Heterostructures and electronic devices derived therefrom

Abstract

The advent of graphene and related 2D materials has recently led to a new technology: heterostructures based on these atomically thin crystals. The paradigm proved itself extremely versatile and led to rapid demonstration of tunnelling diodes with negative differential resistance, tunnelling transistors, photovoltaic devices and so on. In the present invention, the complexity and functionality of such van der Waals heterostructures is taken to the next level by introducing quantum wells (QWs) engineered with one atomic plane precision. We describe light-emitting diodes (LEDs) made by stacking metallic graphene, insulating hexagonal boron nitride and various semiconducting monolayers into complex but carefully designed sequences.

Claims

1. A graphene-based vertical heterostructure comprising at least the following layers in sequence: a first graphene layer comprising graphene or modified graphene; a first insulating layer; a first TMDC layer; a second insulating layer; a second TMDC layer; a third insulating layer; and a second graphene layer comprising graphene or modified graphene; wherein the layers are stacked to form a laminate structure.

2. A heterostructure as claimed in claim 1, wherein at least one additional layer is present on a surface of the first graphene layer opposite the first insulating layer.

3. A heterostructure as claimed in claim 2, wherein the at least one additional layer is selected from the group consisting of hexagonal boron nitride (hBN), SiO.sub.2 and Si.

4. A heterostructure as claimed in claim 1, wherein at least one additional layer is present on a surface of the second graphene layer opposite the third insulating layer.

5. A heterostructure as claimed in claim 1, wherein the heterostructure is mounted on a substrate.

6. A heterostructure as claimed in claim 1, wherein one or more component layers of the heterostructure is formed of a single crystal.

7. A heterostructure as claimed in claim 1, wherein the graphene of at least one of the first graphene layer and the second graphene layer is pristine graphene.

8. A heterostructure as claimed in claim 1, wherein the graphene of at least one of the first graphene layer and the second graphene layer is chemically modified graphene.

9. A heterostructure as claimed in claim 1, wherein the TMDC is selected from the group consisting of: MoS.sub.2, WS.sub.2, WSe.sub.2, MoSe.sub.2, MoTe.sub.2, and WTe.sub.2.

10. A heterostructure as claimed in claim 1 wherein at least one insulating layer comprises hBN.

11. A heterostructure as claimed in claim 1 wherein each insulating layer comprises hBN.

12. A heterostructure as claimed in claim 1 wherein at least one insulating layer consists essentially of hBN.

13. A heterostructure as claimed in claim 1 wherein each insulating layer consists essentially of hBN.

14. A graphene-based vertical heterostructure comprising at least the following layers in sequence: a first graphene layer comprising graphene or modified graphene; a first insulating layer; a first TMDC layer; a second insulating layer; a second TMDC layer; a third insulating layer; a third TMDC layer; a fourth insulating layer; and the second graphene layer comprising graphene or modified graphene; wherein the layers are stacked to form a laminate structure.

15. A graphene-based vertical heterostructure comprising at least the following layers in sequence: a first graphene layer comprising graphene or modified graphene; a first insulating layer; a first TMDC layer; a second insulating layer; a second TMDC layer; a third insulating layer; a third TMDC layer; a fourth insulating layer; a fourth TMDC layer; a fifth insulating layer; and a second graphene layer comprising graphene or modified graphene; wherein the layers are stacked to form a laminate structure.

16. A graphene-based vertical heterostructure comprising at least the following layers in sequence: a first graphene layer comprising graphene or modified graphene; a first insulating layer; a first TMDC layer; a second insulating layer; a second TMDC layer; a third insulating layer; a third TMDC layer; a fourth insulating layer; a fourth TMDC layer; a fifth insulating layer; a fifth TMDC layer; a sixth insulating layer; and a second graphene layer comprising graphene or modified graphene; wherein the layers are stacked to form a laminate structure.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) Embodiments of the invention are further described hereinafter with reference to the accompanying drawings.

