Gate-tunable mid-infrared electroluminescence from Te/MoS2 p-n heterojunctions Authors: Shiyu Wang, Delang Liang, Zhi Zheng, Mingyang Qin, Yuchun Chen, Jie Sheng, Shula Chen, Lin Li, Changgan Zeng, Anlian Pan, Jinluo Cheng, Dong Sun Published: 2026 ABSTRACT AbstractMid-infrared (MIR) emitters are critical components in advanced photonic systems, driving progress in fields such as chemical sensing, environmental monitoring, medical diagnostics, thermal imaging and free-space communications. Conventional MIR emitters based on III–V heterostructures rely on complex epitaxial growth on rigid lattice-matched substrates and suffer from limited integration compatibility with CMOS or flexible platforms. Recent MIR emitters based on two-dimensional (2D) materials such as black phosphorus (BP) is more suitable for on-chip applications but faces challenges related to stability and emission efficiency. Based on the recently discovered highly efficient photoluminescence of Te, we demonstrate a gate-tunable mid-infrared light-emitting diode based on a van der Waals heterojunction formed by multilayer transition metal dichalcogenide (TMD) MoS2 and tellurium (Te). The device emits polarized electroluminescence (EL) centered at 3.5 μm under forward bias at 25 K, and the EL persists up to 80 K with reduced intensity. Gate control of the MoS2 Fermi level modulates the band alignment and injection efficiency, enabling dynamic tuning of the EL intensity. The emission remains spectrally stable under varying bias and gating, indicating robust band-edge recombination. These results establish the Te/TMD heterostructure as a promising platform for integrated polarized mid-infrared optoelectronics.Subject terms: Lasers, LEDs and light sources; Mid-infrared photonics; Electronics, photonics and device physics Introduction The mid-infrared (MIR, 2.5–25 μm) spectral region is essential for numerous applications, including gas sensing, infrared spectroscopy, thermal imaging, and free-space optical communication1–3. Conventional MIR light-emitting technologies have been developed primarily on epitaxial III–V platforms, including InSb/AlInSb quantum-well light-emitting diodes4, GaInAsSbP alloy light-emitting diodes5, and intersubband devices based on III–V compounds6,7. Within the same epitaxial technology landscape, type-II InAs/GaSb superlattice interband LEDs established cascaded active-region concepts for 3–4 μm electroluminescence8 and were subsequently advanced toward higher radiance and output power through optimized cascaded designs, including work from John P. Prineas and co-workers9. Interband cascade LEDs further pushed continuous-wave mid-IR LED power into the multi-milliwatt regime, with notable progress from industrial development efforts such as nanoplus10. More recently, mercury-chalcogenide colloidal quantum-dot LEDs, including HgTe and HgSe-based systems from Philippe Guyot-Sionnest’s group, have emerged as a solution-processable route to mid-IR electroluminescence11,12. Despite their strong performance, these established approaches typically rely on epitaxial growth on lattice-matched substrates and device stacks that are less compatible with CMOS back-end processing or mechanically flexible platforms. Moreover, polarization-defined emission and spectral engineering in many mid-IR emitters are commonly implemented by incorporating additional photonic structures such as resonant cavities or gratings, while the growing demand for on-chip mid-IR polarimetry motivates compact sources with intrinsic and electrically controllable polarization. These constraints have spurred interest in alternative optoelectronic systems that combine mechanical flexibility, process scalability, and spectral tunability. Two-dimensional (2D) van der Waals (vdW) semiconductors offer distinct advantages for light emitters13–19, including atomic thinness, electrostatic tunability, and seamless integration on arbitrary substrates without lattice-matching constraints. Following the isolation of monolayer transition metal dichalcogenides (TMDs), electroluminescence (EL) was rapidly demonstrated in monolayer MoS214 and WSe₂, whose direct bandgaps and strong excitonic effects enable bright emission in the visible and near-infrared (NIR) regimes. Various device architectures—such as lateral p-n junctions20, vertical heterostructures15, and tunneling diodes16–18—have since been developed to realize valley polarization-tunable and gate-controlled EL19 based on TMDs. Despite these advances, the relatively large bandgaps of TMDs (>1 eV) fundamentally limit their operation wavelength range to the visible-NIR region, thereby limiting their applicability in the mid-infrared region. In addition, light emission from TMDs is not polarized due to the nearly isotropic in-plane optical response of TMDs, and it typically requires external photonic structures to achieve strong linear polarization21. This is in contrast to intrinsically anisotropic 2D semiconductors such as black phosphorus or Te. As a result, TMD-based emitters are not ideal for applications in the mid-infrared region. To overcome these challenges, there has been growing interest in anisotropic 2D semiconductors with narrower bandgaps and polarization-selective optical properties. Black phosphorus (BP) has emerged as a promising MIR emitter, with a thickness-dependent direct bandgap ranging from ~0.3 eV (bulk) to ~2.0 eV (monolayer)22–24 and pronounced in-plane anisotropy that enables linearly polarized emission22,25–27. BP-based MIR light-emitting diodes have been investigated across diverse architectures, including heterojunction emitters, waveguide-integrated implementations, resonant-cavity designs, and van der Waals heterostructures, with recent progress toward scalable device formats26–30. A major impediment, however, is BP’s limited ambient stability31,32, which necessitates effective encapsulation for reliable operation. Encouragingly, long operating lifetimes have been demonstrated for properly