Efficient solid-state infrared-to-visible photon upconversion on atomically thin monolayer semiconductors

Abstract

Upconverting infrared light into visible light via the triplet-triplet annihilation process in solid state is important for various applications including photovoltaics, photodetection, and bioimaging. Although inorganic semiconductors with broad absorption and negligible exchange energy loss have emerged as promising alternative to molecular sensitizers, currently, they have exclusively suffered from low efficiency and contained toxic elements in near-infrared (NIR)–to–visible upconversion. Here, we report an ultrathin bilayer film for NIR-to-visible upconversion based on atomically thin two-dimensional (2D) monolayer semiconductors. The atomic flatness and strong light absorption of 2D monolayer semiconductors enable ultrafast energy transfer and robust NIR-to-visible emission with a high upconversion quantum yield (1.1 ± 0.2%) at modest incident power (260 mW cm⁻²). Increasing layer thickness adversely quenches the upconversion emission, highlighting the 2D advantage. Considering the whole library of 2D semiconductors, the facile large-scale production and the ultrathin solid-state architecture open a new research field for solid-state upconversion applications.

2D monolayer semiconductors enable efficient photon upconversion through triplet sensitization and triplet-triplet annihilation.

INTRODUCTION

Photon upconversion (UC) is a process of converting multiple low-energy photons to a higher-energy one, which has potential applications in photonics, energy conversion, and biological fields (1, 2). In particular, the UC from infrared (IR)–to–visible light is inspiring and promising as it enables sub-bandgap photon harvesting for third-generation photovoltaics, photochemical conversion, and low-cost IR photodetection (3–7). Among different UC mechanisms, triplet-triplet annihilation (TTA)–based UC has exhibited unique advantages and attracted strong interest because it can be achieved under low intensity and noncoherent irradiation (e.g., sunlight). To date, TTA-UC in solution mediated by molecular migration has been demonstrated extensively, and efficient near-IR (NIR) (>700-nm)–to–visible UC with a quantum yield [Ф_UC (8), the number of emitted UC photons per the number of absorbed ones with maximum value of 50%] of ~5.6% has been achieved (9–12). In contrast, the TTA-UC in rigid solid state, which is the prerequisite for photovoltaic or photodetection device integration and long-term operation, has remained a big challenge because of inefficient triplet diffusion and the complex interplay among different photophysical processes (13).

Solid-state NIR-to-visible UC based on traditional organometallic sensitizers has generally suffered from severe nonradiative recombination loss due to sensitizer aggregation and energy loss associated with singlet-to-triplet intersystem crossing, showing a consistently low Ф_UC of <0.3% (see table S1) (14, 15). A latest study demonstrates a special UC system of organic bilayer heterojunction with a record-high Ф_UC of ~2.5% (table S1) (16). There, triplets are generated by an interfacial charge transfer/recombination process, bypassing the conventional exchange energy loss. Alternatively, inorganic semiconductors, including lead chalcogenide nanocrystals and lead halide perovskite films, have recently been implemented as triplet sensitizers in solid-state NIR-to-visible UC devices (17–19). Compared to organic counterparts, inorganic semiconductor sensitizers exhibit a few distinct advantages including broad and tunable light absorption, small exchange energy, and better photostability (10, 11, 13, 20–23). However, the insulating organic ligands on a nanocrystal surface hinder exciton migration and the interfacial triplet energy transfer (TET) process, limiting the optimized nanocrystal thickness of one to two monolayers with extremely weak absorbance (~0.1%) and a low Ф_UC (0.53% for PbS) for conventional bilayer structured devices (17, 24, 25), although Ф_UC can be further improved to 1.6% by a complex multilayer structure with interference enhancement (table S1) (25, 26). In contrast to nanocrystals, lead halide perovskite bulk films, on the other hand, have a long carrier diffusion length; thus, the light absorption can be enhanced with increasing film thickness. Unfortunately, previous results have shown that increasing film thickness also equally increases energy loss due to light reabsorption and back energy transfer (18, 19, 27). Ф_UC is 0.15% for a 14-nm perovskite film (18) and even lower for thicker films (see table S1) (28). The soft and ionic nature of lead halide perovskites also makes perovskite/organic interface and UC performance volatile to the solution fabrication process (27, 29, 30). Moreover, both lead chalcogenide nanocrystals and lead halide perovskites contain Pb, which is toxic.