(2) FIG. 1 schematically shows the architecture of single.sub.[sH1]-quantum-well (SQW) and MQW structures along with optical images of a typical device (FIG. 1e). FIG. 1a, represents a schematic diagram of the SQW heterostructure hBN/GrB/2hBN/WS2/2hBN/GrT/hBN. FIG. 1b is a cross-sectional bright-field STEM image of the type of heterostructure presented in a. Scale bar, 5 nm. FIGS. 1c and d represent schematic and STEM images of the MQW heterostructure hBN/GrB/2hBN/MoS2/2hBN/MoS2/2hBN/MoS2/2hBN/MoS2/2hBN/GrT/hBN. The number of hBN layers between MoS2 QWs in FIG. 1d varies. Scale bar, 5 nm. FIG. 1e shows an optical image of an operational device (hBN/GrB/3hBN/MoS2/3hBN/GrT/hBN). The dashed curve outlines the heterostructure area. Scale bar, 10 m. FIG. 1f shows an optical image of EL from the same device. Vb=2.5V, T=300 K. 2hBN and 3hBN stand for bi- and trilayer hBN, respectively. FIG. 1g shows a schematic of another heterostructure according to the invention consisting of Si/SiO2/hBN/GrB/3hBN/MoS2/3hBN/GrT/hBN. FIGS. 1h-j are band diagrams for the case of zero applied bias (h), intermediate applied bias (i) and high bias (j) for the heterostructure presented in FIG. 1g.

(3) FIG. 2 shows optical and transport characterisation of a SQW devices, T=7K. FIG. 2a illustrates a colour map of the PL spectra as a function of Vb for a MoS2-based SQW. The white curve is the dl/dVb of the device. Excitation energy EL=2.33 eV. FIG. 2b shows EL spectra as a function of Vb for the same device as in FIG. 2a. White curve: its j-Vb characteristic (j is the current density). FIG. 2c illustrates a comparison of the PL and EL spectra for the same device. As PL and EL occur in the same spectral range, we measured them separately. FIGS. 2d-g illustrate the same as in FIGS. 2b and c but for the bilayer (FIGS. 2d and e) and monolayer (FIGS. 2f and g) WS2 QWs. The PL curves were taken at Vb=2.4V (FIG. 2c), 2.5 V (FIG. 2e) and 2 V (FIG. 2g); the EL curves were taken at Vb=2.5 V (FIG. 2c), 2.5 V (FIG. 2e) and 2.3 V (FIG. 2g).

(4) FIG. 3 shows optical and transport characteristics of MQW devices, T=7K. FIG. 3a shows the modulus of the current density through a triple QW structure based on MoS2. FIG. 3b shows its schematic structure. FIGS. 3c and d show maps of PL and EL spectra for this device. EL=2.33 eV. FIG. 3e shows individual EL spectra plotted on a logarithmic scale which show the onset of EL at 1.8 nA m-2 (blue curve). Olive and red: j=18 and 130 nA m-2, respectively. FIG. 3f is a comparison of the EL (taken at Vb=8.3 V) and PL (taken at Vb=4.5 V) spectra.

(5) FIG. 4 shows devices combining different QW materials and on flexible substrates. FIGS. 4a-c show EL at negative (a) and positive (c) bias voltages for the device with two QWs made from MoS2 and WSe2 schematically shown in the inset in d. Its PL bias dependence is shown in FIG. 4b, for laser excitation EL=2.33 eV, T=7 K. White curve: |j|-Vb characteristics of the device. FIG. 4d shows the temperature dependence of EQE for a device with two QWs made from MoS2 and WSe2. Inset: schematic representation of a device with two QWs produced from different materials. FIG. 4e shows an optical micrograph taken in reflection mode of a SQW (MoS2) device on PET. FIG. 4f shows an optical micrograph of the same device as in FIG. 4e taken in transmission mode. For FIGS. 4e and f, the area of the stack is marked by red rectangles; scale bars are 10 m. g, EL spectra for the device in FIGS. 4e and f at zero (blue dots) and 1% (red dots) strain. Vb=2.3 V, I=40 A at room T.

(6) FIG. 5 shows a WSe2 Quantum well structure. (A) Schematic of WSe.sub.2 singe quantum well. (B) Band alignment diagram for the QW LED shown in (A). (C) cross sectional scanning transmission electron microscope (STEM) high angle annular dark field (HAADF) image of the QW LED. (D) Electron energy loss spectroscopy (EELS) chemical map of nitrogen and selenium confirming device structure. Scale bar 5 nm.

