Up-Conversion Electroluminescence (UCEL)
- UCEL is a phenomenon where the emitted photon energy exceeds the electrical bias, achieved by multi-electron and many-body storage mechanisms.
- Recent studies demonstrate UCEL in varied regimes—from single-molecule STML to 2D heterostructures—using dark states, triplet shelving, and excitonic Auger processes.
- Key design strategies involve suppressing nonradiative quenching, precise state alignment, and engineered interfaces to control energy-accumulation pathways.
Up-conversion electroluminescence (UCEL) denotes electroluminescence in which the emitted photon energy exceeds the electrical energy supplied per injected or tunneling electron, expressed as in scanning tunneling microscope-induced luminescence (STML) and as in vertical tunneling devices (Rai et al., 15 Aug 2025, Svatek et al., 2019). Conventional electroluminescence satisfies ; UCEL therefore requires multi-electron, energy-storage, or many-body pathways that accumulate energy in molecular or excitonic states before radiative decay (Rai et al., 15 Aug 2025, Binder et al., 2019). Recent work spans four experimentally distinct UCEL regimes: adsorption-geometry-gated charge-injection cascades in a single radical molecule, triplet-shelving UCEL in a plasmonic STM nanocavity, sequential inelastic tunneling through a supramolecular monolayer in a van der Waals tunnel diode, and excitonic Auger upconversion from dense interlayer-exciton reservoirs in type-II transition-metal dichalcogenide heterostructures (Rai et al., 15 Aug 2025, Zheng et al., 10 Sep 2025, Svatek et al., 2019, Binder et al., 2019).
1. Energetic criterion and microscopic requirements
The defining energetic inequality of UCEL is that the photon energy exceeds the single-electron electrical energy budget. In STML, the condition is , while in the molecular/2D tunneling-diode literature it is written as (Rai et al., 15 Aug 2025, Svatek et al., 2019). This distinction in notation does not alter the physical requirement: single-electron excitation of the emissive state is energetically forbidden, so the junction must realize a relay mechanism.
In the reported systems, the relay takes one of three forms. In single-molecule STML of radicals and heavy-metal porphyrins, the relay is a sequence of transiently charged and neutral excited states, often including dark doublet or triplet intermediates (Rai et al., 15 Aug 2025, Zheng et al., 10 Sep 2025). In a supramolecular PTCDI tunnel diode, the relay is an intermediate triplet state reached by inelastic electron scattering and then promoted to the emissive singlet by a second inelastic event (Svatek et al., 2019). In WSe/MoS heterostructures, the relay is not a single molecular state but a dense reservoir of long-lived interlayer excitons whose mutual collisions activate excitonic Auger upconversion into intralayer excitons (Binder et al., 2019).
Elementary energy conservation remains local even when the net emission is overbias. For the charge-injection mechanism in VOPc, each charge-addition or charge-removal step obeys , where is the energy difference between the initial and final molecular states involved in a charge addition/removal step (Rai et al., 15 Aug 2025). In the PTCDI tunnel diode, inelastic excitation rates follow Fermi’s golden rule, 0, and the threshold for triplet excitation is reached when 1 (Svatek et al., 2019). In the TMD heterostructure, the reservoir kinetics are summarized by
2
with UCEL intensity
3
making explicit that overbias emission can emerge from quadratic many-body loss channels rather than direct one-particle injection (Binder et al., 2019).
2. Adsorption-induced gating in a single radical molecule
A particularly direct realization of UCEL control is provided by single vanadyl phthalocyanine (VOPc) molecules adsorbed on three monolayers of NaCl on Au(111) and studied at 4 K with STML and DFT/TD-DFT (Rai et al., 15 Aug 2025). VOPc is a neutral, stable radical with total spin 5, a doublet ground state 6, and an unpaired electron localized predominantly in a vanadium-centered orbital of the vanadyl unit. On NaCl/Au(111), the molecule adopts two adsorption geometries: O-up, in which the vanadyl oxygen points away from the surface and appears lower in STM topographs, and O-down, in which the oxygen points toward the surface and appears higher. The two forms are reversibly interconverted by tip manipulation, consistent with identical chemical composition. The NaCl layer suppresses direct hybridization and nonradiative quenching to the metal while still allowing charge injection through the molecule. TD-DFT places the two lowest neutral excited doublet states 7 at 8–9 eV with very small oscillator strengths, and bright neutral excited states 0 at 1 eV that produce the observed 2 band. The cation 3 has a triplet ground state 4, and in the O-down geometry additional low-lying triplet excited states with very low oscillator strengths, labeled 5, appear only in O-down.