Here, we report a solid-state NIR-to-visible TTA-UC device based on atomically thin two-dimensional (2D) monolayer semiconductors, e.g., monolayer transition metal dichalcogenides (TMDs). Despite atomic thinness (~0.6 nm), the light-matter interaction of these monolayer semiconductors is particularly strong because of prominent excitonic resonance with giant oscillator strength (31, 32). For example, the lowest energy exciton peak of MoSe₂ and WSe₂ monolayers has an absorbance of 5 to 15% (32), which is orders of magnitude larger than that of monolayer PbS nanocrystals (~0.1%) (17) and comparable to that of ~45-nm-thick lead iodide perovskite films (~10% at 785 nm) (28). The energy splitting between singlet-like and triplet-like excitons for A exciton states in TMDs is also fairly small (<20 meV) (33, 34), allowing rapid spin flip for triplet transfer. Furthermore, the flat geometry and atomic smoothness of the 2D TMD monolayer allow close contact with annihilator molecules at ultimate proximity and facile integration into solid-state devices with ultrathinness. Recent spectroscopy studies have demonstrated efficient charge/energy transfer at the organic-TMD interface (35–40), inspiring us to explore TTA-UC on TMD monolayers here. In this study, we use model rubrene film doped with dibenzotetraphenylperiflanthene (DBP) as the NIR-to-visible annihilator-emitter system (17, 18) and MoSe₂ and WSe₂ monolayers with bandgaps larger than rubrene triplet-state energy [1.14 eV (17)] as NIR sensitizers. As shown in Fig. 1A, photoexcitation of TMD monolayers can produce triplets in rubrene through interfacial TET, and the encounter of two triplets in rubrene can generate a high-energy singlet state through TTA, which rapidly transfers to emissive DBP molecules through the Förster resonance energy transfer (FRET) process. Radiative recombination of DBP singlets produces anti-Stokes shifted photoluminescence (PL) in the visible range. The introduction of DBP suppresses the nonradiative loss of singlet state in solid-state rubrene by outcompeting the triplet pair separation and substantially boosts the PL efficiency (17, 18, 41).

Fig. 1. Scheme of the TTA-UC system sensitized by a monolayer semiconductor and the spectral characterization.

(A) Energy diagram and (B) schematic of the UC bilayer device showing the process of light absorption by a monolayer semiconductor, TET to annihilator rubrene layer, TTA in rubrene, and a subsequent FRET to DBP emitter, which emits upconverted emission. (C) Optical image of a representative Rub/MoSe₂ bilayer sample with the bilayer heterostructure region and pure rubrene region labeled. Scale bar, 5 μm. (D) Absorption (solid lines) and emission spectra (dashed lines, excitation wavelength of 405 nm) of the MoSe₂ monolayer and rubrene film, respectively. a.u., arbitrary unit.

RESULTS

We fabricated inorganic/organic bilayer UC samples (Fig. 1B) by mechanical exfoliation of MoSe₂ monolayers onto substrate and then spin coating a thin layer of rubrene film doped with a small amount (0.25%) of DBP on top (see section S1 for details). We refer to such rubrene/DBP binary film as “rubrene” or “Rub” hereafter for simplicity. The optical image in Fig. 1C shows the rubrene/MoSe₂ heterostructure and pure rubrene regions. The atomic force microscopy (AFM) image in fig. S1 indicates a highly smooth and compact rubrene film with a thickness of ~15 nm on both the MoSe₂ monolayer and fused silica substrate. The basic optical spectroscopy characterization including absorption and PL spectra (under 405-nm excitation) of the MoSe₂ monolayer and the rubrene film is shown in Fig. 1D. The MoSe₂ monolayer exhibits a distinct A exciton absorption/PL peak at ~780 nm, and the rubrene film shows a PL peak at 610 nm.

