1. Introduction
In the last decade, molybdenum disulfide (2H-MoS
2) emerged as the most investigated two-dimensional (2D) semiconductor material of the transition metal dichalcogenides (TMDs) family, due to its unique physical properties, combined to a good chemical stability and its abundance in nature [
1]. MoS
2 crystals (and in general all TMDs) are characterized by strong in-plane bonds between the chalcogen (X) and the transition metal (M) atoms and weak van der Waals (vdW) interactions between the stacked layers [
2,
3,
4,
5]. In particular, the 2H-MoS
2 polytype exhibits a tunable bandgap as a function of the thickness, i.e., an indirect bandgap of 1.2 eV in the bulk form and a direct bandgap of 1.8–1.9 eV for a monolayer (1L) MoS
2 [
6,
7]. One layer and few-layers MoS
2 have been employed as channel materials in field-effect transistors (FET), showing very promising performances in terms of the on/off state current ratio (≥10
8) and decent mobility values (up to ~200 cm
2/Vs under proper conditions) [
1]. These properties make MoS
2 one of the potential replacements of Si for the continuation of the Moore scaling law in digital electronics [
8]. Furthermore, 2H-MoS
2 is very appealing for a wide range of More-than-Moore applications [
1,
9,
10,
11,
12], including sensing [
13,
14], photocatalysis [
15,
16], photovoltaics [
17,
18], and photonics [
19,
20] until reaching the more exotic spin-valley physics [
21,
22,
23].
In this context, the dangling bonds-free MoS
2 surface allows the creation of several vdW heterostructures by the combination of various 2D materials (2D/2D vdW heterostructures) [
24,
25,
26,
27], by integration of MoS
2 with semiconductor nanowires (1D core-shell heterostructures) and with bulk semiconductors (2D/3D vdW heterostructures) [
1,
28,
29,
30,
31,
32]. In particular, increasing research efforts have been directed in the last years to the integration of MoS
2 with wide-bandgap (WBG) semiconductors, including silicon carbide (SiC), gallium nitride (GaN), and related group-III nitrides (AlN and AlGaN alloys). The combination of the unique physical properties of MoS
2 with the robust properties of highly mature WBG semiconductors (such as high breakdown field and electron saturation velocity [
33,
34]) set the basis for the realization of new heterojunction diodes that exploit the vertical current at the MoS
2/WBG interface [
35,
36,
37] and of advanced photodetectors covering both the UV and the visible spectral range [
38,
39,
40].
2H-MoS
2 exhibits a very low lattice constant mismatch with the basal planes of 4H-SiC (~2.9%) [
41] and 2H-GaN (<1%) [
42] crystals, which represents a favorable condition for highly oriented epitaxial growth of MoS
2 on these hexagonal substrates [
43]. Furthermore, the small difference between the thermal coefficient expansion of the MoS
2/GaN heterostructure (α
MoS2 − α
GaN ≈ 0.97 × 10
−6 K
−1 [
40,
44,
45]) permits the reduction of the residual strain induced by the cooling of the system from the higher growth conditions to the room temperature [
40]. The promising performances of devices obtained by the integration of MoS
2 on GaN have been demonstrated by several research groups [
46,
47]. As an example, innovative heteroepitaxial devices have been recently reported, such as vertical heterojunction devices [
48], Esaki tunnel diodes obtained by the combination of degenerate p
+-MoS
2 on n
+-GaN/Si [
49], self-powered broadband (UV–vis–NIR) photodetectors [
50,
51,
52,
53,
54], light emitting diodes [
55], and photovoltaics applications [
56].