encapsulated BP LEDs, highlighting the practical importance of stringent environmental isolation for BP-based MIR emitters. Moreover, BP exhibits strong electric field-induced spectral tunability via the Stark effect—an attractive route for active wavelength control, yet one that can also introduce emission-energy shifts under vertical electric fields33. Recently, tellurium (Te)34–37, a layered narrow-bandgap (~0.35 eV) p-type semiconductor, has attracted growing attention as an alternative to BP for MIR optoelectronics. Te possesses a thickness-independent bandgap that is nominally indirect with a small k-offset but effectively quasi-direct for optical transitions, along with excellent air stability, high mobility38,39, and intrinsic linear dichroism40. Prior studies have demonstrated MIR photodetection40–42 and highly efficient photoluminescence43 in Te-based heterostructures, highlighting its potential for MIR optoelectronics. Very recently, during the preparation of this work, electrically driven mid-infrared electroluminescence has been demonstrated in Te-based van der Waals LEDs that rely on dual-electrode bipolar injection, and the polarization of the emission can be tuned by bias44. These advances also link Te-based MIR emitters to Weyl physics, opening research avenues beyond the scope of conventional band-edge optoelectronics45,46. In this work, we develop a gate-tunable heterojunction of Te that enables systematic modulation of the built-in junction electrostatics and interfacial band alignment to control injection and recombination. Based on a vdW p-n heterojunction formed by n-type multilayer MoS2 and p-type Te, as schematically illustrated in Fig. 1a, we demonstrate a mid-infrared light-emitting diode. The band diagram highlights carrier injection and radiative recombination processes localized within Te, giving rise to EL emission centered at 3.5 μm. Gate-tunable carrier modulation enables control over the band alignment and interfacial injection efficiency. The device exhibits linearly polarized EL under forward bias, with the lowest turn-on voltage of ~1.7 V. Spectral analyses confirm that EL arises from direct band-edge transitions in Te, consistent with its PL response. Importantly, the Te/MoS2 platform demonstrates wavelength-stable and polarization-locked MIR emission under electrostatic control, and we further verify its operational robustness through extended storage (~10 months) without the need for strict hermetic encapsulation. These findings establish the Te/MoS2 heterostructure as a promising platform for gate-tunable, polarized mid-infrared light sources in reconfigurable photonic and optoelectronic circuits. Schematic and optical properties of the Te/MoS2 heterojunction. a Schematic illustration of the Te/MoS2 heterojunction device. b Optical micrograph showing the Te/MoS2 heterojunction region (outlined by blue/green dashed lines) on a Si/SiO₂ (285 nm) substrate used as the back-gate. Scale bar: 30 μm. c Raman spectra of isolated Te (purple), MoS2 (blue), and the heterojunction region (yellow). d Spatially resolved EL intensity map of the Te/MoS2 heterojunction LED, recorded in the 3.5 µm band under forward bias (Vg=20V,Vds=4V;25K). The Te flake and MoS2 region are outlined by red and green dashed lines, respectively. e PL (red) and EL (blue) spectra of the Te/MoS2 diode. f Polarization characterization of PL, EL, and the A₁ Raman mode. All curves are fit by I=Imax−Imincos2θ+Imin, where θ is measured from the a-axis of Te, and Imax and Imin denote the intensities along the a-axis and c-axis for each polar plot, respectively. The crystal orientations are labeled with black arrows Results Characterization of transport and optical properties The device was fabricated via standard mechanical exfoliation and van der Waals (vdW) assembly processes, where multilayer MoS2 was positioned on top of a Te nanosheet. The source and drain electrodes were patterned by electron beam evaporation of Pd/Au (20 nm/100 nm) and Cr/Au (20 nm/100 nm) onto the Te flake and MoS2 layers, respectively. The 285 nm SiO₂ layer serves as the back-gate dielectric, with the p-type silicon substrate used as the back gate. Figure 1b shows an optical micrograph of a typical Te/MoS2 heterostructure, where the boundaries of the MoS2 and Te nanosheets are outlined by dashed blue and green lines, respectively. The Au electrodes are visible as bright metallic regions, outlined by orange dashed lines. To determine the thickness and surface morphology of the MoS2/Te heterojunction, atomic force microscopy (AFM) measurements were performed over the sample area. The thicknesses of MoS2 and Te were measured to be ~17 and 234 nm, respectively (see Supplementary Information Fig. S1). Raman spectroscopy was performed to confirm the composition and crystalline quality of the heterostructure. For Raman measurement, a 532 nm laser was utilized to excite the characteristic resonant peaks in Te, MoS2, and the heterojunction region. As shown in Fig. 1c, the Te-related peaks at 92, 120, and 140 cm−1 are assigned to the E1 (bond-bending mode), A (chain expansion mode), and E2 (bond-stretching mode) phonon vibrational modes; the MoS2-related peaks at 385 and 409 cm−1 are assigned to the E2g (in-plane Mo-S vibration mode) and A1g (out-of-plane mode), respectively. In the heterostructure region, all peaks retain their spectral positions without broadening or shifting, demonstrating negligible interlayer strain and preserved lattice integrity at the interface. No additional Raman modes are observed in the overlap region, consistent with a structurally intact van der Waals heterojunction. We focused on characterizing the device’s EL emission under source-drain bias first. To gain deeper insight into the origin of this EL emission, we also carried out photoluminescence (PL) measurements of the Te/MoS2 heterostructure under 1064 nm laser excitation. As shown in Fig. 1e, the resulting PL and EL spectra are presented together for direct