High-Speed Electroluminescence Modulation in Monolayer WS 2

The high‐speed modulation of the nanoscale light sources is of fundamental interest in nanophotonics. Here, electrically driven light emission from a metal–insulator–semiconductor heterostructure consisting of graphene, hexagonal boron nitride (h‐BN), and monolayer tungsten disulfide (WS2) is demonstrated. Electroluminescence in these devices originates from radiative recombination of majority carriers (electrons) accumulated by electrostatic doping and hot minority carriers (holes) injected into monolayer WS2 from graphene through an ultrathin h‐BN tunnel barrier. The devices are electrically driven with a radio frequency signal and electrical modulation of the light emission at frequencies up to 1.5 GHz is demonstrated. The high‐speed WS2 tunnel diodes provide a promising path for on‐chip nanophotonics.


High-Speed Electroluminescence Modulation in Monolayer WS 2
Dohyun Kwak,* Matthias Paur, Kenji Watanabe, Takashi Taniguchi, and Thomas Mueller DOI: 10.1002/admt.202100915 driven thermal emitters based on graphene encapsulated by hexagonal boron nitride (h-BN) have shown a modulation bandwidth of 3 GHz. [11] However, the lack of an energy gap in graphene limits its application to broadband thermal emission with low efficiency. By contrast, single layer semiconducting transition metal dichalcogenides (TMDs) such as WS 2 and WSe 2 have direct bandgaps between the red-end of the visible spectrum and the near-IR. [6,7] Furthermore, they exhibit strongly bound excitons at room temperature due to the strong Coulomb interactions from the quantum confinement in the direction perpendicular to the 2D plane and weak dielectric screening in single layer TMDs. [12] The excitonic features of single layer TMDs provide a unique platform to investigate light-matter interactions. [12,13] To explore the exciton physics of single layer TMDs, various device concepts for electrically driven light emitters have been explored. [14][15][16][17][18][19] Time-resolved experiments and theoretical studies of single layer TMDs have demonstrated very short exciton lifetime on a picosecond timescale. [20][21][22][23] These investigations suggest that light emitting diodes (LEDs) based on single layer TMDs could enable ultrafast modulation of the emitted light. Here, we demonstrate electrically driven light emission at gigahertz frequencies from h-BN encapsulated WS 2 tunnel diodes, consisting of a van der Waals (vdW) heterostack (graphene/h-BN/graphene/ WS 2 /h-BN) on a SiO 2 /Si substrate. Our tunnel WS 2 diodes based on a metal-insulator-semiconductor (MIS) structure reveal efficient electroluminescence (EL), which is dominated by negatively charged excitons due to electron doing of WS 2 by the applied gate voltage. The light emission from our devices can be electrically modulated with a RF signal up to a frequency of 1.5 GHz.

Results and Discussion
The sample structure is schematically depicted in Figure 1a. Monolayer WS 2 is completely encapsulated by top and bottom h-BN. The encapsulation by h-BN is known to preserve the high optical quality of TMDs, preventing doping from the SiO 2 substrate or adsorption of ambient molecules. [24,25] In our devices, monolayer WS 2 serves as an active layer for EL emission and it is electrostatically doped by a back-gate voltage across the bottom h-BN/SiO 2 layer to increase its electron density. We used graphene as contact electrode to single layer WS 2 because it shows low contact resistance, as well as efficient charge carrier injection for light emission. [26] In our device structure, The high-speed modulation of the nanoscale light sources is of fundamental interest in nanophotonics. Here, electrically driven light emission from a metal-insulator-semiconductor heterostructure consisting of graphene, hexagonal boron nitride (h-BN), and monolayer tungsten disulfide (WS 2 ) is demonstrated. Electroluminescence in these devices originates from radiative recombination of majority carriers (electrons) accumulated by electrostatic doping and hot minority carriers (holes) injected into monolayer WS 2 from graphene through an ultrathin h-BN tunnel barrier. The devices are electrically driven with a radio frequency signal and electrical modulation of the light emission at frequencies up to 1.5 GHz is demonstrated. The high-speed WS 2 tunnel diodes provide a promising path for on-chip nanophotonics.