Nevertheless, understanding and controlling the interface properties of the MoS
2/GaN heterostructures represent the key steps to optimize the performances of demonstrated devices and, eventually, to demonstrate new ones. In fact, the interface of such vdW systems plays a crucial role in terms of electronic transport and carrier transfer. As an example, Poudel et al. reported an increase in photoluminescence (PL) emission from MoS
2 and a consequent decrease from GaN, which was attributed to electron–phonon coupling and energy transfer at the MoS
2/GaN interface [
57]. Furthermore, angle-resolved photoemission spectroscopy (ARPES) measurements performed on
n-MoS
2 flakes transferred on
p-doped GaN displayed a modification of the band structure caused by the formation of an interface dipole of 0.2 eV [
58]. Recently, Zhang et at. [
59] performed a nitridation of the GaN surface by N
2 plasma treatment before transferring MoS
2 on top of GaN. A modification of the MoS
2/GaN band structure with respect to a not-nitridated surface and a corresponding enhancement of photo-catalytic properties of the heterojunction were demonstrated, which can be exploited for hydrogen generation.
Besides experimental studies, several theoretical investigations based on the density functional theory (DFT) approach have been performed during the last few years to predict the interfacial properties and energy band-alignment in MoS
2/GaN heterostructures. Most of these simulation studies considered an ideal lattice-matched interface between monolayer MoS
2 and the Ga-terminated GaN(0001) surface, which resulted in the prediction of a covalent-like bond at the interface [
57,
59]. These theoretical results contradict the experimental evidence of a van der Waals (vdW) bond between MoS
2 and GaN, reported by different authors [
36,
37,
42]. Clearly, studies combining experimental investigations and more refined modeling of the interface are necessary to better understand the properties of this heterostructure.
In this paper, we combined experimental characterizations and DFT calculations to provide a detailed evaluation of the MoS2/GaN interface structure and the strain, do**, and the optical emission properties of MoS2 domains grown by CVD on GaN. X-ray photoelectron spectroscopy (XPS) displayed the formation of stochiometric MoS2 ([S]/[Mo] ≈ 2) without the presence of Mo-oxide (MoOx) components. Raman map** showed that the MoS2 domains mainly consisted of monolayers, with a small bilayer fraction, consistently with the intense light emission peak revealed by micro-photoluminescence (μ-PL) map**. Furthermore, an average n-type do** of (0.1–0.2) × 1013 cm−2 and a very low tensile strain of ~0.25% was evaluated by the correlative plot of the E’ and A’1 Raman peaks. The obtained strain was in perfect agreement with the one derived by the exciton peak positions obtained by μ-PL spectra. Cross-sectional scanning transmission electron microscopy (STEM) measurements confirmed both the monolayer MoS2 thickness, the presence of a van der Waals (vdW) gap at the interface with GaN, and a modification of the topmost GaN layers with respect to the bulk crystal. Finally, we employed DFT calculations to better understand the structural and electronic properties of the interface between 1L of MoS2 and GaN. In particular, three configurations of the GaN surface were considered within this heterostructure: (i) an ideal Ga-terminated GaN(0001) surface, (ii) the passivation of Ga terminations with a monolayer coverage of oxygen (O) atoms, and (iii) the presence of an ultrathin Ga2O3 film on the GaN surface. The first two configurations resulted in a strong covalent bond at the interface, very different from the experimentally observed vdW interaction. On the other hand, the formation of a vdW gap of 3.05 Å and a significant n-type do** of 1L of MoS2 was predicted in the presence of an ultrathin Ga2O3 film at the MoS2/GaN interface, which is in close agreement with the experimental results.
2. Materials and Methods
The starting material for these experiments was an unintentionally doped GaN(0001) template grown on a c-sapphire substrate. The pristine GaN surface showed a low root mean square (RMS) roughness of ~0.3 nm, evaluated from the AFM image in
Figure S1 of the Supplementary Materials.