The spectroscopic consequence is geometry-gated UCEL. At sufficiently negative bias, both geometries show a 6 band at 7 eV and an 8 band at 9 eV. The 0 peak is narrow in O-down, with Lorentzian width 1 meV, and broader in O-up, 2 meV; the 3 band is instrument-limited at 4 meV. Differential conductance shows the positive ion resonance at 5 V and negative ion resonances at 6 V in O-up and at 7 V in O-down, with onset at 8 V. Bias-dependent photon counts then separate ordinary EL from true upconversion: in O-up, no emission occurs for 9 V 0, and both 1 and 2 appear only for 3 V 4 V and above, where 5 is equal to or exceeds the photon energies. In O-down, by contrast, 6 emission appears already at 7 V, exactly at the positive ion resonance onset; since 8 eV and 9 eV, the condition 0 is satisfied. The current dependence follows 1 with 2 for both geometries and is closer to 3 in O-down, while second-order photon correlation measurements show antibunching and a slower emission cycle for O-down with a fitted time constant of about 4 ns; O-up shows no antibunching (Rai et al., 15 Aug 2025).
The mechanism is a purely charge-injection cascade gated by adsorption-induced spin interaction with the substrate. The reported sequence is:
- Step 1: 5 at 6 V by electron removal to the tip.
- Step 2: 7 by electron capture from the substrate, producing a dark neutral excited state.
- Step 3: 8 by a second electron removal; this is accessible only in O-down because the low-lying dark cationic triplets are absent in O-up.
- Step 4: 9 by electron capture.
- Step 5: Radiative decay 0, yielding the 1 photon at 2 eV.
Quadratic-response TD-DFT further shows that the summed oscillator strength for 3 higher neutral states within the experimental window is 4 in O-down versus 5 in O-up, dominated by 6. The study attributes the gating of UCEL to a reordering of excited states and enhanced excited-state transition probabilities caused by altered screening of the vanadyl unit in O-down. It also states that there is no experimental indication for inelastic energy transfer being needed for UCEL, that plasmonic assistance is not required to explain the band positions or bias thresholds, and that vibrational sidebands modulate line shapes but are not the energy source for upconversion (Rai et al., 15 Aug 2025).
3. Triplet shelving in a plasmonic STM nanocavity
A second single-molecule implementation uses Pd-octaethylporphyrin (PdOEP) decoupled from Ag(111) or Ag(100) by an ultrathin NaCl layer and measured in a plasmonic STM nanocavity at 7 K (Zheng et al., 10 Sep 2025). On 8 ML NaCl/Ag(111), the 9 spectrum shows a positive ion resonance onset at 0 V and a negative ion resonance at 1 V. STML resolves two narrow electronic transitions: 2 fluorescence at 3 nm (4 eV), with FWHM 5 meV, and 6 phosphorescence at 7 nm (8 eV), with FWHM 9 meV. The linewidths imply lower bounds 0 fs and 1 ps, while the 2 lifetime is estimated to lie in the 3 ps–4 ns range after including charge-exchange quenching. The tip–sample junction provides wavelength-dependent Purcell enhancement; the local plasmon spectrum is strong between 5–6 eV, favoring 7 visibility, and weaker at 8 eV, so plasmon normalization is required to compare 9 and 0 intensities. After normalization, the 1 peak area ratio is 2.
UCEL is observed for the singlet line when the sample bias magnitude drops below 3 V. Bias scans from 4 to 5 V show that the 6 peak persists across the transition from energy-allowed excitation to UCEL, and at 7 V the inequality 8 is satisfied. Simultaneous monitoring of 9 and 00 from 01 to 02 V shows both lines present at all voltages, with an approximately constant intensity ratio across the UCEL transition at fixed current 03 pA. This behavior supports a common threshold governed by the positive ion resonance and a pumping route in which 04 occupancy mediates subsequent population of 05 even when direct 06 excitation is energetically inaccessible (Zheng et al., 10 Sep 2025).
