Ruina
Liu
a,
Baoxin
Liao
a,
Xiangdong
Guo
a,
Debo
Hu
a,
Hai
Hu
a,
Luojun
Du
b,
Hua
Yu
b,
Guangyu
Zhang
b,
Xiaoxia
Yang
*a and
Qing
Dai
*a
aChina CAS Center for Excellence in Nanoscience, National Center for Nanoscience and Technology, Beijing 100190, P. R. China. E-mail: daiq@nanoctr.cn; yangxx@nanoctr.cn; Tel: +86-010-82545720
bBeijing National Laboratory for Condensed Matter Physics and Institute of Physics, Chinese Academy of Sciences, Beijing 100190, P. R. China
First published on 18th November 2016
The performance of electronic circuits is becoming limited by on-chip digital information transmission. Graphene plasmons with ultra-high confinement and low damping rates offer an effective solution to this problem as they allow for the implementation of optical interconnects. However, direct contact with the semiconductor always deteriorates the plasmonic properties due to large damping of the plasmon in the semiconductor. Here, we studied graphene plasmons in heterostructures of graphene and monolayer MoS2 which represents a promising semiconductor for next-generation electronic devices. The graphene plasmons in the heterostructures were systematically evaluated in terms of their resonance frequency, intensity, lifetime and electrical tunability and were found to keep their main characteristics. Experimental results exhibited both a redshift in the plasmon frequencies and a decrease in the resonance intensity for both graphene/MoS2 and MoS2/graphene devices when compared with graphene devices at the same gate bias. These results could be attributed to changes in the dielectric constant and effective doping of graphene. Furthermore, the conductivity saturation on the electron branch in the back-gated graphene/MoS2 device cancelled the electron plasmons. These findings demonstrate that electrically tunable graphene plasmons can be realized in contact with semiconducting MoS2. Our results provide a platform for the investigation of the integration of semiconductor-based electronic and optoelectronic devices with plasmonic devices through van der Waals heterostructures.
Based on the two aspects outlined above, graphene plasmonic waveguides can perform a dual function and simultaneously carry both optical and electrical signals, giving rise to exciting new capabilities. However, the integration of plasmons into semiconductors always deteriorates the lifetime of the plasmon due to extensive absorption inside the semiconductor medium.16 Although we chose one of the thinnest semiconductors, monolayer MoS2, to decrease the damping, the plasmonic properties of the graphene should be exactly studied before any practical applications are explored. Moreover, graphene’s plasmonic properties depend principally on its electrical properties, such as carrier density and mobility.7,17 Previous studies have reported charge transfer at the interface of graphene and monolayer MoS2. For example, in a graphene/MoS2 heterostructure with a back gate, saturation of graphene's transport on the electron branch was observed due to the negative compressibility of the MoS2 electron system.18 Carrier doping and transfer result in the formation of graphene p–n junction between the graphene/SiO2 and graphene/MoS2 boundary.19 These properties suggest that the graphene in graphene–MoS2 heterostructures possesses rich and tunable electrical transport properties, thus we expect graphene–MoS2 heterostructures to possess rich and tunable plasmonic properties.
Here, we report on a combined experimental and simulatory investigation of the plasmonic properties of graphene–MoS2 heterostructures. The plasmonic properties of the MoS2/graphene, and graphene/MoS2 devices supported on SiO2 substrates were comparatively studied with those of graphene devices on SiO2 using far-field infrared spectroscopy. The main features of the plasmons in the graphene–MoS2 heterostructures remained, although the resonance frequency redshifted and the resonance intensity decreased compared to the bare graphene device at the same gate bias (ΔCNP = Vg − VCNP, where ΔCNP is the gate bias, Vg is the gate voltage, and VCNP is the gate voltage at the charge neutral point). The electron plasmons of the graphene/MoS2 device were too weak to be distinguished due to the negative compressibility of MoS2, while the hole plasmons could be normally observed. Our results demonstrate the possibility of integrating graphene plasmonic devices with semiconductor-based electronic and optoelectronic devices.