MoS2 was grown on a GaN/c-sapphire substrate by single step CVD at a temperature of 700 °C for 10 min at atmospheric pressure. The process was carried out in a quartz tube furnace with two-heating zones, the first employed for the evaporation of the sulfur powders (7–10 mg) at 150 °C, and the second for the evaporation of the MoO3 powders (2–3 mg) at 700 °C. The GaN substrate was placed in the second zone of the furnace above the MoO3 crucible. The reaction between the S vapors (transported by an Ar flux of 100 sccm) and MoO3 vapors occurred in the gas phase close to the GaN surface, resulting in the nucleation and growth of MoS2 domains.
The MoS
2 domain coverage on GaN was evaluated by scanning electron microscopy (SEM) using a ThermoFisher Scios 2 dual-beam microscope. X-ray photoelectron spectroscopy (XPS) analysis was carried out by using Escalab ** (~10
16 cm
−3) of the GaN substrate [
36]. Focusing on the wavenumber region between 365 and 425 cm
−1 reported in
Figure 3b, a baseline subtraction and a normalization of the A
1g peak were applied with the purpose of extrapolating detailed information on the crystal quality of the CVD-grown MoS
2 flakes. Despite a low-medium A
1g/E
2g intensity ratio (~0.5), the two main Raman modes could be fitted by narrow single Gaussian peaks. In addition, the deconvolution analysis revealed the presence of a small LO(M) component near the E
2g mode, associated with defects or with the domain boundaries [
68].
To obtain statistically relevant information, a wide number of Raman spectra were collected in a 10 μm × 10 μm sample. From this array of spectra, we evaluated the homogeneity of the MoS
2 number of layers, by extracting the wavenumber difference of the A
1g and E
2g Raman modes (Δω = ω
A1g − ω
E2g), which is known to be dependent on MoS
2 thickness [
69]. In particular, the statistical distribution of the MoS
2 thickness was obtained from the Δω histogram in
Figure 3c, which shows a mean value of Δω = 20.9 cm
−1 with a standard deviation of 0.9 cm
−1. This distribution shows that the MoS
2 mainly consisted of monolayers, with a small percentage of bilayers.
In addition to the thickness assessment, the A
1g and E
2g Raman peak positions provide information on the strain and do** of the thin MoS
2 domains, according to the procedure discussed in several recent papers [
9,
35,
70,
71]. These do** and strain effects can be due to the CVD growth conditions and to the interaction with the GaN substrate.
Figure 4a shows a correlative E
2g vs. A
1g plot, where the graph is separated in four quadrants by the intersection of the ideal strain (red) and do** (black) lines. The intersection point represented by the light blue square corresponds to the ideal (E
2g, A
1g) Raman modes of unstrained and undoped monolayer MoS
2. To this aim, the literature values of the (E
2g,A
1g) peak positions (
;
) for suspended MoS
2 flakes [
5,
72] were taken as the best approximation to this ideal condition, since the effects of the interaction with substrate are excluded in that case.
The red and black arrows indicate the directions of tensile strain and
n-type do** regions, respectively, while the opposite side of the red and black lines correspond the compressive strain and
p-type do**. The
n-type do** region was indicated by the yellow area to better distinguish it from the
p-type region (white) in the upper-side of the graph. The experimental values of the peak positions from the same array of Raman spectra used in
Figure 3 are reported by the empty circles in the graph of
Figure 4a. The corresponding histograms of the E
2g and A
1g peak values are also reported on the upper-side and right-side of the graph (grey bars). In
Figure 4a, the blue and green points correspond to the peaks’ positions obtained in the 1L and 2L (or multilayer) regions, respectively, as determined in the histogram of
Figure 3c. For 1L of MoS
2, an average tensile strain of around 0.2% and light
n-type do** (<0.1 × 10
13 cm
−2) is deduced from the correlative plot in
Figure 4a. A more precise evaluation was obtained by evaluating the strain and do** for each data point of 1L-MoS
2 and by building the histograms of the strain and do** distribution, as reported in
Figure 4b,c. A tensile strain of 0.25 ± 0.10% and a
n-type do** of (0.11 ± 0.12) × 10
13 cm
−2 were deduced from the mean value and the standard deviation of these two distributions. Notably, nearly unstrained monolayer MoS
2 on GaN has been recently reported also under different CVD growth conditions, resulting in the formation of micrometer size triangular domains [
42] or continuous monolayer MoS
2 films [
36,
42]. These observations confirm the key role played by the low mismatch of the in-plane lattice constants (<1%) and of thermal expansion coefficients between MoS
2 and GaN. Furthermore, the low
n-type do** is consistent with the typically reported unintentional
n-type do** of MoS
2 obtained by exfoliation from bulk crystals, probably induced by the presence of native defects (such as sulfur vacancies) [
73,
74]. On the other hand,
n- or
p-type do** behavior has been reported for MoS
2 grown by CVD approaches, depending on several factors, such as the content of MoO
3 residues in the films (typically responsible for
p-type do**) or the peculiar interaction with the underlying substrate. In this regard, the average
n-type do** of the CVD-grown MoS
2 on GaN in the present work is consistent with the absence of MoO
3 residues, as indicated by XPS analyses in
Figure 2.