The proposed pathway is triplet-mediated upconversion via charged relay states. Electron extraction at the positive ion resonance produces 07. Substrate-mediated neutralization then yields 08 or 09; the 10 state is not directly accessible from 11 in the UCEL window due to energy conservation. The triplet can decay radiatively to 12 or be quenched by charge exchange through 13 or 14. The decisive UCEL step is electron-assisted upconversion from 15, where the excess electron tunnels to the tip and promotes 16. Direct 17 spin-flip by a tunneling electron and direct 18 spin flip are described as inefficient here.
A minimal steady-state rate model for the ground, triplet, and singlet populations 19, 20, and 21 uses four effective rates 22–23: 24
25
26
with 27 and 28, 29. A central consequence is
30
so the singlet-to-triplet population ratio is proportional to current because 31 while 32 is constant. Joint fits reproduce superlinear 33 scaling and near-linear 34 scaling, with exponents 35–36 for 37 at 38 V over 39–40 pA and 41–42 for 43 at 44 V over 45–46 pA. The best-fit parameters are
47
48
The study interprets these results as verification of a triplet-mediated UCEL mechanism in which 49 acts as a shelving state and a subsequent electron event upconverts 50 to 51 via 52 (Zheng et al., 10 Sep 2025).
4. Inelastic-tunneling UCEL in a molecular/2D hybrid diode
A solid-state UCEL platform is realized by embedding a hydrogen-bonded supramolecular monolayer of perylene tetracarboxylic di-imide (PTCDI) within hexagonal boron nitride tunnel barriers and contacting it with few-layer graphene, forming a vertical FLG/hBN/PTCDI/hBN/FLG heterostructure (Svatek et al., 2019). The upper hBN barrier is thin, with thickness 53 nm, and the lower hBN barrier is 54–55 hBN monolayers. PTCDI forms planar monolayer islands stabilized by hydrogen bonding, with AFM lattice vectors 56 nm and 57 nm subtending 58, island sizes 59–60 61m, and 62 surface coverage under the reported deposition conditions. Under applied 63, electrons tunnel across the junction and excite molecular states by inelastic scattering.
The optical signatures resolve both singlet and triplet electroluminescence. The dominant visible EL peak is the 64 65–66 transition at 67 nm (68 eV), accompanied by a 69–70 vibronic satellite at 71 nm (72 eV). A weak additional visible peak appears at 73 nm (74 eV). Direct triplet electroluminescence is observed at 75 nm (76 eV), absent in PL. Using these measured EL energies, the singlet–triplet gap is 77 eV. The visible singlet emission persists down to 78 V at 79 K, giving 80 eV for the 81 eV line, and to 82 V at room temperature, giving 83 eV; using the 84 eV feature extends the inferred upconversion at room temperature to 85 eV. The work therefore reports UCEL “up to 86 eV” (Svatek et al., 2019).
The mechanistic picture is sequential inelastic excitation via an intermediate triplet. When the electron energy gain satisfies 87, inelastic tunneling excites the molecule from 88 to 89. A subsequent tunneling electron then excites a 90 molecule to 91. Radiative decay yields both 92 visible EL and direct 93 near-IR EL. A broad peak in 94 near 95 V matches the triplet excitation energy scale and is used as evidence for the inelastic scattering threshold associated with 96. The conceptual rate-equation framework is
97
98
Here 99 is negligible under much of the UCEL range because 00, so the singlet population is fed predominantly from the triplet manifold (Svatek et al., 2019).
Several diagnostics constrain alternative interpretations. EL peak energies remain effectively bias-independent for both polarities, supporting emission from discrete molecular levels rather than band-to-band processes. A log–log plot gives 01 with 02 for positive bias and 03 for negative bias, which rules out triplet–triplet annihilation in the low-triplet-density regime, where quadratic or higher dependence would be expected. Thermal mechanisms are excluded because UCEL persists with 04, and rapid vibronic excitation of 05 is excluded because the intermediate must persist over average intervals of 06 07s per electron per molecule at 08 pA/nm09. Plasmon-assisted emission is described as unlikely because strong local plasmons are improbable with few-layer graphene contacts and graphene plasmon energies are in the mid-IR rather than the visible. The overall photon-per-electron efficiency is 10–11 photons per injected electron, comparable to STML experiments (Svatek et al., 2019).
