Main

Atomically thin vdW materials have become a very attractive platform for realizing a plethora of fundamental phenomena and technological innovations owing to their highly desirable electrical, optical, mechanical and thermal properties. Of these vdW materials, TMDs have become extremely attractive for optoelectronics owing to their unprecedented strength of interaction with light. Combined with other vdW materials, such as graphene, a conductor, and hexagonal boron nitride (hBN), an insulator, one can realize the entire gamut of electrically driven semiconductor devices, such as LEDs, photodetectors, sensors and energy storage devices10,11,12. Owing to their large exciton binding energy in the monolayer limit combined with properties such as valley polarization, the TMDs have also become a highly sought-after platform for realizing strongly coupled exciton-polariton devices with largely unexplored characteristics, such as the valley degree of freedom13,14,15,16, charged excitons17 and long-distance propagation18. Most of the work on exciton polaritons based on 2D TMDs is done via optical excitation, which is the scenario for most fields of exciton polaritons. However, with the recent emergence of polaritonic devices for applications that range from ultrafast LEDs to polaritonic circuits19, there is much interest in realizing electrically driven polariton emitters. Such emitters are highly desirable and also markedly distinct from their optically driven counterparts due to the device complexity. Polariton LEDs have been demonstrated in traditional inorganic semiconductors20,21,22 as well as in organic materials8,23,24,25 using bulk materials or with multiple quantum wells. Although there are few reports of the control of strong coupling in 2D TMDs via electric field gating17,26,27, there is no demonstration of the electrical injection of exciton polaritons in 2D material systems. The attractiveness of the 2D material platform stems from the possibility to realize devices that have atomically thin emissive layers that can be integrated with other vdW materials for contacts (graphene) and tunnel barriers (hBN). Furthermore, the 2D material platform also presents a unique opportunity to integrate these polariton LEDs with other vdW materials with magnetic28, superconducting29 and topological transport properties30, which will result in hitherto uncharted device features.

Shown in Fig. 1a is a schematic of the device, which consists of a 12-period bottom distributed Bragg reflector (DBR) and a 40 nm silver film that act as the bottom and top mirrors, respectively, of the microcavity. In the active region, we have the tunnel area as well as two more hBN-encapsulated monolayers of WS2. The tunnel area consists of a vdW heterostructure with monolayer WS2 as the light emitter, thin layers of hBN on either side of monolayer acting as the tunnel barrier and monolayer graphene electrodes to inject electrons and holes. The extra two hBN-encapsulated WS2 layers are included to increase the overall exciton density and thereby result in pronounced Rabi splitting of the polariton states. This is done to compensate the decrease in the cavity quality factor due to the loss that arises from graphene and the non-uniformity in the active surface created during the sample preparation. Comparison of devices with and without the additional WS2 layers are given in Supplementary Section 6. A capping layer of PMMA is spin coated on the vdW heterostructure and the cavity mode is tuned by the thickness of this layer. In the present demonstration, the PMMA thickness was tuned so that the cavity thickness was 3λ/2. Details of the sample preparation and the optical response of the empty cavity are discussed in Methods and Supplementary Section 1, respectively. Figure 1b shows the optical microscope image of the vdW heterostructure on the bottom DBR.

Fig. 1: Device schematic and tunnelling mechanism.
figure 1

a, The schematic of the device shows the different layers of the vdW heterostructure embedded inside a microcavity that consists of a bottom DBR and top silver mirrors. The thickness of each layer is given in Supplementary Section 1. b, Optical image of the vdW stack before the top mirror is grown. The gold contacts and top hBN are labelled. Owing to the large reflection from the DBR substrate, only the top hBN is observable. More images are shown in Supplementary Section 2. Scale bar, 10 µm. c, Band diagrams at a high bias above threshold. GrT, top graphene; GrB, bottom graphene. d, The tunnelling current density as a function of the bias voltage shows symmetric current–voltage characteristics.

Full size image

The band diagram of the vdW heterostructure in the tunnel geometry under bias is shown in Fig. 1c. EL is observed above the threshold voltage when the Fermi level of the top (bottom) graphene is biased above (below) the conduction (valence) band of WS2, which allows electrons (holes) to tunnel into the WS2 conduction (valence) band and eventual radiative recombination. Figure 1d shows the electrical characteristics of tunnelling current density J as a function of bias voltage V between the graphene electrodes. The sharp rise in current for both positive and negative voltages at around ±4 V indicates the onset of a tunnelling current through the structure. With the optimum thickness of hBN layers (~2 nm), we observed a significant tunnel current and an increased lifetime of the injected carriers for radiative recombination. Reducing the hBN thickness decreased the turn-on voltage, but also increased the likelihood of a breakdown at a higher current. In addition, the use of an ambient transfer technique resulted in air bubbles being trapped at the interfaces, which caused an additional increase in the turn-on voltage.