To explore the NIR-to-visible UC, we excited the rubrene/MoSe₂ bilayer device with a 772-nm continuous-wave (CW) laser and collected the PL spectra with a wavelength- and intensity-calibrated micro-PL setup (see section S2 for details). In contrast to the MoSe₂ monolayer with a distinct emission at ~780 nm, the rubrene/MoSe₂ bilayer under the same excitation conditions shows negligible MoSe₂ emission (Fig. 2A and fig. S2) but a strong upconverted emission at 610 nm with a tail extending to 850 nm, which corresponds to DBP emission. Control film of just a rubrene layer under the same excitation conditions shows no emission. This is a clear indication of the NIR-to-visible UC process. As the PL system response has been calibrated with intensity calibration to reveal the true sample response, despite the different wavelength range, the PL intensity of UC emission and MoSe₂ monolayer emission can be directly compared. In addition to energy gain, the integrated intensity of UC emission is more than two orders of magnitude larger than that from the MoSe₂ monolayer under the same excitation conditions.

Fig. 2. UC emission spectra and its power dependence under NIR CW excitation.

(A) PL spectrum (solid line) of the Rub/MoSe₂ bilayer device under NIR CW photoexcitation. The red vertical line denotes the excitation wavelength of 772 nm. Also shown in dashed line is the PL spectrum (with intensity multiplied by 10) of the MoSe₂ monolayer under the same excitation. (B) A log-log plot of the upconverted PL intensity as a function of incident power density under 772-nm CW excitation. It shows a transition from quadratic (slope ~ 2) to linear (slope ~ 1) regimes at ~260 mW cm⁻².

The UC emission by TTA mechanism has a unique dependence on the incident power density (I), which changes from quadratic (∝ I²) to linear (∝ I) with increasing I (42). This is due to the change of the dominant triplet decay pathway in rubrene, with primary monomolecular relaxation at low I and bimolecular TTA at high I. To confirm the TTA-UC process in rubrene/MoSe₂, we explored the upconverted emission intensity as a function of incident power density, and the results are plotted in Fig. 2B on a dual-logarithm scale. The dependence of upconverted PL intensity on I shows a characteristic quadratic-to-linear transition, with slope changing from 2.08 at small I to 0.92 at large I. This is a clear signature of UC through the TTA mechanism (42). The transition point of quadratic-to-linear dependence is an important parameter, i.e., the threshold power density (I_th), above which TTA efficiency saturates. Figure 2B shows an I_th of ~260 mW cm⁻² for the rubrene/MoSe₂ bilayer device under 772-nm CW light. By integrating the spectral range (600 to 800 nm) of AM1.5 solar irradiation where MoSe₂ can absorb but rubrene cannot, this I_th corresponds to an incident power of ~10 suns (see section S3 for details). Fundamentally, I_th depends on the triplet properties of annihilator molecules and depends inversely on the sensitizer absorbance at excitation wavelength and sensitization yield (42). Similar excitation wavelength (780 to 800 nm) and rubrene annihilator layer were used in previous UC devices on PbS nanocrystals (17) and lead iodide perovskite films (18), which allows a direct comparison across different NIR semiconductor sensitizers. As the sensitizer absorption increases with thickness, here, we chose reported UC devices of comparable thicknesses to highlight the strong light-matter interaction of atomically thin TMD monolayers. The I_th of the MoSe₂ monolayer is much smaller than that of monolayer PbS nanocrystals (~12 W cm⁻²) (17) and twofold less than that of the MA_0.15FA_0.85PbI₃ perovskite thin film with 14 nm thickness (~500 mW cm⁻²) (18). Considering the atomic thinness, this result demonstrates the superior properties of the TMD monolayer as an ultrathin absorber for light harvesting and triplet sensitization. We note, in perovskite UC devices, that increasing perovskite film thickness can further reduce I_th, and a record-low I_th of subsolar fluxes (~7.1 mW cm⁻²) has been achieved with perovskites of 380 nm thickness (19).