comparison. Under forward-bias conditions (Vg = 20 V, Vds = 4 V), the device exhibits EL emission centered at ~3.5 μm. All measurements were performed at 25 K, unless otherwise noted, to enable high signal-to-noise detection of gate-modulated mid-infrared emission. As shown in Supplementary Note 4, electroluminescence persists up to 80 K, although the emission intensity decreases with increasing temperature. A direct spectral comparison with PL reveals that both EL and PL exhibit nearly identical peak positions and spectral linewidths (Fig. 1e), suggesting that both emissions originate from interband transitions of Te, a quasi-direct transition between the valence band maximum and conduction band minimum of Te. Spatially resolved EL mapping (Fig. 1d) shows that the mid-infrared emission primarily originates from the Te/MoS2 overlap (junction) region. The device geometry in Fig. 1d was determined by comparing the EL map to a reflection micrograph of the same area, which is provided in Supplementary Fig. S5. Given the in-plane anisotropic chain-like crystal structure of Te, the EL emission from Te-based heterostructures is expected to be strongly linearly polarized, similar to the polarized photoluminescence (PL) reported in our recent work47. The polarized emission is crucial for polarization-sensitive optoelectronic applications. Figure 1f presents measurements of the polarization characteristics of the EL emission. Notably, both EL and PL emissions reach maximum intensity when the polarization is aligned along the armchair direction (a-axis) and minimum intensity when the polarization is aligned along the chain direction (c-axis), reflecting the intrinsic optical anisotropy dictated by selection rules of the dipole matrix elements near the Te bandgap, where radiative transitions are preferentially oriented along the a-axis. To quantify the polarization, we use the degree of linear polarization (DOP), defined as DOP=Imax−Imin/Imax+Imin. The DOP of the measured device is also bias dependent, consistent with a recent report on graphene-Te-graphene mid-infrared LEDs44. The EL polarization plot shown in Fig. 1f yields a DOP of 0.70. We note that the polarization state varies significantly across different devices. Different Te flakes in Te/TMD heterojunctions exhibit different degrees of linear polarization in EL, while in an independent Te/WSe2 device, we observe a nearly unity EL DOP (see Supplementary Fig. S6). The complex DOP results may arise from the flake-to-flake variations in the intrinsic doping of Te, which shifts the equilibrium Fermi level and thereby influences the polarization; and recent Te-based MIR LEDs have shown that the DOP can vary with the injection condition44. Local strain and substrate-induced effects are also possible contributing factors. A comprehensive bias‑ and gate‑dependent study of the polarization behavior is beyond the scope of this work and requires further investigation. Having established that the Te/MoS2 heterostructure functions as a polarized mid‑infrared light emitter, we next examine its electronic transport properties and interfacial band alignment. In the present device platform, Te flakes were selected from hydrothermal batches to ensure sufficient morphological uniformity for reliable junction formation, while few-layer MoS2 was employed to provide robust electron transport and stable current drive for reproducible EL turn-on. A brief rationale for the thickness selection is provided in Supplementary Note 1. The electrical properties of the Te/MoS2 heterostructure were systematically investigated to elucidate its p-n junction behavior and gate-dependent transport characteristics at 25 K. The output characteristics (Ids-Vds) measured at back-gate voltages (Vg) of 30, 0, and –30 V show pronounced, gate-tunable rectification behavior, as depicted in Fig. 2. At Vg = +30 V, when the MoS2 channel is strongly n-doped, the device exhibits clear rectification behavior. At Vg = 0 V, the rectification is strongly suppressed, and the currents remain in the submicroampere range over the entire Vds window. At Vg = –30 V, the current stays at the measurement noise floor over the Vds range, indicating that carrier injection across the junction is effectively quenched by gate-induced depletion. Band alignment and electrical transport of the Te/MoS2 heterojunction. a Qualitative equilibrium band alignment of the Te/MoS2 heterojunction at Vg = 0 V, with the depletion region marked by gray dashed lines. b Output characteristics (Ids-Vds) of the heterostructure measured at 25 K for back-gate voltages Vg = 30, 0, and −30 V, showing gate-tunable rectification behavior. The inset provides a magnified view of the low-current region. c Transfer characteristics (Ids-Vg) recorded at Vds = 4 V. d Schematic band alignment of the Te/MoS2 heterojunction under reverse bias (Vds < 0) for three representative gate voltages: (i) Vg > 0, (ii) Vg = 0, and (iii) Vg < 0. The yellow dashed line indicates the position of the MoS2 conduction band edge We first recall the equilibrium band alignment and then consider how the applied gate voltage and bias voltage modify carrier injection at the junction. Figure 2a shows a schematic of the equilibrium band alignment in the Te/MoS2 heterojunction at zero gate voltage. In this configuration, the p-type Te and n-type MoS2 result in hole accumulation in Te and electron accumulation in MoS2. Hall measurements on representative Te flakes exhibit a typical natural hole doping on the order of 5 × 1018 cm−3 (ref. 48). At the interface, the redistribution of free carriers forms a depletion region (indicated by gray dashed lines), where band bending gives rise to a built-in electric field. Previous reports determined the Te and MoS2 work functions (ΦTe ≈ 4.75 eV and ΦMoS2 ≈ 4.3 eV) by UPS/KPFM41,49. Accordingly, the band diagram in Fig. 2a is constructed using these quantitative data and provides a quantitative basis for the equilibrium band alignment. Importantly, the