Before we performed the EL experiment, we characterized our device by angle-resolved white light reflectivity and photoluminescence (PL) to ascertain that we were, indeed, in the strong coupling regime. These measurements, as well as the EL, were carried out using a Fourier space (k space) imaging set-up to map out the energy versus in-plane momentum dispersion. Figure 2a shows the angle-resolved reflection spectra from the active area of our device, which demonstrates an anticrossing behaviour. We further observed the PL under non-resonant excitation (460 nm), which showed an intense emission from the lower polariton branch and a weaker emission from the upper polariton branch, as shown in Fig. 2b (Methods gives the optical measurement details). The EL measurements were carried out under an external d.c. bias applied using a Keithley 2400 source meter. The angle-resolved dispersion of the polariton EL at a 0.1 µA µm–2 injection is shown in Fig. 2c and was found to be identical to the PL dispersion (Fig. 2b). A spatial image of the EL is shown in Supplementary Section 3. The Rabi splitting and cavity detuning derived using a coupled oscillator mode fit (shown by solid and dashed white lines in Fig. 2c) to the EL experiment were ∼33 meV and ∼−13 meV, respectively. The cavity detuning is defined by δ = Ec − Ex, where Ex is the exciton energy and Ec is the cavity photon energy with zero in-plane momentum. The cavity linewidth (γc) and exciton linewidth (γx) at room temperature were 24 meV (Supplementary Fig. 1) and 29 meV, respectively. Thus, the Rabi splitting, 33 meV, was larger than \(\sqrt {\left( {{\mathrm{\gamma }}_{\rm{c}}^2 + {\mathrm{\gamma }}_{\rm{x}}^2} \right)/\left. 2 \right)}\), which satisfies the strong coupling criteria. The presence of graphene in the cavity layer lowered the quality factor and the larger thickness of the cavity (3λ/2) increased the mode volume, and thereby reduced the Rabi splitting compared to that in previous reports7. The sectional slice, at different angles, from the EL dispersion is shown in Fig. 2d. The dispersion of the upper and lower polariton modes can be clearly seen here as the anticrossing occurs in the vicinity of the exciton resonance (solid line). In this experiment, the EL emission was collected from a large area (~20 μm2) and non-uniformity across the sample surface created during the stacking and transfer processes (as shown by the EL spatial image in Supplementary Section 3) caused a further broadening of the cavity mode. In contrast, the PL showed narrower polariton branches because the PL was collected within a 1 μm2 spot size (Supplementary Section 4). The use of a transparent contact along with sample preparation protocols that leave less residue will help to minimize the broadening.

Fig. 2: Polariton dispersion.
figure 2

a,b, Angle-resolved reflectance (a) and PL (b). The white solid lines in b are guidelines that follow the polariton branches. c, Angle-resolved EL at current injection of 0.10 µA µm–2. The solid lines are a coupled oscillator model fit to the polariton branches. The dashed lines are bare cavity and exciton resonances. d, EL spectral plots at different angles from sectional slice in c. The (blue) solid vertical line indicates the exciton emission wavelength (λexciton) at 620.1 nm. The dashed trend lines indicate the upper and lower polariton branches. a.u., arbitrary units.

Full size image

As the tunnelling current increased, the overall intensity of the EL went up. A weak EL from the polaritons was observed near the threshold bias (Fig. 3a), whereas at a sufficiently higher bias above the threshold, the polaritonic emission became distinctively bright (Fig. 3b). Shown in Fig. 3c is the polar plot of EL intensity as a function of angle that depicts a narrow emission cone of ±15°. The radiation pattern remained almost unchanged for both the minimum (green curve) and maximum (orange curve) driving currents. The integrated intensity under different driving tunnel currents is shown in Fig. 3d (black circles, left axis) and follows an almost linear trend. Increasing the current to sufficiently higher values could lead to a successful polariton scattering along the lower branch and create an extremely narrow emission pattern due to polariton lasing. However, in our case we were limited by the dielectric breakdown of hBN tunnelling barrier and hence could not reach this regime. Improvement in the quality of the hBN could further increase the damage threshold. The external quantum efficiency (EQE), which is the ratio of the number of extracted photons to the number of injected charge particles, is also plotted in Fig. 3d (red circles, right axis) as a function of current density. The observed EQE is comparable to those in other reports of polariton LEDs, such as in organic materials8 and carbon nanotubes9, albeit the light-emitting layer of the present device is only few atom layers thick (~0.7 nm) compared to the much thicker active material used in previous demonstrations. However, the observed EQE is lower than that reported for similar tunnelling devices not confined to a cavity geometry10. The reduced efficiency is probably due to the poor light extraction from our cavity, which needs further improvement as well as in-plane waveguiding. An alternative way to increase EQE is to stack more monolayers inside the tunnel region separated by thin hBN10. Details of the EQE estimation are given in Supplementary Section 7.

Fig. 3: Current-dependent polariton EL intensity.
figure 3

a,b, Polariton dispersion from EL under current injections of 0.08 µA µm–2 (a) and 0.28 µA µm–2 (b). White solid lines are guidelines that follow the polariton branches. c, Polar plot from different current density. The emission angular distribution pattern does not change within the range of applied current. d, Integrated EL intensity (black) and EQE (red) as a function of current density. The EL process is in the linear regime within the range of current applied.