The MoS2, graphene, MoS2/graphene, and graphene/MoS2 films were patterned into electrically continuous nanoribbon arrays with a 1:1 width-to-pitch ratio. The ribbon widths were characterized by atomic force microscopy (AFM) (Fig. 1c–e and S1b in the ESI†) to be 100 nm, with the graphene, MoS2, MoS2/graphene, and graphene/MoS2 nanoribbons possessing a uniform thickness of ca. 2.1 nm, 1.9 nm, 3.1 nm, and 4.0 nm, respectively. These values are larger than the thickness of a single layer of graphene (0.35 nm (ref. 25)) and MoS2 (0.6 nm (ref. 26)), which may be attributed to residual poly(methyl methacrylate) (PMMA) on the graphene during the transfer process and partial etching of the SiO2 substrate in the O2 plasma etching process.Ti/Au (5 nm/60 nm) metal stacks were deposited as the source and drain electrodes (an optical microscope image of the device can be viewed in Fig. S1a in the ESI†). A titanium buffer layer was used to form an Ohmic contact with the MoS2 or graphene layer and to increase the adhesion force of the Au electrodes. The tunneling barrier height was zero between the Ti and MoS2 under the electrodes, due to the metallic behavior of the Ti–MoS2 system.27 P+ doped Si beneath the 300 nm SiO2 layer was used as a back gate electrode. The Isd–Vsd curves measured from the bare graphene and MoS2 devices both exhibit a linear relationship (Fig. S2 in the ESI†), confirming the formation of an Ohmic contact.
The transfer curves of the graphene, MoS2/graphene and graphene/MoS2 nanoribbon devices are shown in Fig. 2b. Three key trends can be identified in these curves. Firstly, the Isd values of the MoS2/graphene and graphene/MoS2 devices are in the same order of magnitude as those of the bare graphene device, but are two orders of magnitude larger than those of the MoS2 device, suggesting that transport is mainly graphene controlled. Secondly, the VCNP (at ca. 50 V) value for the graphene/MoS2 device was considerably smaller than for the graphene and the MoS2/graphene devices (not observed in the region less than 100 V). Graphene is heavily p-doped here, which may be due to residual PMMA and H2O molecules31 or surface dangling bonds on the SiO2 substrate.32 Once the graphene was brought into contact with the MoS2, electrons transferred from the MoS2 to the graphene, as the work function of MoS2 is smaller than that of graphene (as displayed in Fig. 2c). This process weakens the p-doping effect in the graphene and leads to the formation of a Schottky barrier at the interface in the absence of a gate voltage (Vg) (Fig. 2d). MoS2 beneath the graphene can further reduce the p-doping from the SiO2 substrate. Thus, the VCNP value of the graphene/MoS2 device is the lowest among the three systems. Thirdly, the electron branch of the transfer curve of the graphene/MoS2 device is very different from those of the graphene and MoS2/graphene devices.33,34 As shown, the increase in Isd for the graphene/MoS2 device is extremely slow and reaches saturation when Vg approaches 100 V. This result is attributed to the lower mobility of electrons in MoS2 compared to graphene and also the band arrangement in the graphene/MoS2 heterostructure, which is induced by a sufficiently large positive gate voltage.18 Once the Fermi level of graphene is higher than the MoS2 conduction band edge, gate-induced electrons in graphene can easily transfer to MoS2 without a barrier, resulting in an electron increase in MoS2 and an electron decrease in graphene.