(7) FIG. 6 shows (A) Contour map of the electroluminescence for negative bias with current density plotted against bias (right axis) (B) Contour map of the photoluminescence spectra for different bias voltage with the modulus of the current density plotted (right axis) (P=10 uW, E=2.33 eV) (C) Contour map of the electroluminescence for positive bias voltage with current density plotted (right axis).

(8) FIG. 7 shows (A) current density of j=(D) Ratio of the integrated electroluminescence intensity from T=6 K to T=300 K.

(9) FIG. 8 shows (A) Temperature dependence of the quantum efficiency for 6 separate WSe.sub.2 single quantum wells measured from T=6 K to T=300 K. (B) Room temperature bias dependence of the electroluminescence spectra measured from Vb=1.3V to 2.3V. (C) (Left/bottom axis) bias voltage dependence of the quantum efficiency (right/top axis) current density dependence of the quantum efficiency.

DETAILED DESCRIPTION

(10) The term vertical heterostructure refers to a plurality of two-dimensional crystals arranged in a stack. A heterostructure comprises at least two different materials. The two-dimensional crystals are arranged such that the heterostructures are substantially parallel, being arranged face-to-face, forming a laminate. Such heterostructures may also be called two-dimensional heterostructures.

(11) For the purposes of the present invention, the heterostructure is formed entirely of two-dimensional crystals. This does not preclude the heterostructure from being mounted on a substrate and/or have a protective coating. A two-dimensional heterostructure is so-called because it is comprised of two-dimensional crystals. It will itself, of course, be a three dimensional structure.

(12) Examples of two-dimensional crystals which may be included in the heterostructures of the invention include graphene, modified graphene (e.g. doped graphene, graphane, fluorographene, chlorinated graphene), BN, MoS.sub.2, NbSe.sub.2, Bi.sub.2Te.sub.3, MgB.sub.2, WS.sub.2, MoSe.sub.2, TaSe.sub.2, NiTe.sub.2.

(13) Heterostructures may be formed by placing two-dimensional crystals upon one another mechanically, epitaxially, from solution and/or using any other means which would be apparent to the person skilled in the art.

(14) A graphene heterostructure comprises at least layers of a two-dimensional crystal of graphene or modified graphene.

(15) The term two dimensional crystal means a crystal which is so thin that it exhibits different properties than the same material when in bulk. Not all of the properties of the material will differ between a two-dimensional crystal and a bulk material but one or more properties are likely to be different. A more convenient definition would be that the term two-dimensional crystal refers to a crystal that is 10 or fewer molecular layers thick, e.g. one molecular layer thick, but this depends on the material. Crystals of graphene which have more than 10 molecular layers (i.e. 10 atomic layers) generally exhibit properties more similar to graphite than to graphene. A molecular layer is the minimum thickness chemically possible for that material. In the case of graphene one molecular layer is a single atom thick. The same is true of boron-nitride. In the case of the transition metal dichalcogenides (e.g. MoS.sub.2 and WS.sub.2), a molecular layer is three atoms thick (one transition metal atom and two chalcogen atoms). Thus, two-dimensional crystals are generally less than 50 nm thick, depending on the material and are preferably less than 20 nm thick. Graphene two-dimensional crystals are generally less than 3.5 nm thick and may be less than 2 nm thick.

(16) The term two dimensional crystal includes crystals which are doped, as described below.

(17) The term modified graphene refers to a graphene-like structure that has been modified in some way. Thus, the modified graphene may be graphene which has been doped. This may have the purpose of modifying the work function of graphene without significantly reducing its conductivity. Examples of compounds which can be used to dope graphene are: NO.sub.2, H.sub.2O and I.sub.2, which act as acceptors to provide a p-doped graphene; or NH.sub.3, CO and C.sub.1-C.sub.3 alcohols (e.g. ethanol), which act as donors to provide an n-doped graphene. Small amounts of doping can increase the transparency of the doped graphene relative to graphene but the dopant itself may absorb or reflect light. Conventional methods of doping the graphene can be used to improve the functionality of the graphene, including its transparency to actinic radiation. These methods of doping are described in the literature and are not therefore reproduced here. An alternative approach to doping is to place metal (e.g. gold) nanostructures on the surface of the graphene. This will both dope the graphene and increase the local electric field. A preferred dopant is one which is not chemically bonded to graphene but which is able to transfer charge to graphene, effectively altering the graphene's work function.