Full size image

We also investigated the effect of the cavity detuning on the polariton EL by fabricating a similar device but with a larger cavity detuning (−43 meV). Further details of this device (device 2) are discussed in Supplementary Section 5. Figure 4a shows the angle-resolved EL spectra from the highly negatively detuned device at a current density of 0.2 µA µm–2. Owing to the larger detuning, this device showed a strong bottle neck effect in the EL with an emission maximum that occurred at a large angle. This is further confirmed in Fig. 4b, which compares the normalized polar plot from device 2 with that obtained from device 1 (Fig. 2c). For the higher negative detuning, the emission maximum occurs at 18° as compared with device 1, which centres at 0°. This bottleneck effect for the larger detuning sample can be understood as a result of the poor polariton scattering to k|| = 0 owing to the short polariton lifetime in these cavities and the larger negative detuning of device 2. The Rabi splitting of device 2 is 27 meV and can be thought of as in the limit of strong coupling using the criterion discussed previously. Both the graphene and mode volume play a role in reducing the Rabi splitting, but in the case of device 2, the additional WS2 layers are not at the field maxima in the cavity due to the thicker hBN encapsulating layers (Supplementary Section 5). This further reduces the Rabi splitting.

Fig. 4: Negatively detuned polariton LED.
figure 4

a, Angle-resolved EL (0.2 µA/µm2) of device 2 with a large negative detuning (−43 meV). A strong bottle neck effect was observed, which resulted in emission at a larger in-plane momentum. b, The normalized angular emission pattern from device 1 (Fig. 2c) and device 2 (a) shows the modification in the emission pattern owing to the bottleneck effect in device 2.

Full size image

In summary, we have demonstrated room-temperature polariton EL from a vdW heterostructure embedded in a microcavity. The tunnelling architecture of our device enables electron and hole injection and recombination in the WS2 monolayer, which acts as the light-emitting layer. Above the threshold bias, the bands are aligned, which favours carrier tunnelling from the graphene electrodes to the monolayer WS2 through the ultrathin hBN barriers. Varying the current injection above the threshold leads to a significant increase in emission intensity. The EL is also found to be highly directional owing to the cavity dispersion. Further improvements in cavity Q factor and higher current injection should help realize more efficient microcavity LEDs and the possibility of an electrically driven low-threshold microcavity polariton and/or a photon laser. The present demonstration of EL from TMD exciton polaritons in a microcavity is an important step towards realizing electrically driven integrated microcavity light emitters using 2D vdW materials for potential applications as ultrafast LEDs and low threshold lasers.

Methods

Sample preparation

The DBR. which consisted of 12 periods of alternate layers of SiO2 (106.2 nm) and Si3N4 (77.5 nm), was grown on a silicon substrate by plasma-enhanced chemical vapour deposition using a combination of nitrous oxide, silane and ammonia at a temperature of 350 °C. Two gold contacts were then fabricated onto the DBR top surface. We used electron beam lithography to pattern the contacts and deposited Ti/Au (2 nm/8 nm) by electron beam evaporation. Monolayer WS2, graphene and multilayer hBN were exfoliated from bulk crystals (WS2 and graphene from HQ Graphene and hBN from 2Dsemiconductor Inc.) using scotch tape onto a 300 nm SiO2/Si substrate. Heterostructure stacking and transfer were done using the well-known polypropylene carbonate transfer technique31. We first identified a thick hBN layer (40 nm) and then used it to stack the top two WS2 monolayers followed by stacking the tunnel region. The final stack structure from top to bottom was hBN/WS2/hBN/WS2/hBN/graphene/hBN/WS2/hBN/graphene/hBN. There are 11 separate layers and the stacking was done continuously from top to bottom. Several stacking images are shown in Supplementary Section 2. The entire stack of the vdW heterostructure was then transferred onto the DBR held at a temperature of 120 °C. Alignment was carefully done to make sure each graphene flake sat exactly on top of corresponding gold contact pad. After the transfer, the entire structure was soaked in chloroform for 2 h to remove the polypropylene carbonate residue followed by polymethylmethacrylate (495 A4 from Michrochem) spin coating to form a 200 nm top spacer layer. The final silver layer (40 nm) was deposited via electron-beam evaporation for the top mirror of the microcavity. Details of each layer thickness and cavity response are given in Supplementary Section 1.

Optical measurements

Angle-resolved spectra were recorded using a homemade spectroscopy set-up that comprised white light (broadband halogen source for reflection) and a laser (PL measurement). The set-up was coupled with a Princeton Instruments monochromator with a PIXIS: 256 EMCCD camera. A ×100, 0.7 numerical aperture objective was used for all the measurements. The polariton dispersion was revealed by imaging the back aperture of the microscope objective (Fourier plane) on to the camera. All the measurements were done at room temperature.

EL measurements

The EL measurements were carried out under an external d.c. bias applied using a Keithley 2400 source meter. The emission from the device was imaged using the same Fourier imaging system as used for PL measurements.