As shown in Fig. 3a–c, plasmons in the heterostructures of graphene and monolayer MoS2 exhibit similar properties to those in bare graphene. Monolayer MoS2 does not introduce new peaks into the extinction spectra in our detection region (from 675 cm−1 to 4000 cm−1), because there are no IR-active phonons nor any plasmon resonance in this region.39,40 Control experiments of the bare MoS2 device and the MoS2 film/graphene device further prove that there is no resonance absorption for MoS2, as shown in Fig. S4 in the ESI.† This may be attributed to the low electron concentration (1012–1013 cm−2) of MoS2 in our experiments. Theoretical studies indicate that plasmon resonances of two-dimensional MoS2 with a carrier concentration of 1012–1014 cm−2 appear in the far-infrared and terahertz regimes.39
The electrical tunability of graphene plasmons remained in the heterostructures of graphene and monolayer MoS2. As displayed in Fig. 3a and b, three resonance peaks shift to blue as |ΔCNP| increases due to the increased charge density. We extracted the value of all of these resonance peaks at each gate voltage and plotted them as a function of ΔCNP in Fig. 4a. From this figure, the tunability of plasmons in the heterostructures is obviously shown. Although the main features of the plasmons in the heterostructures remain, the introduction of monolayer MoS2 induced frequency and intensity changes in the resonance peaks. Compared with bare graphene devices, the resonance peaks of both the MoS2/graphene and graphene/MoS2 devices exhibit a redshift, as shown in Fig. 4a. For instance, when ΔCNP = −120 V, peak 1, peak 2, and peak 3 of the MoS2/graphene device redshift by 43 cm−1, 88 cm−1, and 125 cm−1, and the peaks of the graphene/MoS2 device redshift by 65 cm−1, 111 cm−1, and 133 cm−1 compared to the frequencies of the bare graphene device at 814 cm−1, 954 cm−1, and 1382 cm−1. Furthermore, the peak intensities of the heterostructures are weaker than that of the bare graphene device. For example, when ΔCNP = −120 V, the highest peak intensities for the graphene, MoS2/graphene, and graphene/MoS2 devices were 0.73%, 0.48%, and 0.23%, respectively.
The frequency redshift and strength reduction of plasmons in the heterostructures of graphene and monolayer MoS2 are due to changes in the environmental dielectric and the effective Fermi level of graphene, according to the equation:9,41,42, where e is the electron charge, EF is the Fermi level of graphene, q is the wave vector which can be determined by the ribbon width W via q = π/W, ℏ is the reduced Planck constant, ε0 is the vacuum permittivity of free space, and εr is the average dielectric constant of the environment around graphene. Here, for all the devices, W = 100 nm, which means that the value of q is fixed at a value of 3.14 × 105 cm−1. Thus, the resonance frequency is determined by two variables: EF and εr of graphene. The εr value of the graphene/SiO2 devices is , where εair = 1, and εSiO2 = 3.9.43 Because the reported static relative permittivity for single-layer MoS2 is εSL−MoS2 = 3.7,44 which is slightly smaller than that of SiO2 and much larger than εair, the εr value for the graphene/MoS2 device is slightly smaller than εr(G/SiO2), and the εr value for the MoS2/graphene device is large than εr(G/SiO2). However, the resonance frequencies of the bare graphene devices are much higher than those of the two heterostructured devices, and the frequencies of the graphene/MoS2 devices are lower than those of the MoS2/graphene devices at every ΔCNP value (Fig. 4a). These results indicate that the EF of graphene in the graphene/MoS2 device is the lowest at the same value of ΔCNP. The EF value in the bare graphene device was calculated using a parallel plate capacitor model (details in ESI†), as marked in Fig. 3f.
The plasmonic behavior of the MoS2/graphene, graphene/MoS2 and bare graphene devices were also studied via a finite element method. The simulated results corresponding to Fig. 3a–c are shown in Fig. 3d–f. An effective electrical ribbon width of 68 nm (ref. 37) and a width-to-pitch ratio of 1:2 was used for all of the devices (see more details of the simulation method in the ESI†). For the bare graphene devices, the calculated EF values were used in the simulation and the obtained simulated curves (Fig. 3f) were found to be in good agreement with experimental results (Fig. 3c). For the heterostructured devices, the EF value of graphene was used as a variable in the simulation and was adjusted to get the best fitting curves (the EF values used are shown in Fig. 3d and e). The resonance frequencies of each of the three peaks correspond closely to their experimental values and their resonance intensities were also in good agreement with the experimental values. The resonance intensities of the bare graphene devices were found to be approximately 2 times that of the MoS2/graphene devices and 4 times that of the graphene/MoS2 devices. The plasmon resonance intensity of graphene is closely related to its EF value, with large |EF| values tending to indicate a strong resonance intensity.45,46 Thus, the simulation results indicate that the graphene plasmon keeps its main features in the heterostructures of graphene and monolayer MoS2 while the effective EF and εr values determine its resonance frequency and intensity.