(18) When graphene is placed in contact with a transition metal dichalcogenide layer, there is a dopant effect, i.e. the graphene's work function is changed.

(19) As used in this specification, a layer of a material refers to a plane of that material. Each layer may comprise any number of molecular layers of the same chemical composition. Thus a layer of graphene does not necessarily mean a graphene monolayer, although it might. Likewise, a layer of WS.sub.2 does not necessarily refer to a WS.sub.2 monolayer, although it might. In many embodiments of this invention, a layer of any material means a two dimensional crystal of that material.

(20) FIG. 1 schematically shows the architecture of single-quantum-well (SQW) and MQW structures along with optical images of a typical device (FIG. 1e). We used a peel/lift van der Waals technique to produce the devices of the invention. In total we measured more than a dozen of such QW structures comprising single and multiple layers of TMDC flakes from different materials: MoS2, WS2 and WSe2. The yield was 100% with every device showing strong EL that remains unchanged after months of periodic measurements, which demonstrates the robustness of the technology and materials involved.

(21) Cross-sectional bright-field scanning transmission electron microscope (STEM) images of our SQW and MQW devices demonstrate that the heterostructures are atomically flat and free from interlayer contamination (FIG. 1b,d). The large atomic numbers for TMDCs allow the semiconductor crystals to be clearly identified owing to strong electron-beam scattering (dark contrast observed in FIG. 1b, d). Other layers were identified by energy-dispersive X-ray spectroscopy. The large intensity variation partially obscures the lattice contrast between adjacent layers but, despite this, the hBN lattice fringes can clearly be seen in FIG. 1b, d. The different contrast of the four MoS2 monolayers in the MQW of FIG. 1d is attributed to their different crystallographic orientations (confirmed by rotating the sample around the heterostructure's vertical direction, which changes the relative intensity of different layers).

(22) For brevity we concentrate on current-voltage (I-V) characteristics, photoluminescence (PL) and EL spectra from symmetric devices based on MoS2 (FIG. 2a-c). Devices based on WS2 and devices with asymmetric barriers were also investigated.

(23) At low Vb, the PL in FIG. 2a is dominated by the neutral A exciton, X0, peak at 1.93 eV. We attribute the two weaker and broader peaks at 1.87 and 1.79 eV to bound excitons. At a certain Vb, the PL spectrum changes abruptly with another peak emerging at 1.90 eV. This transition is correlated with an increase in the differential conductivity (FIG. 2a). We explain this transition as being due to the fact that at this voltage the Fermi level in the bottom graphene electrode (GrB) rises above the conduction band in MoS2, allowing injection of electrons into the QW (FIG. 1i). This allows us to determine the band alignment between the Dirac point in graphene and the bottom of the conductance band in MoS2: the offset equals half of the bias voltage at which the tunnelling through states in the conductance band of MoS2 is first observed. To take into account the effects of possible variance in the thickness of hBN barriers and small intrinsic doping of graphene, we average the onset of tunnelling through MoS2 for positive and negative bias voltages (FIG. 2a), which yields the offset to be 0.5 eVin agreement with theoretical prediction.

(24) The injection of electrons into the conduction band of MoS2 leads not only to an increase in tunnelling conductivity but, also, to accumulation of electrons in MoS2 and results in formation of negatively charged excitons, X. The X peak is positioned at a lower energy compared with the X0 peak owing to the binding energy, EB, of X. In the case of MoS2 we estimate EB as 36 meV near the onset of X. As the bias increases, the energy of the X peak shifts to lower values, which can be attributed either to the Stark effect or to the increase in the Fermi energy in MoS2.

(25) In contrast to PL, EL starts only at Vb above a certain threshold (FIG. 2b). We associate such behaviour with the Fermi level of the top graphene (GrT) being brought below the edge of the valence band so that holes can be injected into MoS2 from GrT (in addition to electrons already injected from GrB) as sketched in FIG. 1j. This creates conditions for exciton formation inside the QW and their radiative recombination. We find that the EL frequency is close to that of PL at Vb2.4 V (FIG. 2a-c), which allows us to attribute the EL to radiative recombination of X. Qualitatively similar behaviour is observed for WS2 QWs (FIG. 2d-g).