Another important parameter in TTA-UC is the UC quantum yield (Ф_UC), which is defined as the number of emitted UC photons per the number of photons absorbed (8). Following a previous method of determining the PL quantum yield of a MoS₂ monolayer (43), we determined Ф_UC by measuring the UC emission intensity from the rubrene/MoSe₂ bilayer device and excitation intensity from a diffuse reflector with a micro-PL setup, respectively (see section S4). A conservative estimate on three rubrene/MoSe₂ bilayer devices yields a Ф_UC of 1.1 ± 0.2% for rubrene/MoSe₂ at 772-nm excitation and an incident power of ~100 W cm⁻². This value should be considered as a lower limit (see section S4) and is already much higher than the Ф_UC for PbS nanocrystals (0.53%) (17, 24, 25) and MA_0.15FA_0.85PbI₃ perovskite thin film (0.15%) (18) and comparable to that in the latest organic bilayer heterojunction UC device {2.5% for rubrene/ITIC-Cl [3,9-bis(2-methylene-((3-(1,1-dicyanomethylene)-6,7-dichloro)-indanone))-5,5,11,11-tetrakis(4-hexylphenyl)-dithieno[2,3-d:2′,3′-d′]-s-indaceno[1,2-b:5,6-b′]dithiophene]} (16), putting it among the most efficient solid-state NIR-to-visible UC systems so far (see table S1).

To reveal the underlying photophysical processes and their time scales in the rubrene/MoSe₂ UC device, we turn to time-resolved (TR) PL measurements (see section S2 for details). Briefly, we excited samples using a 700-nm picosecond laser pulse, which excited MoSe₂ only and collected 750- to 850-nm emission by time-correlated single-photon counting (TCSPC). The TCSPC result from blank substrate confirms no scattered excitation light (fig. S3). As shown in Fig. 3A, the PL decay kinetics of MoSe₂ can be mostly fitted by a single exponential decay function with a lifetime of ~0.5 ns in a pristine monolayer and is obviously accelerated in a bilayer device. The almost overlap between MoSe₂ TRPL kinetics in bilayer and instrument response function (IRF) implies ultrafast (<20-ps) photoexcitation energy transfer from MoSe₂ to rubrene (see section S2 for details). This is consistent with the recent ~5-ps excitation transfer from ReS₂ to tetracene (35). Compared to PbS nanocrystals and lead iodide perovskite films with tens to hundreds of nanoseconds of excited-state carrier lifetime, which is beneficial for triplet sensitization, despite the fast exciton decay in the MoSe₂ monolayer, the photoexcitation energy transfer from MoSe₂ to rubrene is even faster and outcompetes the exciton recombination due to the ultimate contact between rubrene molecules and the MoSe₂ layer. Molecular triplet sensitization by inorganic semiconductors can occur through a concerted electron-hole transfer process (i.e., Dexter exchange mechanism) or a sequential charge transfer process (charge transfer–mediated mechanism) (2, 11, 13). Considering the exciton energy of the MoSe₂ monolayer (~1.59 eV) and rubrene triplet (1.14 eV) and their band alignment (Fig. 3B), the triplet sensitization of rubrene by MoSe₂ can occur through a one-step Dexter exchange process or sequential charge transfer with initial ultrafast hole transfer followed by electron transfer to rubrene triplet state (18, 44). A conclusive and unambiguous determination of interfacial charge/energy transfer process and associated time scales requires species-resolved ultrafast transient absorption measurements, which are quite challenging for such ultrathin bilayer sample and requires future studies.