5. Excitonic Auger UCEL in type-II van der Waals heterostructures
In WSe12/MoS13 type-II heterostructures, UCEL is obtained not from sequential charging of a single emitter but from carrier-carrier interactions within a dense interlayer-exciton reservoir (Binder et al., 2019). The device stack is Gr / 14–15L hBN / 16L WSe17 / 18–19L hBN / 20L MoS21 / 22–23L hBN / Gr on Si/SiO24, with top and bottom graphene acting as tunnel electrodes. Under bias 25 above the interlayer threshold, holes are selectively injected into the WSe26 valence band and electrons into the MoS27 conduction band. Because the carriers are injected into different layers, direct intralayer recombination is avoided at sub-threshold biases and interlayer excitons accumulate at the interface. The conduction-band minimum resides in MoS28, the valence-band maximum in WSe29, and the reported offsets are 30–31 eV, 32–33 eV, with an interlayer gap 34–35 eV.
The interlayer excitons are spatially indirect and, because of lattice mismatch and small twist, also 36-space indirect. The momentum mismatch is estimated as 37 for zero twist and 38 for 39 twist, so radiative recombination requires momentum assistance. The interlayer exciton emission energy exhibits a linear Stark blueshift with applied electric field,
40
with slope 41–42 meV/V and extrapolated zero-injection interlayer-exciton energy 43 eV. Introducing a monolayer hBN spacer increases the electron-hole separation, suppresses radiative recombination, and enables a much larger exciton density (Binder et al., 2019).
This density increase activates excitonic Auger upconversion. Without a spacer, electroluminescence shows interlayer-exciton EL at 44–45 eV appearing first at 46 V, then WSe47 intralayer EL at 48 eV near 49 V, and MoS50 intralayer EL at 51 eV near 52 V. With a monolayer hBN spacer, intralayer EL from both WSe53 (54 eV) and MoS55 (56 eV) appears already at 57 V, well below the monolayer excitonic thresholds and corresponding to an upconversion of 58 eV relative to the interlayer-exciton EL. The onset voltages for WSe59 and MoS60 UCEL are similar, and UCEL dominates over interlayer-exciton EL in spacer devices (Binder et al., 2019).
The proposed process is an excitonic Auger collision between two interlayer excitons. One exciton annihilates nonradiatively and transfers its energy and momentum to the other, promoting it into the excitonic continuum with near-delocalized character. From this continuum, relaxation can return to the interlayer-exciton ground state, or proceed into MoS61 intralayer excitons followed by radiative recombination at 62 eV, or into WSe63 intralayer excitons followed by radiative recombination at 64 eV. Because the continuum contains many excited excitonic states in both layers, the two intralayer channels occur with comparable probability, explaining the nearly equal onset and intensity of WSe65 and MoS66 upconverted EL. Independent density estimates give 67 cm68 for a spacer device and 69 cm70 for a device without spacer. The mean interlayer-exciton spacing
71
is 72 nm in the spacer device and 73 nm without spacer; with 74 nm, this gives 75 versus 76, matching the emergence of many-body Auger processes only in spacer devices (Binder et al., 2019).
The evidence is explicitly framed against alternative explanations. Selective electrical injection excludes optical nonlinearities such as two-photon absorption. The equal onset and comparable intensity of WSe77 and MoS78 UCEL contradict a single-particle Auger picture, which would favor electrons over holes, and device A1 shows a superlinear near-quadratic rise of EL with current in the UCEL regime, transitioning to nearly linear behavior when direct intralayer injection dominates. Hot-carrier EL and trap-assisted recombination do not explain the equal upconversion in both layers at the same threshold or the dependence on the hBN spacer (Binder et al., 2019).