large valence band offset at the Te/MoS2 interface serves as an effective energy barrier, preventing hole leakage from Te into MoS2. To rationalize the gate‑dependent band diagrams in Fig. 2d, it is important to consider how the back gate voltage couples to the Te and MoS2 layers in this vertical geometry. The Te/MoS2 heterojunction and the back silicon electrode form a capacitor. Because the MoS2 layer is relatively thin and lies directly on the gate dielectric outside Te, the back-gate voltage primarily modulates the carrier density in the edge regions of MoS2 rather than in the region where MoS2 overlies Te. In this geometry, the gate field effectively penetrates the full thickness of the ~17 nm MoS2 channel and tunes its Fermi level, while the much thicker Te is strongly screened and remains only weakly affected by the back gate. This diagram of the gating effect is consistent with the EL spatial map shown in Fig. 1d. The EL-active region extends across the Te/MoS2 overlap area. Although the back gate couples most efficiently to the MoS2 regions that are not covered by Te, carriers can be efficiently injected into the whole overlap area with Te, which contributes to the EL emission from the overlapped area. Consequently, back-gating provides an efficient handle to tune the effective band alignment and injection across the Te/MoS2 interface and thus the EL output from the overlap region. For Vg = 0 V and +30 V, a finite reverse-bias current persists at sufficiently negative Vds, which can be attributed to band-to-band tunneling. As illustrated in Fig. 2d(i, ii), the combined effect of the drain-source bias Vds and the back-gate voltage Vg bends the bands such that the conduction band minimum of MoS2 lies below the valence band maximum of Te, resulting in a type-II band configuration. When the magnitudes of Vds are fixed in Fig. 2d, positive Vg shifts the MoS2 conduction band downward in energy relative to Te and increases the overlap between the MoS2 conduction band and Te valence band. This effect enhances the tunneling probability and produces a larger reverse current at Vg > 0 than at Vg = 0, which is consistent with the transport data shown in Fig. 2b. In contrast, for Vg < 0 (Fig. 2d(iii)), the MoS2 conduction band is shifted upward in energy relative to the Te valence-band edge, so the band-to-band tunneling window effectively closes, and the reverse current is strongly suppressed. The transfer characteristics (Ids-Vg) measured at Vds = 4 V confirm that the device exhibits pronounced rectifying behavior, with the current strongly dependent on gate-induced carrier modulation. The threshold voltage (Vth) of the MoS2 FET was determined to be ~−7 V. This value was extracted from the logarithmic Ids-Vg transfer curve (Fig. 2c) as the gate voltage at which the drain current departs from the subthreshold regime and begins to increase rapidly. These results demonstrate that the Te/MoS2 heterostructure operates as a robust and gate-tunable p-n junction, with rectifying behavior and suppressed leakage arising from the built-in potential at the interface. These characteristics provide a consistent electronic framework for understanding the bias‑ and gate‑dependent electroluminescence discussed later. Mid-infrared electroluminescence and bias-dependent spectra of Te/MoS2 heterojunctions We next discuss the electroluminescence and underlying recombination mechanisms under different bias conditions. To directly correlate this band‑alignment diagram with light emission, we investigate how the mid‑infrared EL depends on the applied source-drain bias at a fixed gate voltage. Bias-dependent EL spectra (Fig. 3b) are obtained by sweeping the source-drain voltage (Vds) from −4 to 4 V while keeping the gate voltage fixed at Vg = 20 V. EL emission is observed only when the forward bias exceeds an EL turn-on voltage (Von). We define Von as the minimum Vds at which the integrated EL intensity clearly exceeds the background signal. Von is not a fixed constant but varies systematically with gate voltage. At Vg = 20 V, the spectra shown in Fig. 3b indicate that Von > 2 V and Von is accurately determined to be below 1.7 V using the higher sensitivity measurement described in Supplementary Note 7. Notably, the optical turn-on voltage (Von ~ 1.7 V) is larger than both the Te bandgap scale (Eg/q ~ 0.35 V) and the estimated built-in potential (~0.46 V). This is expected because Von reflects the bias required to reach sufficient interlayer injection and quasi-Fermi level splitting in the overlap region for detectable radiative recombination, together with additional voltage drops from contact/access series resistance. Near the onset, trap filling and defect-assisted nonradiative recombination can further delay the emergence of EL to larger voltage. For Vds above Von, the EL intensity increases steadily when the applied bias increases further. The EL peak position remains essentially unchanged under different carrier injection levels, indicating a stable underlying radiative transition. The linewidth shows a weak narrowing trend with increasing bias (Supplementary Note 12). As the emission becomes increasingly dominated by band-edge recombination, the relative contribution from the low-energy-side emission decreases, leading to a slight reduction in the full width at half maximum. The low-energy in-gap emission may originate from shallow impurity states, the Franz-Keldysh effect induced by the built-in electric field, and excitonic emission. Figure 3a illustrates the band alignment under forward bias: the applied voltage flattens the electron barrier on the MoS2 side and thereby lowers the effective electron injection barrier. Under these conditions, the lowered and narrowed interfacial barrier favors tunneling-assisted electron injection from MoS2 into Te. At 25 K, the thermal energy (kBT ≈ 2.15 meV) is much smaller than the tens of meV barrier scale, and no obvious high-energy spectral broadening is observed, indicating that