Fig. 3. Photophysical process of rubrene/MoSe₂ UC emission.

(A) TRPL kinetics of MoSe₂ emission in a monolayer and the Rub/MoSe₂ bilayer, showing the acceleration of MoSe₂ PL decay by rubrene. (B) Schematic of the triplet sensitization process in rubrene/MoSe₂ through either the one-step Dexter exchange process or sequential charge transfer with initial ultrafast hole transfer followed by electron transfer to rubrene triplet state. CB, conduction band; VB, valence band; LUMO, lowest unoccupied molecular orbital; HOMO, highest occupied molecular orbital. (C) TRPL kinetics of upconverted emission under 700-nm excitation, showing slow rising and decay in microseconds due to the contribution of slowly diffusing and long-lived triplets. (D) UC PL intensity from the Rub/MoSe₂ heterostructure with different MoSe₂ layer numbers (1, 2, 3, and bulk-like). A thicker MoSe₂ layer shows a weaker UC emission. Inset shows the Rub/MoSe₂ samples with different MoSe₂ layer numbers. Scale bar, 5 μm.

To confirm the TTA mechanism for UC emission, the PL kinetics of upconverted emission is also measured. Representative TRPL kinetics is shown in Fig. 3C, and kinetics under a different power density or repetition rate of excitations is shown in fig. S4. Unlike the TRPL of MoSe₂ with instantaneous rising from direct photoexcitation, the kinetics of UC emission rises in nanoseconds and decays slowly in microseconds. A biexponential fitting on the rising process in Fig. 3C indicates a biphasic behavior with a fast component of τ₁ ~ 14 ns and a slow component of τ₂ ~ 261 ns. Moreover, the rising process becomes faster with increasing excitation power and repetition rate (i.e., increasing triplet concentration), consistent with previous results (17). According to Fig. 1A, the TTA-UC emission originates from the overall consequence of TET from MoSe₂ to rubrene, TTA, and then FRET, and radiative decay in rubrene. As shown above, TET occurs in picoseconds. The FRET and radiative decay in the rubrene film can be obtained from the TRPL kinetics by directly exciting rubrene, which shows a lifetime of ~4.6 ns (fig. S5). Therefore, the slow and strong concentration-dependent UC PL kinetics confirms the TTA process for UC emission, which depends on the encounter between two triplets. In that regard, the 14-ns fast-rising component can be attributed to bimolecular TTA at the rubrene/MoSe₂ interface where triplet concentration is high and the 261-ns slow component can be attributed to TTA between triplets which have diffused away from the interface in rubrene, similar to previously reported rubrene/perovskite UC devices (27). The long lifetime of triplets in the rubrene film leads to the slow decay of UC PL in microseconds.

A unique property of 2D TMDs is the facile tuning layer thickness with atomic precision. We investigated the UC performance of rubrene/MoSe₂ devices with different MoSe₂ layer numbers (Fig. 3D, inset). As demonstrated previously (45), increasing MoSe₂ layer thickness from monolayer to multilayer shifts the direct bandgap nature to an indirect one. Fortunately, the absorption of multilayer TMD is dominated mostly by direct transition with negligible contribution from the indirect one, while the layer thickness affects the only indirect transition. Therefore, the absorption of TMD increases addictively with increasing layer thickness. With increasing MoSe₂ layer numbers, the NIR absorption (in unit of percentage) increases almost linearly (fig. S6). On the contrary, the upconverted emission intensity decreases monotonically, with negligible UC on bulk-like MoSe₂ (Fig. 3D). In multilayer MoSe₂, the photogenerated excitons in the underneath layers have to transport along the out-of-plane direction between layers to the rubrene/MoSe₂ interface before TET, which has to compete with the fast exciton decay process in TMDs. On the other hand, increasing layer number increases back energy transfer from emissive singlets in rubrene to MoSe₂ (fig. S7), quenching UC emission. This highlights the dimension advantage of 2D monolayer semiconductors for ultrathin and efficient TTA-UC devices.