From the obtained |EF| values, we calculated the effective gate voltage of graphene at certain |ΔCNP| values in the heterostructures, as plotted in Fig. 4b. In the heterostructures, only part of the applied gate voltage works on the graphene layer due to screening and charge transfer between monolayer MoS2. Ignoring the carrier transfer between graphene and MoS2, we roughly estimated the carrier distribution in each layer of the graphene–MoS2 heterostructures via Thomas–Fermi (T–F) charge screening theory. Details can be found in the ESI.† The utilised charge screening lengths of graphene and MoS2 were λgraphene = 0.6 nm and λMoS2 = 7 nm.25,47 The calculated carrier density of graphene in the MoS2/graphene and graphene/MoS2 devices was ca. 73% and ca. 48% of that in the bare graphene device, respectively. This evaluation exhibited a similar tendency to previous results. The deviation between the exact values may arise from carrier transfer between graphene, and impurities, and the simplification of the T–F model for a two-atom thick layered heterostructure. We want to clarify that, due to its large screening length and ultra-thin thickness, the screening effect is much weaker than in bulk semiconductors, which may totally screen the gate electric field.
The lifetime of the plasmon, T, can be extracted from the infrared extinction spectra using the equation T = 2ħ/Γ, where Γ is the full width at half-maximum (FWHM) of the resonance peaks.48 Due to plasmon–phonon coupling with the SiO2 phonons, which can be described with the electromagnetic induced transparent (EIT) model or the Fano model, the Γ value of the plasmon resonance in the MoS2/graphene and graphene/MoS2 heterostructures were obtained via curve fitting.35 The details can be found in Fig. S5 in the ESI.† The calculated lifetimes for the MoS2/graphene devices are between 80 fs and 100 fs, and for the graphene/MoS2 devices are about 80 fs. These values are comparable with previously reported values for graphene/SiO2 devices in the same frequency range.35 This suggests that the monolayer MoS2 causes little loss to plasmons in the graphene and MoS2 heterostructures, which is beneficial for the support of plasmons with high performance in optoelectronic integrated devices. This is another advantage of MoS2 compared to bulk semiconductors for graphene plasmons.
Finally, we want to discuss the electron plasmons in the graphene/MoS2 heterostructure. The electrical tunability of the Fermi energy of the graphene device with the back gate is symmetric with respect to the CNP,33 however, it is asymmetric for graphene/MoS2. We measured the hole plasmons and the electron plasmons at both sides of the CNP of the graphene/MoS2 devices (Fig. 5a). The results show that the hole plasmons are distinct. However, the electron plasmons were not detectable, even at very high ΔCNP values (80 V). This phenomenon is consistent with the electrical properties of the graphene/MoS2 device, as proven by its transfer characteristics (the red line in Fig. 2b), where the Ids value nearly does not increase with Vg in the n-doped region. When the back gate voltage, Vg, is applied, it causes bending in the bottom-layer MoS2 conduction band (Fig. 5b). When the positive Vg exceeds a critical gate voltage, the EF of graphene becomes higher than the MoS2 conduction band edge, and thus gate-induced electrons in graphene are able to transfer to MoS2 without a barrier. This results in the electrons in MoS2 increasing dramatically and a reduction in the electron density in graphene (Fig. 5b), which is called the negative compressibility of the MoS2 electron system.18 Meanwhile, the contact between graphene and MoS2 transforms from a Schottky barrier into an Ohmic contact. Based on this property, we suggest a new kind of plasmonic structure based on the graphene/MoS2 heterostructure, where graphene is kept as a complete plane to keep its high quality and nanostructures are designed using MoS2.
Raman spectra were taken with a Horiba Jobin Yvon LabRAM HR800 microscope. The electrical transport properties were analyzed using a semiconductor parameter analyzer (Keithley 4200-SCS), which was equipped with three wolframium probes. AFM measurements were conducted using a scattering SNOM (Neaspec GmbH). Infrared transmission measurements were performed by FTIR microscopy (Thermo Fisher Nicolet iN10). All characterization of the devices’ performance was conducted under ambient conditions.
Footnote |
† Electronic supplementary information (ESI) available. See DOI: 10.1039/c6nr07081g |
This journal is © The Royal Society of Chemistry 2017 |