thermally activated over-barrier injection is not the dominant mechanism. At the same time, the large valence band offset at the Te/MoS2 interface maintains a high barrier for holes in Te. These two conditions together ensure that radiative recombination is spatially localized within the Te layer, which is responsible for the observed mid-infrared emission. The spatial profiles of the quasi-Fermi levels under forward bias lend further support to this interpretation. Specifically, the electron quasi-Fermi level in MoS2 exhibits a gradual downward slope as it approaches the interface, confirming efficient electron injection into Te. In contrast, the hole quasi-Fermi level in Te rises steeply near the boundary, forming a sharp gradient that signifies strong hole confinement and suppressed injection into MoS2. Under reverse bias, as shown in Fig. 2d(i), the applied voltage increases the built-in potential, thereby steepening the quasi-Fermi level gradients on both sides of the heterojunction. This enhancement strengthens carrier separation within the depletion region and effectively suppresses both electron and hole injection. As a result, radiative recombination is inhibited, leading to suppressed electroluminescence under reverse bias. These quasi-Fermi level profiles qualitatively explain the bias-dependent switch of EL emission. Long‑term operational stability data, showing that the device maintains robust mid‑infrared electroluminescence after extended ambient storage without any encapsulation, are provided in Supplementary Fig. S8. Band alignment and bias-dependent electroluminescence of the Te/MoS2 heterojunction. a Vds > Von (Vg = 20 V), illustrating the light-emission process. Blue and orange dashed lines denote quasi-Fermi levels for electrons and holes, respectively. b Bias-dependent EL spectra at Vg = 20 V (25 K). c Integrated EL intensity as a function of bias voltage Gate voltage tunable electroluminescence of Te/MoS2 Having identified the bias window over which efficient mid‑infrared EL occurs, we next explored how electrostatic gating of the MoS2 channel modulates the EL intensity at fixed forward bias. Since EL emission in the Te/MoS2 heterostructure arises from radiative recombination between electrons injected from MoS2 and holes localized in Te, the EL intensity is highly sensitive to the doping level of MoS2. In particular, gate tuning of the doping level of MoS2 modulates the interfacial band alignment, thereby influencing the efficiency of carrier injection into the Te layer and ultimately tuning the light emission intensity. Since the back gate has little effect on the quasi-ohmic Te/Pd contacts (Fig. S2), its modulation of the doping level in MoS2 can slightly affect the built-in potential of MoS2/Cr contact, and further modulates the effective voltage applied on the MoS2/Te heterojunction. However, because the Schottky barrier of MoS2/Cr contact is only tens of meV, the effect of the built-in potential changes by the doping is expected to be limited, providing a minor correction compared with the heterojunction band alignment that governs the gate-tunable EL and is therefore omitted from the schematic band diagrams for clarity50. To elucidate this effect, we measured the EL response as a function of back-gate voltage Vg at Vds = 4 V. Figure 4a displays the EL emission-energy map under a fixed lateral bias (Vds = 4 V), revealing a nonmonotonic intensity profile: the EL emission shows a maximum at a critical gate voltage (Vg-crit) of ~20 V. Gate voltage dependence of Te/MoS2 electroluminescence. a Plot of EL intensity as a function of photon energy and gate voltage Vg under a 4 V bias voltage. b Interfacial energy band alignment diagram of Te/MoS2 under critical gate voltages and threshold gate voltages, where diagrams in (i) and (ii) correspond to the energy bands of the yellow dashed lines in Fig. 4a. The blue and orange dashed lines in (i) and (ii) indicate the quasi-Fermi levels for electrons and holes, respectively. c Interfacial energy band alignment of Te/MoS2 across different gate voltage regions: (i) nonradiative states under Vg < −2.7 V, (ii) intermediate regime where Vg < Vg-crit, and (iii) radiative bands at Vg > Vg-crit. d Selected EL spectra at different gate voltages, ranging from 30 to −30 V, measured at 25 K. e Integrated intensity of the EL emission as a function of gate voltage (with fixed Vds = 4 V), measured at 25 K This trend results from competition between enhanced electron accumulation and the formation of a gate-induced effective interfacial injection barrier. The underlying mechanism is illustrated by the band diagrams in Fig. 4b, c. Specifically, Fig. 4b highlights two representative gating conditions that define the onset and maximum of EL emission, while Fig. 4c traces the full evolution of the equilibrium band structure across the entire gating range. Figure 4b(i) corresponds to the critical gate voltage (Vg-crit = 20 V), where the EL emission is maximized because enhanced electron accumulation in MoS2 improves the electron supply while the gate-induced interfacial injection barrier increases. The competition between these two effects reaches an optimum balance at Vg-crit. The quasi-Fermi-level profile in Fig. 4b(i) indicates that the MoS2 channel resistance is reduced at this gate bias due to a high carrier density, so that under the same applied Vds, a larger fraction of the voltage can drop across the junction region. Consistently, the electron quasi-Fermi level in MoS2 varies more gently toward the interface, which reflects a smaller series voltage drop in the MoS2 access region and thus more favorable injection conditions into Te. Meanwhile, the large valence-band offset at the Te/MoS2 interface blocks hole injection from Te into MoS2. Consistently, the hole quasi-Fermi level on the Te side varies sharply near the interface, indicating strong hole confinement. Figure 4b(ii) represents the threshold voltage (Vg-th = 0 V), where the gate-induced