WSe₂ homojunctions and quantum dots created by patterned hydrogenation of epitaxial graphene substrates

Two-dimensional (2D) transition metal dichalcogenides (TMDs) [1] are atomically thin semiconductors that have great potential for future electronic and optoelectronic devices [2, 3]. To achieve this, viable control of their electronic band structure is required to modulate charge carrier polarity [4–12] or facilitate low-resistance metal contacts [13, 14]. As yet, junctions based on these materials were realized by electrostatic gating [4, 11], and the fabrication of lateral [7–10] and vertical [8, 12] heterostructures. In addition, chemical doping [5, 6] was used to modify the majority carrier type and establish p-n diode characteristics.

In this Letter, we demonstrate an alternative path for creating lateral TMD homojunctions by utilizing patterned hydrogenation of the epitaxial graphene (EG) that the WSe₂ layer resides on. Using a patterned template with coexisting areas of bare and hydrogenated graphene [15], we show that abrupt and structurally perfect homojunctions can be formed. These homojunctions do not form in the same manner as a conventional p-n junction—by carrier transfer from one side of the junction to the other, and concomitant production of space charge (from dopant cores) on the two sides of the junction. Rather, the effects observed here are fully explained in terms of the electrostatic potential of the underlying graphene, with this potential then superimposed on the WSe₂ overlayer. We argue that this mechanism explains the lateral junction formation also in other related 2D heterostructure systems [16, 17] in which interlayer coupling is weak and gap states are absent. It is also shown that this approach enables one to create WSe₂ quantum dots that confine either valence band (VB) or conduction band (CB) states, paving the way for tailoring TMD-based quantum materials.

Sample preparation

STM experiment

Theoretical calculations

We used cryogenic STM at 5 K to study the MOCVD-grown WSe₂ layers on EG which was formed on SiC. As described in the preceding section, a subsequent hydrogenation process was applied to intercalate hydrogen at the SiC–graphene interface and convert the initial monolayer (1L) graphene plus carbon buffer to QF2L [20]. We performed this treatment in a way such that only partial hydrogenation occurred, that is, the graphene substrate consisted of QF2L areas coexisting with 1L areas [26]. On such patterned graphene, WSe₂ forms a continuous layer by overgrowing steps between adjacent QF2L and 1L areas. This is evident from the STM image in figure 1(a) showing QF2L islands overgrown by a single layer (SL) of WSe₂. STM imaging at atomic resolution (figure 1(b)) reveals the hexagonal atomic arrangement within the topmost Se sheet and also shows that the lateral order remains unaffected by the graphene step beneath.

Zoom In Zoom Out Reset image size

To verify that the underlying graphene substrate consists of 1L and QF2L areas, we used the STM tip to remove the WSe₂ layer in the vicinity of a step. Working on a surface area that had a continuous coverage of two layers of WSe₂ (double layer, DL), the tip was held at constant height z at the lateral position marked in figure 1(c) and a voltage pulse of 4 V applied (figure 1 in the supplementary information (SI) (stacks.iop.org/TDM/6/021001/mmedia)). Figure 1(d) demonstrates that this treatment locally desorbs the WSe₂ layer, leaving behind the bare graphene substrate with a step structure that appears to be the footprint of the initial WSe₂ surface topography (figure 1(c)). In line with previous work [27–29], STM imaging of 1L graphene at low sample bias (figure 1(d), right panel) reveals the honeycomb structure of graphene together with the 6  ×  6 reconstruction of the SiC buffer beneath. The corrugation of this reconstruction is absent on the QF2L area since there the buffer was converted to graphene by hydrogen intercalation [20]. The tip-induced desorption procedure used here is highly reproducible, allowing us to create defined pits in the WSe₂ overlayer along with atomically clean and intact graphene at the bottom.

The underlying substrate modifies the electronic properties of the WSe₂ overlayer. To quantify this, we measured the local DOS by recording the conductance dI/dV as a function of the sample bias V. We start the discussion with SL-WSe₂ covering adjacent 1L and QF2L, see the STM image in the upper panel of figure 2(a). The corresponding conductance spectrum acquired on the 1L region (red curve in figure 2(b)) indicates a bandgap of 2.05 eV between the CB and VB edges, along with VB states ~0.7 eV below the VB edge (statistical average over various independent measurements yields a bandgap of 2.00  ±  0.06 eV). These observations are consistent with prior work [22, 30] which assigned the CB edge (denoted CBM) to a combination of valleys at the Q and K points, the VB edge to a valley at the K point (K_V), and the next lowest VB edge arising from a local VB maximum (Γ_V) at the Γ point of the surface Brillouin zone. The spectrum acquired on the QF2L region (blue curve) exhibits the same spectral features, but shifted ~0.25 eV higher in energy, suggesting a rigid upward shift of the WSe₂ bands when moving from 1L to QF2L. For the DL-WSe₂ case shown in the lower panel of figure 2(a), the corresponding spectra (orange and green curves in figure 2(b)) corroborate the expected trend of a reduced bandgap of 1.83 eV (statistical average 1.85  ±  0.04 eV) and a VB splitting (Γ_V1 and Γ_V2) due to interlayer coupling [30]. Nonetheless, also here a rigid energy shift of somewhat smaller magnitude is found between spectra taken at the two different graphene areas.