Introduction
The study of nanoscale light sources is an active field of research as it may enable high-density photonic integrated circuits with a compact size. A number of photonic materials such as organic materials, semiconducting quantum dots, and nanowires have been studied and used as nanoscale light sources. [1][2][3][4][5] Achieving high electrical bandwidth and fast modulation speed has, however, remained a fundamental challenge. [2] Recently, 2D materials have shown exciting properties for optoelectronic applications. [6,7] Graphene, consisting of a single layer of carbon atoms in a honeycomb lattice, has attracted a lot of interest in light detection, modulation, and manipulation applications because of its exceptional electrical and optical properties. [8,9] As a candidate of nanoscale light sources, graphene can electrically emit bright and broadband light in the visible range from the spatially localized accumulation of hot electrons. [10,11] Electrically bottom graphene under top h-BN is contacted to WS 2 , and top graphene is used as hole injection layer in which carriers tunnel to WS 2 across the barrier of top h-BN by applying a bias voltage. The thickness of top h-BN is ≈2 nm. Finally, Ti/Au electrodes of 3/60 nm thickness were deposited on the bottom graphene as a source and the top graphene as a drain electrode, respectively. Figure 1b shows the electrical characteristic of our device at a gate voltage of V g = 80 V. The device exhibits diode-like rectifying behavior with a rectification ratio of ≈10 3 . The rectification ratio is defined at 4 V and −4 V. The current increases exponentially in forward bias region, while it is strongly suppressed under reverse bias. The I-V curve in forward bias region can be divided into two regimes: before and after tunneling of charge carriers from graphene to WS 2 across the top h-BN barrier. The resistance of the device in the first regime (i.e. below threshold) is dominated by the tunnel barrier and the device current is suppressed due to insufficient energy of the carriers to overcome the barrier. The threshold voltage of the device depends on the thickness of the top h-BN ( Figure S1, Supporting information). In case of an h-BN monolayer, the device exhibits ohmic-like behavior, indicating that the atomically thin h-BN is no longer a tunnel barrier. By contrast, thick h-BN as a tunnel barrier causes high threshold voltage of the device and an increase of the turn-on voltage. We achieved best results by using h-BN thicknesses ranging from 2 to 4 nm. With increasing device current above threshold voltage, the tunnelresistance becomes comparable to the sheet resistance of WS 2 . The device current in the sheet-resistance-limited regime, where the lateral resistance of WS 2 layer is dominant, can be controlled by electrostatic electron doping. I-V curves of our device reveal diode-like behavior above a gate voltage of V g = 20 V as shown in the inset of Figure 1b. The more electrostatic n-type doping of the WS 2 layer, the larger the device current in forward bias region. The device current for gates voltages below V g = 20 V is completely suppressed, consistent with the n-type behavior of WS 2 . [27] In Figure 1c we present a band diagram which schematically depicts the EL process in our devices. In general, EL from unipolar devices based on semiconductor-metal junctions results from radiative recombination of electron-hole pairs by carriers injected from a metal into a semiconductor through a Schottky barrier. [5,28] Similarly, EL from our devices is generated by radiative recombination between electrons in the WS 2 layer and holes injected from graphene when a bias voltage is applied. Holes from graphene below threshold are blocked by the top h-BN tunnel barrier. Above threshold voltage, holes from graphene can transfer to the WS 2 layer by overcoming the h-BN tunnel barrier, which is known to serve as a hole-transporting and electron-blocking layer. [28] No EL at a negative gate voltage is observed because of electron depletion in the WS 2 layer. Thus, efficient light emission in our devices emerges above threshold voltage when a positive gate voltage is applied to the device. In Figure 1d, we show photoluminescence (PL) and EL spectra at room temperature. The EL was measured at a bias voltage of V d = 6.5 V and a gate voltage of V g = 80 V. The PL was acquired with a continuous wave laser (λ = 532 nm, P = 5 µW) at room temperature. The EL exhibits an emission maximum centered at 1.864 eV and it is broadened as compared to the PL peak showing its maximum at 1.998 eV. While the PL is dominated by neutral exciton emission, the EL is predominantly due to negatively charged excitons, as will be discussed later. Figure 1e,f shows optical images of the device and the EL emission, recorded with a CCD. PL from the device can be observed across the whole face of the WS 2 flake ( Figure S2, Supporting information), but EL occurs mostly at the interface of the graphene/h-BN/WS 2 junction in Figure 1f. This observation indicates that most holes injected from graphene strongly recombine with electrons at the interface of the graphene/h-BN/ WS 2 junction. The unevenly distributed light emission is attributed to as variation of the tunnel-resistance across the device.