electron population in MoS2 first becomes sufficient to enable injection into Te, thereby lowering the effective electron injection barrier. In this case, the electron quasi-Fermi level exhibits a steeper spatial variation toward the junction, consistent with a larger series drop in MoS2 and injection that is only just initiated. Figure 4c illustrates the full evolution of the band alignment in the Te/MoS2 heterostructure, complementing the specific gating conditions in Fig. 4b to offer a more comprehensive view. Three distinct gate-controlled regimes are identified: (i) Vg < Vg-th, Fig. 4c(i), (the depleted regime): below the threshold (Vg-th), MoS2 is depleted of free electrons. The lack of available carriers quenches EL emission. (ii) Vg-th < Vg < Vg-crit, Fig. 4c(ii), (intermediate regime): as Vg increases above Vg-th, electron accumulation in MoS2 increases, improving the electron supply for injection. The gate reshapes the electrostatic band bending near the interface and can increase the effective electron injection barrier. In this gate range, the benefit from increased electron supply dominates over the barrier penalty, so the net electron injection into Te and the EL intensity still increase with Vg and reach a maximum at Vg-crit. (iii) Vg > Vg-crit, Fig. 4c(iii), (above 20 V, the high gate voltage regime): a higher gate voltage reshapes the electrostatic potential and band bending near the interface. This increases the effective electron injection barrier for carriers entering Te. The increased barrier suppresses electron injection into Te and becomes the dominant factor. It outweighs the benefit from additional electron accumulation, so the EL intensity no longer increases and decreases beyond Vg-crit. To gain further insight into the gate-dependent EL behavior, we investigated whether the underlying radiative transitions responsible for the observed EL remain constant across different gate voltages. As shown in Fig. 4d, the emission peak position remains fixed at 3.5 μm with negligible linewidth variation, indicating a stable recombination pathway governed by band-edge transitions in Te, even as the intensity varies. This confirms that gate modulation primarily affects the carrier injection efficiency rather than the fundamental nature of the recombination mechanism in Te. Together with the bias-invariant spectrum shown in Fig. 3b, these results underscore the robustness of Te band-edge transitions against electrical perturbations. In addition, the EL polarization state remains essentially unchanged during back‑gate sweeps at a fixed drain bias. As shown in Supplementary Fig. S10, the extracted degree of linear polarization (DOP) stays within ~0.58–0.59 over a broad gate range while the EL intensity varies, demonstrating a polarization‑locked intensity control enabled by electrostatic tuning. In contrast to materials such as BP, whose emission-energy shifts with applied vertical fields33, the gate bias in our Te/MoS2 heterojunction modulates EL intensity without inducing measurable changes in emission wavelength, demonstrating stable band-edge emission under field control. To further establish the gate dependence of EL, we measured the EL intensity as a function of Vg, as shown in Fig. 4e. As the gate voltage is swept from positive toward negative values, the EL intensity decreases rapidly and is effectively quenched. Based on the EL-Vg intensity curve shown in Fig. 4e, the EL threshold is more precisely identified as Vg-th = −2.7 V. The refined threshold provides a clear reference point for subsequent analysis of the current-dependent EL efficiency (see Fig. 5). Current-dependent electroluminescence in Te/MoS2 heterojunctions. Log-log plots of integrated EL intensity versus drain-source current density (Jds), with red and blue solid lines indicating power-law fits to different current regimes. a Variation under different source-drain bias, showing distinct low- and high-current-density scaling. b Variation under different gate voltages, exhibiting segmented power-law behavior with fitted exponents reported in the text. c External quantum efficiency (EQE) versus Jds; red and blue curves correspond to the bias sweep and gate sweep configurations in (a and b), respectively Driving-current dependence of mid-infrared electroluminescence To probe how the bias and gate voltage affect the internal quantum efficiency and carrier injection efficiency, we extract the differential power-law exponent k from log-log plots of IEL vs Jds(calculatedusingajunctionareaA~300μm2,Jds=Ids/A). Following the k-analysis commonly used for LEDs, the integrated EL intensity can be written as the product of the source-drain current term (Ids), the injection efficiency (ηinj), and the internal quantum efficiency (ηIQE)51–53, so that In the active-region-limited regime, this leads to where ηinj characterizes the effective injection efficiency of carriers that can cross the Te/MoS2 interface barrier and enter the Te recombination zone. This term accounts for current losses before radiative recombination in Te, including nonradiative current paths in the MoS2 access region and interfacial nonradiative recombination. k⊥ reflects the competition between radiative and nonradiative processes inside Te. In conventional LED recombination models, Shockley-Read-Hall (SRH) defect recombination, band-to-band radiative recombination, and Auger recombination correspond to characteristic values of k⊥≈2, 1, and 2/3, respectively54–56. Experimentally, we sweep Vds at a fixed gate voltage (Vg=20V, Fig. 5a) and sweep Vg at a fixed bias (Vds=4 V, Fig. 5b) to study the dependence of EL (IEL) on the drain current (Ids). Figure 5a shows IEL versus current density Jds on a double-logarithmic scale at Vg=20V, a bias chosen to maximize the EL signal-to-noise ratio. Between the EL onset at Jds≈0.05μAμm−2 and Jds≈0.4μAμm−2, the dependence is super-quadratic with an effective exponent k≈2.36. This large k indicates that defect-assisted SRH recombination