Zoom In Zoom Out Reset image size

To further explore this behavior, we recorded a series of consecutive SL-WSe₂ spectra along a line spanning adjacent 1L and QF2L regions. The resulting DOS bias map (figure 2(c), center panel) demonstrates the formation of a lateral homojunction within the continuous WSe₂ layer. A uniform band-edge shift to higher energy is observed as the lateral position is changed from monolayer to free-standing bilayer graphene. The band edge offset ΔE deduced from the data in figure 2(c) is 0.24 eV. A statistical analysis of ΔE values determined from various different SL-WSe₂ junctions yields the histogram shown in figure 2(d). Clearly, besides the main maximum at 242 meV—identified as the band-edge offset of WSe₂ overlying a 1L-QF2L graphene junction as discussed before—there is a second maximum at 112 meV. We attribute this smaller ΔE value to junctions formed between adjacent bilayer (2L) and quasi-freestanding triple layer (QF3L) regions of the graphene substrate (figures 2(a) and (b) in the SI). Their existence on our samples is plausible because 2L areas formed during the initial graphene growth—which are likely to be present at small quantity even at a nominal coverage of one monolayer—will lead to 2L-QF3L junctions upon partial hydrogenation (since the buffer-layer carbon is converted to a new graphene layer upon hydrogenation).

Turning back to figure 2(c), the lower panel shows the variation of the mid-gap energy E_mid across the junction. Within the picture of a conventional p-n junction, the data would suggest electron-doped WSe₂ on 1L graphene and hole-doped character on QF2L graphene, with negligibly small doping concentrations due to the ~1 eV gap between the Fermi energy (at 0 V) and the respective band edges. At the same time, the junction is remarkably abrupt, showing a width of only ~100 Å. Clearly, this is incompatible with the picture of a conventional p-n junction for which the width of the space charge region is inverse to the carrier concentration [31].

We now discuss the origin of the lateral homojunction formation which cannot be explained by carrier depletion in a p-n junction as discussed above. Previous work has demonstrated that metal–semiconductor junctions formed by van der Waals interactions between TMDs and graphene exhibit negligible Fermi level pinning due to the absence of gap states [13, 16, 32]. Hence, a TMD layer on top of a junction formed in the graphene is expected to display a band-edge profile in the TMD that simply follows the potential profile in the graphene junction. For our experiments, EG on SiC is known to exhibit n-type character due to donorlike dangling-bond states existing at the SiC–graphene interface [15, 33, 34]. On the other hand, hydrogen-intercalated quasi-free-standing graphene has p-type character because of (i) the saturation of the dangling-bond states and (ii) the spontaneous polarization of hexagonal SiC(0 0 0 1) [15]. The carrier densities of the 1L and QF2L graphene are relatively high [33], thereby enabling an abrupt, lateral junction to form between them. Taking their vacuum levels (E_vac) as initially aligned, then within the Schottky Mott rule electrons will transfer from one material to the other until Fermi level alignment is reached. A barrier is formed in the lateral junction with height given by the work function difference between QF2L and 1L graphene, ΔE $\cong$ Φ_QF2L  −  Φ_1L, as illustrated in figure 3. Similarly, for the junction between 2L and QF3L graphene, a junction will form with barrier height of ΔE $\cong$ Φ_QF3L  −  Φ_2L. Now let us consider placing WSe₂ (SL or DL) on these 1L-QF2L or 2L-QF3L junctions. Neglecting any charge transfer between WSe₂ and graphene, then the WSe₂ bands will simply align to the laterally varying graphene band structure, i.e. with some fixed offset ΔE_C (ΔE_V) between the WSe₂ conduction (valence) band edge and the Dirac point of the graphene (as determined by ${{\chi }_{{\rm WS}{{{\rm e}}_{{\rm 2}}}}}$ the electron affinity of WSe₂, ~3.55 eV [35]). Hence, in this approximation, we expect the same barrier height in the WSe₂ lateral junctions as in the underlying graphene junctions [16]. We note at this point that the effect observed here is similar to the previously reported case of potential imprinting by charge impurities in graphene-boron nitride heterostructures [36, 37].