6. Comparative mechanisms, misconceptions, and design constraints
The reported UCEL platforms differ sharply in microscopic mechanism, but they can be organized by the nature of the energy-storage state or reservoir (Rai et al., 15 Aug 2025, Zheng et al., 10 Sep 2025, Svatek et al., 2019, Binder et al., 2019).
| Platform | Intermediate or reservoir | Characteristic UCEL signature |
|---|---|---|
| VOPc on 3 ML NaCl/Au(111) (Rai et al., 15 Aug 2025) | Dark 79 and low-lying dark cationic triplets 80 in O-down | 81 emission at 82 eV appears already at 83 V |
| PdOEP on 3 ML NaCl/Ag (Zheng et al., 10 Sep 2025) | 84 shelving state with 85 relay states | 86 at 87 eV persists for 88 V |
| FLG/hBN/PTCDI/hBN/FLG (Svatek et al., 2019) | Intermediate triplet 89 excited by inelastic tunneling | Visible EL at 90 eV down to 91 V at 92 K |
| WSe93/hBN/MoS94 (Binder et al., 2019) | Dense interlayer-exciton reservoir enabling excitonic Auger scattering | Intralayer EL at 95 and 96 eV already at 97 V |
A recurrent misconception is that overbias emission necessarily implies plasmonic assistance. The available evidence is system-specific and does not support that generalization. In VOPc, plasmonic assistance is not required to explain the band positions or bias thresholds, and there is no experimental indication that inelastic energy transfer is needed for UCEL (Rai et al., 15 Aug 2025). In the PTCDI tunnel diode, plasmon-assisted emission is described as unlikely because few-layer graphene contacts do not provide visible plasmons of the relevant scale (Svatek et al., 2019). In PdOEP, by contrast, the plasmonic nanocavity clearly reshapes radiative rates and detection efficiency through Purcell enhancement, but the microscopic UCEL pathway is still triplet-mediated and charge-exchange-driven rather than a purely plasmonic energy-source mechanism (Zheng et al., 10 Sep 2025).
A second misconception is that UCEL in organics is generically triplet–triplet annihilation. The reported systems show otherwise. The PTCDI monolayer yields nearly linear 98 versus 99, excluding TTA as the 00 step (Svatek et al., 2019). The PdOEP single-molecule case is reproduced by a one-molecule rate model with two-electron character encoded in 01 and 02, while a preliminary dimer result with nearly linear 03 is specifically interpreted as potentially involving cooperative TTA-like physics not present in the single-molecule model (Zheng et al., 10 Sep 2025). The TMD heterostructure case is instead a many-body excitonic Auger process, and the VOPc radical case is a charge-injection cascade through dark cationic triplets (Binder et al., 2019, Rai et al., 15 Aug 2025).
Taken together, these studies suggest three recurrent UCEL design conditions. First, nonradiative quenching must be sufficiently suppressed: 04 ML NaCl is crucial for VOPc and PdOEP, while hBN barriers or spacers are essential in PTCDI and WSe05/MoS06 devices (Rai et al., 15 Aug 2025, Zheng et al., 10 Sep 2025, Svatek et al., 2019, Binder et al., 2019). Second, the system must contain either a long-lived shelving state or a dense dark reservoir: 07 and 08 in O-down VOPc, 09 in PdOEP and PTCDI, or interlayer excitons in the TMD heterostructure. Third, the level alignment must permit sequential energy accumulation without opening faster nonradiative loss channels. This suggests that UCEL optimization is fundamentally a problem of engineering state ordering, charge-exchange rates, and radiative versus nonradiative competition.
The reported limitations are equally consistent across platforms. Low temperatures are central in the STML and TMD demonstrations, although PTCDI UCEL persists at room temperature (Rai et al., 15 Aug 2025, Zheng et al., 10 Sep 2025, Binder et al., 2019, Svatek et al., 2019). Device performance remains sensitive to interface quality, voltage drop across the active region, twist angle in TMD stacks, and the exact balance between state lifetime and tunneling rate. In radicals, adsorption geometry can switch UCEL on and off through substrate screening of the unpaired spin, and the work on VOPc explicitly proposes that this mechanism should generalize to other neutral radicals with a separated unpaired spin center and a decoupled chromophore (Rai et al., 15 Aug 2025). In heavy-metal porphyrins, the visible placement of both 10 and 11 lines enables quantitative validation of triplet-mediated UCEL. In molecular/2D diodes and TMD stacks, scalable vertical architectures provide electrical access to triplet or excitonic reservoirs without requiring the single-tip geometry of STML (Zheng et al., 10 Sep 2025, Svatek et al., 2019, Binder et al., 2019).