in Te remains the dominant recombination channel55–57. As Vds increases, the Te/MoS2 barrier is lowered and thinned, increasing the fraction of the total current, Ids, injected into Te. In the framework of Eq. (1), this corresponds to a positive dlnηinj/dlnIds, which drives the measured exponent slightly above the SRH limit. For higher currents, 0.4μAμm−2≲Jds≲1.3μAμm−2, the slope decreases to a super-linear k≈1.29. In this regime, radiative recombination in Te becomes dominant (~1)56,57, and the injection efficiency continues to improve with increasing Vds, so k remains slightly above 1. To separate the effects of forward bias and gate control while maintaining a high signal-to-noise ratio, we next fix Vds=4V and tune Vg (Fig. 5b). When the gate-induced current lies in the range Jds≈0.03–0.3μAμm−2, IEL again follows a super-linear power law with k≈1.89. This value indicates that defect-related SRH recombination in the active region remains important, so k⊥ is close to the SRH limit (~2)55–57. In this regime, increasing Vg at fixed Vds = 4 V rapidly increases the MoS2 channel conductance and raises the effective Te/MoS2 junction barrier. A smaller fraction of Ids reaches the radiative zone, and a larger fraction is dissipated via nonradiative or interfacial dark paths outside the EL-active region. In the language of Eq. (1), ηinj decreases with Ids, and this negative trend lowers the measured k from the SRH-like value. For a higher current 0.3μAμm−2≲Jds≲1.3μAμm−2, the slope drops to k≈0.71. In this regime, radiative recombination in Te remains relevant56,57, but the gate-induced redistribution of potential promotes interfacial nonradiative recombination and other loss pathways outside the EL-active zone. As a result, ηinj decreases with increasing Ids, providing a negative contribution to dlnηinj/dlnIds and causing the measured k to fall below 1. At the largest drive currents (Jds≳1.3μAμm−2), the EL intensity eventually rolls over and decreases with increasing Jds, corresponding to an effective negative exponent in the IEL-Jds power-law relation. These trends can be consistently interpreted within a standard LED framework that separates injection efficiency from the internal quantum efficiency of the Te active region. In the low-current-density window (0.03–0.3μAμm−2), back-gating accumulates electrons in the MoS2 edge/access regions and lowers the effective injection barrier into Te, thereby suppressing the SRH-type losses in Te and increasing ηinj with current; together, these effects naturally yield super-linear exponents approaching the SRH-like limit. As Vg increases further, the junction enters a regime in which gate-induced band bending on the MoS2 side becomes substantial (cf. Fig. 2), a larger fraction of Ids is either diverted into interfacial nonradiative recombination and other leakage pathways outside the Te radiative zone or shunted through the MoS2 rather than recombining in Te. In other words, ηinj begins to decrease with current (dlnηinj/dlnIds<0), which by itself can drive the observed sub-linear behavior (k≈0.71) even if radiative recombination remains important in Te. At the highest drive currents (Jds≳1.3μAμm−2), the slope k dips and becomes negative. As the carrier density increases, high-order nonradiative processes such as Auger recombination become significant, and self-heating further enhances nonradiative loss. The gate-induced barrier, together with interfacial dark recombination and other leakage pathways, continues to reduce the fraction of Ids that reaches the EL-active region, so k⊥ tends toward the Auger value (~2/3)55,58,59, while dlnηinj/dlnIds remains negative; together, these factors drive the overall k below zero. To quantify the device efficiency, we evaluate the external quantum efficiency (EQE) of the Te/MoS2 heterojunction LED. EQE is defined as the ratio of photons emitted into the far field to electrons passing through the device, i.e., EQE=Φph/Ids/q. The photon flux Φph is obtained from the measured EL intensity after correcting for the calibrated throughput of the collection optics and detector, as detailed in Supplementary Note 5. Figure 5c shows the EQE as a function of drive current Ids for the same device, where the red curve corresponds to sweeping Vds at fixed Vg=20V and the blue curve corresponds to sweeping Vg at fixed Vds=4V. Within our IQE-injection-efficiency framework, the integrated EL intensity scales as IEL∝IdsηinjηIQE, which yields: Thus, k > 1 indicates that the EQE increases with current, k ≈ 1 marks the vicinity of the EQE maximum, and k < 1 signals the onset of efficiency droop. For the forward-bias sweep at fixed Vg=20V (red curve), the EQE continues to increase with Jds over the measured range, as expected from the super-linear exponents k > 1 (Fig. 5a). Under these conditions, a maximum EQE of ~0.10% is obtained. In contrast, for the gate sweep at fixed Vds=4V (blue curve), the EQE initially rises steeply with increasing Jds, reaches a peak (EQE ≈ 0.32%) near the current where k ≈ 1, and then decreases to ~0.10% as k drops below unity and eventually becomes negative (Fig. 5b). The shapes and peak positions of the EQE-Jds trace closely mirror the segmented k value behavior discussed above, providing a consistency check of the combined IQE-injection-efficiency picture. In this three-terminal heterojunction, the drain bias and the gate voltage act as two independent knobs for controlling the EL output. The drain bias mainly sets the overall forward-bias drive and thus the injection level, which naturally modulates the EL intensity. In contrast, the back gate electrostatically tunes the MoS2 carrier density, altering the resistance and, consequently, the injection current. However, the distinct EQE-Jds curves for Vds-sweep and Vg-sweep in Fig. 5c indicate that the back gate provides additional control over the device: it can redistribute the potential drop near the Te/MoS2 junction, thereby modifying the effective interfacial injection