Figure 5a shows a representative PL spectrum of the MoS
2/GaN obtained with a laser excitation wavelength of 532 nm. The intense PL emission is a further confirmation of the good MoS
2 crystal quality achieved by CVD. In fact, a high density of defects in MoS
2 films would involve non-radiative recombination of excitons, causing PL quenching [
75,
76].
In detail, a deconvolution analysis performed on the spectrum of
Figure 5a revealed the coexistence of three components. The A
0 and B peaks located at 1.87 eV and 1.91 eV correspond to the excitonic emissions due to the spin-orbit coupling splitting of the MoS
2 valence band [
77]. Differently, the red-area convoluted peak at lower energy (1.79 eV) is related to the trion (also known as charged exciton) contribution, consisting of the bound state between an electron (or hole) and an exciton [
78,
79]. This deconvolution analysis confirms the absence of the defect-related peak X
D, typically located at lower energy with respect to the trionic component. After a statistical analysis of different MoS
2/GaN areas, we built a histogram of the excitonic peak energy A
0, as reported in
Figure 5b. This distribution showed a standard deviation of 10 meV around a mean value of 1.87 eV, indicating a spatially uniform PL emission from the sample surface. The energy of the main PL peak (A
0) has been shown in the literature to be dependent on the strain of MoS
2, with a red shift of the peak at increasing strain with a rate of −99 meV/% [
72]. By applying this linear relation, the values of the tensile strain were calculated from the experimental values of the A
0 peak energy, as reported in
Figure 5c. From this analysis, strain values in the range between 0.08 and 0.3% were deduced, with a mean value of ε = 0.19 ± 0.05%, in good agreement with the previous estimation by Raman measurements.
The interface properties of the 2D/3D vdW heterostructure were characterized by cross-sectional transmission electron microscopy analyses.
Figure 6a is an overview HR-TEM image, showing a monolayer MoS
2 conformal to the crystalline GaN substrate, similarly to other reports for MoS
2 grown by CVD approaches on GaN or other crystalline hexagonal substrates [
36,
37,
80,
81]. Furthermore, the presence of a vdW gap between the single layer of MoS
2 and GaN surface is clearly demonstrated by the high-resolution HAADF-STEM image in
Figure 6b. This is a direct indication of a weak bond between MoS
2 and the underlying GaN crystal. Notably, this high-resolution STEM analysis reveals a different structure of the first crystalline planes of GaN with respect to the underlying bulk crystal. As reported in previous structural investigations of CVD MoS
2/GaN heterostructures [
37], such differences can be attributed to partial oxidation of the GaN surface during the MoS
2 growth process or some form of surface reconstruction.