(26) An important parameter for any light-emission device is the QE defined as =N2e/I (here e is the electron charge, N is the number of the emitted photons and I is the current). For SQWs we obtain quantum efficiencies of 1%this value by itself is ten times larger than that of conventional planar p-n diodes and 100 times larger than EL from Schottky barrier devices. Our rough estimations show that the external QE (EQE) for PL is lower than that for EL. Relatively low EQE found in PL indicates that the crystal quality itself requires improvement and that even higher EQE in EL may then be achieved.

(27) To enhance QE even further, we have employed multiple QWs stacked in series, which increases the overall thickness of the tunnel barrier and enhances the probability for injected carriers to recombine radiatively. FIG. 3 shows the results for one of such MQW structures with three MoS2 QWs (layer sequence: Si/SiO2/hBN/GrB/3hBN/MoS2/3hBN/MoS2/3hBN/MoS2/3hBN/GrT/hBN) and another MQW with four asymmetric MoS2 QWs (FIG. 1c,d) was also investigated. The current increases with Vb in a step-like manner, which is attributed to sequential switching of the tunnelling current through individual MoS2 QWs. PL for the MQW device is qualitatively similar to that of SQW devices but the X0 peak is replaced with a X peak at Vb=0.4 V (FIG. 3c).

(28) The X0 peak reappears again at Vb>1.2 V. This can be explained by charge redistribution between different QWs. The EL first becomes observable at Vb>3.9 V and j of 1.8 nA m2 (FIG. 3d,e). This current density is nearly 2 orders of magnitude smaller than the threshold current required to see EL in similar SQWs. Importantly, the increased probability of radiative recombination is reflected in higher QE, reaching values of about 8.4% (for the device with quadruple QW, 6% for triple). This high QE is comparable to the efficiencies of the best modern-day organic LEDs (ref. 29).

(29) The described technology of making designer MQWs offers the possibility of combining various semiconductor QWs in one device. FIGS. 4a-c describes an LED made from WSe2 and MoS2 QWs: Si/SiO2/hBN/GrB/3hBN/WSe2/3hBN/MoS2/3hBN/GrT/hBN. EL and PL occur here in the low-E part of the spectra and can be associated with excitons and charged excitons in WSe2. However, in comparison with SQW devices, the combinational device in FIG. 4 exhibits intensities more than an order of magnitude stronger for both PL and EL, yielding about 5% QE. We associate this with charge transfer between the MoS2 and WSe2 layers such that electron-hole pairs are created in both layers but transfer to and recombine in the material with the smaller bandgap. Such a process is expected to depend strongly on band alignment, which is controlled by bias and gate voltages. This explains the complex, asymmetric Vb dependence of PL and EL in FIG. 4.

(30) Generally, the fine control over the tunnelling barriers allows a reduction in the number of electrons and holes escaping from the quantum well, thus enhancing EQE. EQE generally demonstrates a peak at T around 50-150 K, depending on the material. Depending on the particular structure we found that typical values of EQE for MoS2- and WS2-based devices at room T are close or a factor of 2-3 lower than those at low T (FIG. 4d).

(31) Finally, we note that because our typical stacks are only 10-40 atoms thick, they are flexible and bendable and, accordingly, can be used for making flexible and semi-transparent devices. To prove this concept experimentally, we have fabricated a MoS2 SQW on a thin PET (polyethylene terephthalate) film (FIG. 4e,f). The device shows PL and EL very similar to those in FIG. 2a-c. We also tested the device's performance under uniaxial strain of up to 1% (using bending) and found no changes in the EL spectrum (FIG. 4g).

(32) In summary, we have demonstrated band-structure engineering with one atomic layer precision by creating QW heterostructures from various 2D crystals including several TMDCs, hBN and graphene. Our LEDs based on a single QW already exhibit QE of above 1% and line widths down to 18 meV, despite the relatively poor quality of available TMDC layers. This EQE can be improved significantly by using multiple QWs. Consisting of 3 to 4 QWs, these devices show EQEs up to 8.4%. Combining different 2D semiconductor materials allows fine-tuning of the emission spectra and also an enhanced EL with a quantum yield of 5%. These values of QE are comparable to modern-day organic LED lighting and the concept is compatible with the popular idea of flexible and transparent electronics. The rapid progress in technology of chemical vapour deposition growth will allow scaling up of production of such heterostructures.