condition and the radiative probability per injected carrier. As a result, gating can shift the device into a higher-efficiency operating window and enhance the EQE without requiring a proportional increase in Jds. Notably, a peak EQE of ~0.32% is achieved in a simple planar device geometry without any optical cavity, reflective back contact, or waveguide-based light extraction. In light of recent reports on mid-infrared BP-based LEDs employing resonant cavities and reflective electrodes28,60, we anticipate that similar optical and interfacial engineering could further enhance the EQE of Te/MoS2 heterojunction LEDs. Discussion In summary, we demonstrate a gate-tunable Te/MoS2 van der Waals LED emitting at ~3.5 μm, a technologically important mid-IR wavelength, with linearly polarized output and spectra that remain invariant to bias and gate at 25 K. The device architecture enables electrostatic control of band alignment and injection efficiency, while a segmented k-slope analysis identifies injection-efficiency droop as the principal limitation at high drive. These results position the Te/MoS2 heterostructure as a polarization-sensitive MIR source compatible with heterogeneous integration, with immediate implications for on-chip sensing, spectroscopy, and communications in the 3–5 μm window. In particular, polarization-sensitive MIR emission and the potential for integration with mature photonics platforms open opportunities for compact, chip-scale spectrometers, trace-gas sensing, environmental monitoring, and polarized MIR communication systems that leverage atmospheric transmission windows. Looking ahead, optimizing growth methods to produce high-quality Te crystals with lower defect densities and cleaner interfaces will minimize nonradiative recombination and improve carrier-injection efficiency, thereby significantly enhancing quantum efficiency. Building on these material advances, higher brightness and more stable operation could be pursued through contact and barrier engineering, thermal management, and pulsed operation to enable room-temperature performance. Photonic confinement strategies, such as microcavities, waveguide-coupled resonators, and photonic crystals, could further enhance out-coupling. Such photonic designs could improve on-chip compatibility by redirecting the emission of the present simple LED geometry into guided or waveguide-coupled modes27,61. Realization on CMOS-friendly substrates and co-integration with Si or other MIR photonics platforms would enable lab-on-a-chip sensing, portable MIR spectrometers, and field-deployable LIDAR or free-space links62–64. Beyond this system, the segmented k-slope framework provides a generalizable toolkit for vdW emitters and can guide the rational design of reconfigurable, on-chip MIR optoelectronics, including multi-color or tunable polarization-enabled sources29,65. Materials and methods Sample preparation Te nanoflakes were grown through a hydrothermal method38,39. Initially, 3 g of polyvinylpyrrolidone (PVP, molecular weight = 58,000) was dissolved in 32 mL of deionized (DI) water. Subsequently, 92 mg of Na2TeO3 was put into the PVP solution with continuous stirring. Next, 3.32 mL of ammonium hydroxide solution (25–28%, wt/wt%) and 1.68 mL of hydrazine hydrate (80%, wt/wt%) were added to the mixture. After five minutes of magnetic stirring, the solution was transferred into a 50 mL Teflon-lined stainless-steel autoclave and maintained at 180 °C for 10 h. The resulting product was washed thoroughly with deionized water to remove any residual ions. For device fabrication, the Te nanoflakes were redispersed in ethyl alcohol and drop-cast onto a 285-nm SiO2/Si substrate. Multilayer MoS2 flakes were exfoliated using polydimethylsiloxane (PDMS) and transferred onto selected Te flakes via a dry transfer method. Electrode fabrication was carried out in two sequential electron beam lithography (EBL) steps. First, electrodes were patterned on the Te flake, followed by electron beam evaporation of 20 nm Pd and 100 nm Au (Pd/Au). In the second EBL step, electrodes were defined on the MoS2 region, and a 20-nm Cr and 100-nm Au (Cr/Au) metal stack was deposited using the same electron beam evaporation process. Each lithography and deposition step was followed by a standard lift-off process in acetone. Mid-infrared EL measurement EL measurements were performed using a self-built micro-EL setup. The sample was mounted in an open‑loop cryogenic thermostat providing a vacuum environment and temperature control from 25 to 300 K via a liquid helium flow. The cryostat was fixed onto a PDV XY‑50 motorized XY scanning stage to enable spatially resolved measurements. A × 40 reflective objective lens with a numerical aperture of 0.5 was used to collect the EL emission, which was directed into a spectrometer (Princeton Instruments SP2500i) equipped with a 150 g mm−1 grating. The signal was detected using a liquid-nitrogen-cooled single-channel InSb photodetector. To improve the signal-to-noise ratio, the EL signal was modulated by an optical chopper at 521 Hz and demodulated using a lock-in amplifier phase-locked to the chopper reference. The optical chopper was placed in the collection path. During the measurements, the electrical bias was applied using a Keithley 2612B dual-channel source meter, with one channel supplying the back-gate voltage and the other applying the drain-source bias. PL and Raman measurement For photoluminescence (PL) measurements, the same low-temperature cryogenic setup, motorized XY stage, and lock-in detection scheme described for EL were used. A 1064 nm continuous-wave laser was employed as the excitation source. The PL signal was collected by the same objective and detected using the same InSb photodetector. For Raman spectroscopy, a 532-nm laser was used for excitation, and the scattered signal was collected by the same spectrometer with the same grating and detected using a CCD camera. Supplementary information Supplementary Information