pass filter topology can be modified for the design of the match networks.20 We plan to further improve this design to take into account noise parameters for the practical design of MoS2-based GHz LNAs.
In conclusion, we presented embedded gate CVD MoS2 RF FETs with an intrinsic fT and fmax of 20 and 11.4 GHz, respectively. From a minimum Lg of 150 nm, we extract a MoS2vsat of 1.88 × 106 cm/s, higher than previously reported experimental values. Our devices exhibit enhancement mode operation, ION/IOFF ratio of 108, and the highest CVD MoS2gm of 70 μS/μm. Using the measured S-parameters, we demonstrate a MoS2-based cascode amplifier showing gain greater than 15 dB at 1.2 GHz. Our results advance the state-of-the-art cut off frequencies achievable by CVD MoS2 and represent a step towards fully integrated MoS2 smart nanosystems.
Materials and methods
Large area atomic single layer MoS2 was grown by a standard vapor transfer process starting from molybdenum trioxide (MoO3) and sulfur (S) powder in a quartz tube (inner diameter = 22 mm) with a single zone Lindberg/Blue M furnace. Controlled large area growth was accomplished by using masking and target substrates. Both the masking and target substrates used were from the same wafer. The polished side of the substrates faced the MoO3 precursor. By controlling the distance between individual masking substrates, we could control the area of the continuous monolayer region. Additional detailed growth methods are available in Section 1 of the SI.
Raman and photoluminescence (PL) spectroscopy were done using a Witec Alpha 300 micro-Raman confocal microscope, with the laser operating at a wavelength of 488 nm. Parameters in our mapping were (i) grating (Raman) = 1800 g/mm, (PL) = 600 g/mm; (ii) integration time/pixel = 1 s; (iii) resolution = 3 pixels/µm.
The RF MoS2 transistor fabrication begins with patterning two embedded gate fingers on intrinsic Si/SiO2 (>20 k Ω∙cm). Electron beam lithography (EBL) and e-beam evaporation were used to define and deposit the embedded gate metal stack consisting of 2/23 nm Ti/Au. Horizontal gate lengths were verified to be between 100 and 500 nm. The shortest total Lg is 150 nm; the sum of the horizontal length (100 nm) and the two heights of the sidewalls (50 nm). ALD at 200°C was used to deposit 10 nm of Al2O3 as the gate dielectric. Large area CVD MoS2 was then transferred by poly(methyl methacrylate)-assisted wet transfer. A via to connect the embedded gate fingers with the gate pad was etched with phosphoric acid. The active MoS2 channel was etched using Cl2/O2 plasma. A final EBL step patterned source and drain (S/D) contacts consisting of 2/70 nm Cr/Au.
Electrical DC characterization was done on a Cascade Microtech Summit 11000B-AP probe-station using an Agilent B1500A parameter analyzer. The measurements were taken at room temperature, in ambient atmosphere, and in the dark. Microwave performance was characterized from 0.1–15 GHz using an Agilent two-port vector network analyzer (VNA-E8361C) at room temperature, in ambient atmosphere, and in the dark. The OPEN de-embedding was performed on the as-measured device-under-test by etching away the MoS2 in the active regions. The SHORT de-embedding is subsequently performed by depositing a strip of metal across the channel region, shorting out all pads. The intrinsic parameters are extracted using Equation 3 of the SI.
The authors declare that [the/all other] data supporting the findings of this study are available within the paper [and its supplementary information files]. Any additional data that supports the findings of this study are available from the corresponding author upon reasonable request.
Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The authors would like to thank the Army Research Office for partial support of this work under STTR award number W911NF-14-P-0030, and the NSF NASCENT ERC and NNCI programs. D.A. acknowledges the support of Northrop Grumman Aerospace Systems.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.