(33) Methods

(34) Sample fabrication. Flakes of graphene, hBN and TMDCs are prepared by micromechanical exfoliation of bulk crystals. Single- or few-layer flakes are identified by optical contrast and Raman spectroscopy. Heterostructures are assembled using the dry peel/lift method. Electrical contacts to the top and bottom graphene electrodes are patterned using electron-beam lithography followed by evaporation of 5 nm Cr/60 nm Au.

(35) Electrical and optical measurements. Samples are mounted within a liquid helium flow cryostat with a base temperature of T=6 K. Electrical injection is performed using a Keithley 2400 source meter. To measure PL the samples were excited with a continuous wave 532 nm laser, focused to a spot size of 1 m through a 50 objective (NA=0.55) at a power less than required to modify the spectral line shape. The signal was collected and analysed using a single spectrometer and a nitrogen cooled CCD (charge-coupled device).

(36) Scanning transmission electron microscopy. STEM imaging was carried out using a Titan G2 probe-side aberration-corrected STEM operating at 200 kV and equipped with a high-efficiency ChemiSTEM energy-dispersive X-ray detector. The convergence angle was 19 mrad and the third-order spherical aberration was set to zero (5 m). The multilayer structures were oriented along the hkl0 crystallographic direction by taking advantage of the Kikuchi bands of the silicon substrate.

(37) In another embodiment of the invention, we also investigated high efficiency quantum well LEDs based on WSe2 monolayers. Thus in addition to the findings above, we have separately also shown that WSe2 single quantum wells exhibit an unusual temperature dependence in the electroluminescence quantum efficiency. Surprisingly, we have found the EL quantum efficiency to increase in some samples by 2 orders of magnitude when the temperature is increased from T=6 K to T=300 K. The room temperature quantum efficiency approaches 20% which is comparable to current LED lighting. Unlike conventional LED devices our WSe2 LED's show no drop off in the emission efficiency up to 1000 A/cm2. Again, this is an unexpected development and means that such devices could pave the way towards ultra-bright flexible lighting, 2D lasers and future near infra-red communication devices.

(38) As with the heterostructures above, the devices of this embodiment of the invention are carefully fabricated by mechanically transferring individually exfoliated flakes of graphene, few layer hexagonal boron nitride (hBN) and WSe2 monolayers into a quantum well architecture.

(39) More particularly, we found that QW's consisting of single layer WSe2 show 2-orders of magnitude brighter electroluminescence (EL) than the other studied TMDC's such as MoS2, WS2 and MoSe2 and that the quantum efficiency of the EL process increases 250 times to nearly 20% when increasing the temperature from T=6 K to T=300 K, this quantum efficiency is comparable to current commercial LED lighting and such a temperature dependence of the emission efficiency has not been reported elsewhere in any other system. Furthermore we also observed that some devices with thinner hBN tunnel barriers exhibit an absence of droop in the light output vs injection current up to a maximum achieved current density of 1000 A/cm2.

(40) FIG. 5A shows the devices structure. We make use of graphene as a transparent and conductive window which can inject either electrons or holes due to its low intrinsic doping levels, hBN as an atomically flat and defect free tunnel barrier and single layer WSe2 as the semiconducting element of the quantum well. The light emission process occurs when both electrons and holes are simultaneously injected into the conduction band and valence band respectively of the WSe2 layer through the thin hBN tunnel barrier FIG. 5B. This can only occur when a significant threshold bias is applied across the graphene electrodes.

(41) Processing contamination has been found to limit the performance of Van der Waals heterostructure devices so to confirm the geometry of our devices and to access contamination levels within our devices we take a cross sectional slice through the heterostructure stack and image with atomic resolution using scanning transmission electron microscopy (STEM). FIG. 5C shows the bright field cross sectional image of one of our devices showing individual atomic layers of h-BN and a central WSe2 monolayer encapsulated by hBN. Electron energy loss spectroscopy (EELS) mapping, FIG. 5D confirms the presence of nitrogen (corresponding to regions of hBN) and selenium (corresponding to regions of WSe2) confirming the formation of atomically flat and clean interfaces and the encapsulation between few layer hBN tunnel barriers (See methods and Supplementary for more details on cross sectional imaging).