Zoom In Zoom Out Reset image size

Using reported graphene work function values [33], one thus expects barrier heights for the 1L-QF2L and 2L-QF3L junctions of 0.55 and 0.33 eV, respectively. It is apparent that these values exceed the experimental result in figure 2(d) by more than a factor of two. However, it is known that for the 800 °C preparation conditions of our samples, a small amount of partial hydrogenation can occur even on the surface areas that are nominally pristine 1L or 2L graphene [38, 39]. That is, not only do areas of both pristine graphene and quasi-free-standing graphene occur over the surface, but also, the nominally pristine graphene is partially hydrogenated. We tentatively attribute our reduced barrier heights to this effect, since other possible charge transfer effects between graphene and WSe₂ are expected to be relatively small (in-gap interface states are found to be absent at the graphene–WSe₂ interface [16], defect densities in the WSe₂ are low enough such that they would produce electrostatic potential shifts of 1–10 meV at most [40], and short-range interface dipoles [41] between WSe₂ and graphene are not expected to vary across the 1L-QF2L or 2L-QF3L junctions since the topmost graphene layer is chemically unchanged across the junctions).

The lateral junctions of the graphene can be utilized to confine VB and CB states within the TMD layer. Consider, for example, the surface area overgrown by a SL-WSe₂ layer as shown in figure 4(a): following the preceding discussion, along the dotted line in figure 4(a) one expects a spatial variation of the band-edge energies as sketched in figure 4(b). The corresponding DOS bias map (figure 4(c)) measured along the dotted line reveals the resulting band-edge modulation and suggests that states near the VB edge will be confined to the QF2L region. Consistent with this expectation, interference patterns of electron waves backscattered [42] at graphene steps are observed when probing states near the VB edge (figure 3 in the SI).

Zoom In Zoom Out Reset image size

Confinement in sufficiently small islands leads to a measurable discretization of electronic states, thus establishing the formation of a TMD quantum dot [43]. Figure 4(d) shows such a case of a compact QF2L island overgrown by a DL-WSe₂ layer. To visualize the nodal structure of the confined states, spatial DOS maps were recorded (figure 4(e)) showing quasi-hexagonal wave patterns due to the regular shape of the dot. Fourier transform power spectra of these spatial DOS maps yield peaked wave vector components along the Γ–K direction of the surface Brillouin zone (figure 4 in the SI); they are plotted versus energy in figure 4(f). The experimental k_Γ–K values agree well with the VB dispersion (green line) calculated within DFT (figure 5 in the SI). In addition, the data points in figure 4(f) indicate that the confinement imposes wave vector quantization with a mean Δk value of 0.027 Å⁻¹, see figure 4 in the SI. Given the boundary condition k_n  =  nπ/L for confinement, with n an integer and L the dimension of the confining potential, we find L  =  π/Δk $\cong$ 116 Å which fits well with the size of the dot shown in figure 4(d). Conversely to the VB state confinement in figure 4, we find that CB states are confined in overgrown 1L patches embedded in QF2L regions (figure 6 in the SI).

In summary, the band-edge energies of an atomically thin TMD overlayer can be precisely modulated by lateral junction formation in a weakly coupled graphene layer, allowing one to create structurally perfect homojunctions of small lateral size and free of gap states. The modulation arises from the varying surface potential across the graphene which is directly superimposed as a bulk potential in the overlying TMD. We argue that this mechanism also explains the band-edge modulation recently observed in WSe₂, MoSe₂, and Te layers on graphene of varying thickness [16, 17]. The lateral junction formation that occurs in the 2D system studied here (and likely in other systems in which interlayer coupling is weak and gap states are absent [13]) is qualitatively different to what occurs in conventional p-n junctions, the latter implying carrier transfer and the production of a space charge region. It is anticipated that the 2D junctions studied in our work could be utilized for lateral carrier transport by adding a h-BN spacer between the graphene template and the TMD overlayer. Given the use of hydrogenated graphene, possibly electron beam lithography [44] could be employed to selectively desorb the hydrogen, thus writing a surface pattern that would be superimposed as a potential pattern in the 2D layer on top. This would represent a versatile approach to spatially modulate carriers in two dimensions, providing a test ground for fundamental research on quantum-confined electrons and a TMD-based material platform for 2D electronic circuits.

Y Pan and S Fölsch acknowledge financial support by the Deutsche Forschungsgemeinschaft (SFB 658). Y Pan also acknowledges the support from the National Key R&D Program of China (2017YFA0206202) and the National Science Foundation of China (11704303). Y-C Lin, B Jariwala, J A Robinson, and R M Feenstra acknowledge the support from the Center for Low Energy Systems Technology (LEAST). LEAST is one of six Semiconductor Research STARnet centers sponsored by MARCO and DARPA. Y Nie and K Cho were also supported in part by ASCENT, one of six centers in JUMP, a SRC program sponsored by DARPA.