To study the origin of the EL emission, we investigated PL and EL spectra as a function of bias voltage and photon energy at a positive gate voltage of V g = 80 V as shown in Figure 2a,b. Electrostatic electron doping of single layer WS 2 causes a reduction of exciton emission and formation of charged excitons. [29] Consistently, the reduction of the PL intensity from our device was observed with increasing gate voltage due to change of the Fermi level by electrostatic electron doping of the WS 2 layer ( Figure S3, Supporting information).
When a bias voltage is gradually applied to the device at a gate voltage of V g = 80 V, neutral exciton (X 0 ) emission disappears and emission from the negatively charged excitons (X − ) becomes predominant as shown in Figure 2a. The PL peak is predominantly due to neutral excitons without bias voltage, but negatively charged excitons are identified at a bias voltage of V d = 1 V in Figure 2c. The change of the emission spectra is attributed to injection of charge carriers from graphene to WS 2 . The device current from injection of charge carrier is identified at a bias voltage of V d = 1 V as shown in Figure 1b. The injection of charge carriers can cause electrical doping of the WS 2 layer, and the trion binding energies vary with doping level of TMDs. [30] With further increasing bias voltage in Figure 2a, the emission peak continuously shifts to the red and becomes broader due to an increase of doping level by the injection of charge carriers. EL in our devices appears above a bias voltage of V d = 3 V. After EL emergence in Figure 2a, the emission intensity increases with increasing bias voltage in agreement with the results of the EL map in Figure 2b. EL intensity of our device increases exponentially with increasing bias voltage in Figure 2d. In addition, the EL spectra linearly shift to the red with increasing bias voltage in Figure 2e and become broader. Both the peak shift and the broadening of the EL are due to increased temperature at the junction. [31,32] In Figure 2c, the emission spectra at V d = 6 V clearly shift to the red as compared to the energy of the neutral excitons. The results suggest that the EL from the device origins from negatively charged excitons.
To study PL and EL spectra from the device at low temperature, we recorded PL and EL spectral intensity maps. Figure 3a-c exhibits PL intensity maps at different gate voltages and bias voltages as well as the EL map at different bias voltages. Only one emission peak was identified in the PL and EL at room temperature as shown in Figure 1d. However, various emission peaks in monolayer WS 2 can be observed at cryogenic temperature. [13,33,34] In Figure 3d, the PL spectra without application of a gate voltage show various exciton complexes such as neutral exciton, trions, and charged excitons. The different exciton complexes in monolayer WS 2 at cryogenic temperature have been studied extensively. [13,33,34] The neutral exciton (X 0 ) and charged trion (X − ) in the PL spectra of our sample appear at 2.0764 and 2.043 eV, respectively. The additional emission peaks (labeled L 1 , L 2 , and L 3 ) are observed at 2.027, 2.003, and 1.973 eV, respectively. These emission peaks are attributed to recombination of localized excitons, bound to impurities or defects. [33] The PL spectra at negative gate voltage hardly change compared to that without gate voltage except a slight decrease of L 1 and L 2 peak intensities ( Figure S4, Supporting information). With increasing gate voltage, the PL intensity of localized excitons increases, and one localized peak in the PL spectra at a gate voltage of V g = 40 V becomes predominant as shown in Figure 3d. The localized peak continuously redshifts with increasing gate voltage. The results are attributed to electrostatic electron-doping which causes an decrease of the exciton energy. [30,35] We further investigated the PL spectra with different bias voltages at a fixed gate voltage of V g = 80 V, as application of a bias voltage can also cause electrical doping. Figure 3e shows normalized PL spectra with different bias voltages at a gate voltage of V g = 80 V. The neutral exciton and biexciton peaks disappear when a bias voltage is applied to the device. The result is consistent with the PL change at room temperature in Figure 2a. The emission peak of a localized state becomes predominant below a bias voltage of V d = 2 V. The emission spectra at a bias voltage of V d = 3 V exhibit a new peak of ≈1.988 eV, which originates from EL. The emission peak broadens strongly and redshifts with increasing bias voltage. EL emission was observed above a bias voltage of V d = 2 V at 5 K as shown in Figure 3c. The turn-on voltage for EL emission at 5 K is lower than that at room temperature because the threshold current depends on temperature according to J th ≈ exp(ΔT/T 0 ), where J th is a threshold current, ΔT is the temperature difference, and T 0 is a characteristic temperature. [36] The EL at cryogenic temperature shows only one broad peak which continuously shifts to the red with increasing bias voltage as shown in Figure 3f. This observation indicates that EL in our device at 5 K originates from recombination of negatively charged excitons. In addition, with increasing bias voltage, the intensity of the EL increases strongly and a broadening of the emission peak occurs.