(42) We study the PL and EL properties of these QW LED's by utilising a variable temperature flow cryostat with a base temperature of T=6 K (See methods for measurement details) We start by describing the low temperature (T=6 K) photoluminescence and electroluminescence properties of a typical WSe2 quantum well with device structure Grb2L hBN1L WSe2-2L hBN-Grt. FIG. 6(B) shows the bias voltage dependence of the photoluminescence. Our devices show a pronounced peak in the PL at an energy of E=1.70 eV which we attribute to the neutral exciton, the linewidth is found to be 16 meV with a poorly resolved charged exciton state at a lower energy of 1.68 eV with a linewidth of 40 meV. This linewidth is broader than reported elsewhere and can be attributed to inhomogeneous broadening and depends on the source of the material, however similar behaviour can be observed from cleaner WSe2 from different sources (See supplementary information).

(43) The PL line shape shows insensitivity to applied bias voltage only changing in intensity for large bias voltages. As the bias voltage is increased to Vb=2.0 V the PL increases rapidly due to the emergence of electroluminescence, which is clearly seen when the excitation laser is switched off and only EL is collected Figure (A,C). Electroluminescence occurs when the Fermi level of one of the graphene electrodes is coincident with or above the energy of the conduction band in the TMDC and the Fermi level of the other graphene electrode is coincident or below the valence band thus enabling the simultaneous injection of electrons and holes which form excitons and decay releasing photons.

(44) One major limitation of conventional light emission devices is the temperature dependence of the quantum efficiency. In many such devices, quantum efficiency drops by a factor of 10 from low temperature to room temperature. This is commonly caused by the high temperature ionisation of impurities due to crystal growth defects and other defects. The resultant charged impurity left behind can acts as a scattering centre which leads to the increase of non-radiative recombination and thus a reduced efficiency.

(45) Unique to the WSe2 QW we observe in some cases a 250 fold increase in the quantum efficiency as the temperature is increased from T=6 K to T=300 K.

(46) We now describe some particularly interesting results in relation to tungsten-based TMDCs.

(47) The external quantum efficiency (EQE) demonstrated by many conventional van der Waals LEDs is of the order of 1%, and is even smaller for planar devices. We have found that certain tungsten-based TMDCs (such as WSe.sub.2 and WS2), when used as emitting layer in our vertical LED, can offer EQE which increase with temperature and which can easily reach 20% at room temperature. This makes such devices potentially interesting for real life applications. We think that such behaviour is a consequence of the peculiar band structure of tungsten-based TMDC, which have long-living dark excitons as the ground state when incorporated in one of the heterostructures of the invention. Thus, another embodiment of the invention is the provision of a van der Waals structure i.e. a heterostructure such as those described above, in which the external quantum efficiency is greater than 10% and more preferably is greater than 20%.

(48) The strong spin-orbit interaction in W-based TMDCs leads to the lowest energy states in the conductance band and the highest energy states in the valence band to have opposite spin orientation. The interesting effects which we have observed in electroluminescence (EL) for WSe.sub.2 or WS.sub.2 LEDs, arises because electrons and holes are injected separately thereby creating an electron-hole imbalance and giving rise to new channels of exciton recombination.

(49) We have prepared vertical LED van der Waals heterostructures using WSe.sub.2 and WS.sub.2 as the light emitting QW. The structures consist of a monolayer of W-based TMDC separated from graphene electrodes by thin (2-3 monolayers) hBN barriers. This is shown in FIGS. 5C and 5F (the hBN spacers are required to control the lifetime of the charge carriers inside of the QW to allow efficient radiative recombination). The stacks were constructed via a multiple peel/lift procedure making use of the van der Waals interaction between neighbouring crystals. The high quality of the samples is confirmed by cross-sectional TEM measurements, see FIG. 5D, which demonstrate the absence of contamination between the layers.

(50) By applying bias voltage, V.sub.b, between the two graphene electrodes, it is possible to set their Fermi levels in such a way that electron (hole) injection into the conductance (valence) band of TMDC QW occurs, see FIG. 5E. If the dwell time of the quasiparticles in the QW is long enough (controlled by the thickness of hBN) then they can form excitons and recombine with light emission, as shown in FIG. 5F.