In the last section of this paper, DFT calculations have been performed to obtain a deeper physical understanding of the type of interaction and electronic properties of the MoS
2/GaN interface. In this context, it is worth mentioning that DFT calculations of this kind of heterostructure have been recently reported in the literature [
57,
59]. A S-Ga equilibrium distance of 0.232 nm in Ref. [
57] and 0.237 nm in Ref. [
59] was evaluated for the ideal case of a lattice-matched interface between MoS
2 and Ga-terminated GaN, indicating the formation of a covalent-like bond at the interface. Clearly, those calculation results do not match with the results of atomic resolution TEM analyses of the MoS
2/GaN heterostructure obtained in the present work and with those recently reported by different research groups [
37,
47], which showed the presence of a larger vdW gap separating S from Ga atoms.
As a matter of fact, under real experimental conditions, the GaN(0001) surface can be subjected to reconstructions or to oxidations. Hence, to provide a more complete description of the MoS
2/GaN system, we performed DFT calculations of the heterostructure considering three different model configurations of the GaN surface (see
Figure S3 of the Supplementary Materials): (i) the ideal Ga-terminated GaN, analogous to the one reported in the literature; (ii) the passivation of the Ga termination with an oxygen coverage of one monolayer; and (iii) the formation of an ultra-thin crystalline Ga
2O
3 oxide. The analysis of the DFT predictions for these three configurations has been compared with the experimental results for our system.
Figure 7a shows the most stable configuration obtained by DFT calculations of the ideal Ga-terminated GaN surface, where a Ga-S equilibrium distance of 0.241 nm was estimated, in close agreement with recent literature reports [
57,
59]. Furthermore, the calculated band structure for this system (reported in
Figure 7b) shows a high
n-type do** of MoS
2 and strong perturbation of its energy bands. As a matter of fact, this ideal configuration of the MoS
2/GaN(0001) interface does not provide a real representation of the system. For this reason, we performed DFT calculations considering the presence of O atoms bonded to the GaN(0001) surface.
Figure 8a,b shows the results for oxygen-passivated Ga terminations with one monolayer O surface coverage. Also in this case, a covalent interface interaction was obtained, which again deviates from the experimental evidence of a weak van der Waals bonding. The theoretically calculated strong interface coupling had a structural impact only on the topmost Ga layer of the substrate and perturbed the MoS
2 bands with respect to those of a freestanding MoS
2 layer (see
Figure S4 of the Supplementary Materials).
We thereon considered the formation of an ultra-thin layer of surface native oxide Ga
2O
3, which significantly reduces the surface energy of GaN(0001) as compared to other oxidized reconstructions [
82]. This layer is characterized by an O−Ga−O trilayer which inverts the polarization of the GaN layer along the [0001] direction, followed by a Ga−O bilayer that terminates the oxidized surface (
Figure 9a). The interaction of this Ga
2O
3-terminated GaN surface with MoS
2 gave rise to a van der Waals interface with an oxygen-sulfur interface distance of 3.05 Å. This distance is significantly larger than the one reported in the literature for the ideal MoS
2/GaN(0001) system [
57,
59] and it is in better agreement with the experimental observations of a vdW gap at the interface.
Figure 9b–e shows the total and partial electronic contributions (Mo d and Ga s orbitals) in the density of states of the heterosystem, plotted at the close Γ-M-K-Γ path of the Brillouin zone. The pristine bands of MoS
2 are clearly preserved in this case, showing a direct band gap of 1.7 eV at the K point of the Brillouin zone. We note that such an interface induces a significant
n-type do** for the MoS
2 sheet, due to a shift of surface Ga s states deriving from the oxide layer towards lower energies (because of Ga-O bonding). Such a shift brings the Fermi level of the system close to the conduction band of the MoS
2 layer. Overall, the theoretical calculations indicate that a van der Waals interface at the MoS
2/GaN(0001) heterosystem is expected when an ultra-thin Ga
2O
3 native oxide forms at the substrate’s surface, whereas it is rather improbable for low surface oxygen coverages.