To characterize the modulation speed of our devices we used a time-correlated single-photon counting (TCSPC) method. Figure 4a depicts the schematic measurement setup. The device was electrically driven by a radio frequency (RF) sine signal and an offset DC voltage at room temperature. We applied a DC bias voltage to operate the device above the turn-on voltage at a gate voltage of V g = 80 V. The electrically modulated emission from the device was collected by two Si avalanche photodetectors . Then, the collected signals were sent to a time-correlated single photon counter, which measured the delay between photon arrival times in the resulting photon stream. We electrically modulated our device from 100 MHz up to 1.5 GHz with a sine signal of 14 dBm (corresponding to a peak voltage of V ac = 1.6 V) on a DC offset voltage of V d = 6.5 V. Figure 4b-d shows histograms of the optical output from the device at different modulation frequencies of 100, 800, and 1500 MHz for ten oscillation periods. The results for other frequencies are presented in Figure S5 (Supporting information). The raw results of optical output were fitted by a sine wave y 0 + A sin(2πft), where A is the amplitude, f the frequency, and t the time. The fits are in good agreement with the measurement data. Figure 4e-g exhibits the optical output from the device for two oscillation periods. The optical output from our device is well modulated by the input sine signal. The results show that the light emission from our device can be modulated with a frequency of up to 1.5 GHz. In addition, we show the fitted amplitudes as a function of modulated frequency ( Figure S6, Supporting information). The amplitude sharply decreases for f > 500 MHz, and we estimate a 3 dB cut-off frequency of f c ≈ 600 MHz. This fast modulation speed of our device is attributed to the short exciton lifetime of TMDs and the fast tunneling injection of carriers (≈ps) without suffering from low mobility in the channel. [37,38] The WS 2 tunnel diodes are faster than conventional semiconductor LED (e.g., GaAs). The direct gap III-V materials suffer from intrinsic bandwidth limitations due to minority-carrier lifetimes of a few nanoseconds. [39,40] For a high-speed operation, traditional LEDs are heavily doped since the increase of carrier concentration reduces minoritycarrier lifetime. [39,40] By contrast, single layer TMDs show short exciton lifetime of a few picoseconds in theoretical studies and Figure 3. a) PL spectral intensity as a function of gate voltage and photon energy at 5 K. b) PL spectral intensity as a function of bias voltage and photon energy with a gate voltage of 80 V g at 5 K. c) EL spectral intensity as a function of bias voltage and photon energy with a gate voltage of 80 V g at 5 K. d) Normalized PL spectra at different gate voltages in ranging 0 to 80 V g at 5 K. e) Normalized PL spectra at different bias voltages in ranging 0 to 4 V d with a gate voltage of 80 V g at 5 K. f) Normalized EL spectra at different bias voltages in ranging 2.5 to 4 V d with a gate bias of 80 V g at 5 K.
experiments. [20][21][22][23] We also explored the dependence of the electrically modulated EL on the DC offset voltage and an RF modulation voltage. The optical output from the device was collected with a frequency modulation of 500 MHz, different offset DC voltages and RF modulation voltages at the fixed count time and fixed maximum count, respectively ( Figure S7, Supporting information). The raw results of optical output were again fitted by sine wave. The EL intensity at the highest point of a sine wave increases with increasing both DC offset voltage and RF modulation voltage. The amplitude of output sine signal, however, only depends on the RF modulation voltage.

Conclusion
In this work, we reported tunnel WS 2 light emitting diodes based on an MIS vdW heterostructure. The tunnel WS 2 lightemitting devices were operated by application of a bias voltage and a gate voltage at room temperature. The origin of the EL from the devices is due to the radiative recombination of holes injected from graphene across an h-BN tunnel barrier and electrons accumulated by electrically doping on WS 2 . The fasttunneling injection of carriers (≈ps) and short exciton lifetime of picoseconds in TMDs allow a high-speed modulation of EL