(51) FIG. 6A shows PL and EL for one of our WSe.sub.2-based samples. It clearly shows 3 peaks for near-zero V.sub.b, which we identify, as neutral exciton X1.72 eV, charged exciton X.sup.1.70 eV, and localised exciton at 1.67 eV. The amplitude of the peaks depends in a complex way on the V.sub.b, but the general trend is the decrease of the PL intensity with applied bias.

(52) At I V.sub.b|>2V the luminescence is dominated by EL signal. Typically, in most of the.sub.[sH2] samples, the EL signal at sufficiently high V.sub.b is dominated by X.sup. peak, which shifts to lower energy with bias. Most interesting, however, is the temperature behaviour of the EL.

(53) We have found that there is a growth in EL.sub.[SH3] by a factor of 300 from helium to room temperature as is shown in FIGS. 7A and 7C. With temperature, all three peaks usually merge into one, with X.sup. probably dominating. We have found that similar behaviour is observed in LEDs based on WS.sub.2 QW. The strong increase of the EL at room temperature gives rise to a large room temperature external quantum efficiency, which can reach 20% or more in some of our samples. This makes W-containing TMDCs very promising materials for future thin film, transparent and flexible LEDs.

(54) Such behaviour is observed also in our LEDs based on WS.sub.2. At the same time, LEDs based on Mo-containing TMDCs (see FIGS. 7B, 7C, and 7D) demonstrate a strong decrease of the EL with temperature. Thus, we suggest that the mechanism for the unusual T-dependence of EL in W-based LEDs probably lies in the specific band-structure of W-containing TMDCs.

(55) We measured the activation temperature of EL for LEDs based on Mo- and W-containing TMDCs, as is shown in FIGS. 7E and 7F. The results are strikingly different, with the intensity of EL increasing exponentially with T for W-based LED and decreasing for Mo-containing LEDs. The extracted activation temperature for W-based LEDs is of the order of 30 meV.

(56) The exponential increase in EL with temperature opens very interesting opportunities for W-based TMDC to be used for LED applications. The most important parameter for such applications is the EQE. FIG. 8A shows typical behaviour of the EQE for three typical devices. The temperature dependence of the quantum efficiency, QE=2N/j of the WSe2 LEDs always shows the characteristic increase with temperature reaching nominal values of 10-20% for single QW LED's a factor 100 improvement as compared to MoX.sub.2 based TMDC's.

(57) Another interesting property is the persistent high quantum efficiency at high electric fields of 10.sup.7 Vcm.sup.1 and current density of order 1000 A/cm.sup.2. A common disadvantage of commercial and domestic LED lighting is the droop effects at high injection current due to increased non-radiative scattering mechanisms and also heating effects which limit quantum efficiency. Our devices however get brighter at higher temperature and the efficiency remains high even at extremely high current densities. Improvements to crystal quality of the TMDC and reduction of graphene lead resistance could be expected to increase the quantum efficiency even further.

(58) These heterostructures were produced as follows. Firstly, bulk hexagonal boron nitride hBN is mechanically cleaved and exfoliated onto a freshly cleaned Si/SiO.sub.2 substrate. After this a graphene flake is peeled from a PMMA membrane onto the hBN crystal followed by a thin hBN tunnel barrier then a hBN tunnel barrier on PMMA is used to lift a WSe.sub.2 or MoSe2 single layer from a second substrate then both of these crystals are together peeled off the PMMA onto the hBN/Gr/hBN stack forming hBN/Gr/hBN/WX2/hBN. Finally the top graphene electrode is peeled onto the stack thus completing the LED structure. After the stack is completed we either follow standard micro fabrication procedures for adding electrical contacts to the top and bottom graphene electrodes or the whole stack is transferred onto highly reflective distributed Bragg reflector substrate where we are able to collect up to 30% of the emitted light from the LED opposed to just 2% from the Si/SiO.sub.2 substrate.

(59) It can thus be seen that the heterostructures of the present invention offer significant advantages in terms of electroluminescent efficiencies and/or quantum efficiencies and represent a potentially valuable materials for fabricating electronic devices.