Last, to demonstrate the possibility of large-scale production for device integration, we fabricated the UC devices from chemical vapor deposition (CVD)–grown large-area MoSe₂ monolayer samples, including isolated triangles and continuous films by the same spin-coating approach. The optical image of the rubrene/MoSe₂ triangles and the corresponding UC (550 to 650 nm) PL image under 785-nm excitation are shown in Fig. 4 (A and B, respectively). Apparently, all MoSe₂ triangles exhibit uniform UC emission after coating a thin layer of rubrene. The same UC emission was also observed on MoSe₂ continuous films. The photographs of the MoSe₂ continuous film on a transparent substrate before and after rubrene spin coating are compared in Fig. 4C. The grayish film becomes pinkish after rubrene coating. The PL image of a ~100 × 100 μm² area shows a uniform and continuous UC emission across the whole area for the rubrene/MoSe₂ UC device (Fig. 4D). The UC PL inhomogeneity comes from the original inhomogeneous MoSe₂ film, including multilayer patches and unreacted impurities/substance from CVD growth (fig. S8).

Fig. 4. NIR-to-visible UC on large-area MoSe₂ monolayers.

(A) Optical image and (B) corresponding UC PL image of rubrene/MoSe₂ with large-area MoSe₂ isolated triangles. Scale bars, 10 μm. For PL imaging, the excitation light is 785 nm, and the collection wavelength range is 550 to 650 nm. (C) Photographs of the large-area MoSe₂ continuous film on ~1 cm × 1 cm fused silica substrate before and after spin coating of the rubrene film. (D) UC PL image of rubrene/MoSe₂ with the large-area MoSe₂ continuous film. (E) Comparison of integrated PL intensity from MoSe₂ monolayer emission and rubrene/MoSe₂ UC emission from three different types of samples (exfoliated flakes, isolated triangles, and continuous films) under the same excitation conditions. Note that the PL intensity of the MoSe₂ monolayer has been multiplied by 10.

In addition to energy gain, the PL intensity is also noteworthy for rubrene/MoSe₂ UC devices. We compared the integrated PL intensity of pristine MoSe₂ monolayers and rubrene/MoSe₂ UC bilayers for different types of MoSe₂ (exfoliated flakes, isolated triangles, and continuous films). The PL spectra were measured under the same NIR excitation condition with similar absorbance from MoSe₂; therefore, their intensities directly represent the relative emission quantum yields. Regardless of MoSe₂ type, the UC emission is always one to two orders of magnitude stronger than the emission from pristine MoSe₂ monolayers (Fig. 4E). This, together with faster MoSe₂ PL decay in the rubrene/MoSe₂ bilayer than that in the monolayer, suggests that the fast interfacial TET from MoSe₂ to rubrene can compete efficiently with the interior nonradiative recombination loss (e.g., defect trapping) in MoSe₂, generating upconverted PL with both energy and intensity gain. Furthermore, all three types of MoSe₂, despite largely varying PL intensity in the pristine monolayer, exhibit comparable UC emission intensity. This robust UC performance is in contrast to UC devices based on perovskite films that are volatile to the fabrication process and perovskite film conditions (27, 29, 30). Similarly consistent NIR-to-visible UC emission was also observed on WSe₂ monolayers (fig. S9). These results reveal the generality and robustness of TTA-UC on 2D monolayer semiconductors.

DISCUSSION

In summary, we demonstrate efficient and robust solid-state NIR-to-visible TTA-UC on 2D monolayer semiconductors. Compared to previously reported PbS nanocrystals and lead halide perovskites that generally require a rather thick film to achieve notable absorption, despite its atomic thinness, 2D layered materials have extraordinarily strong light absorption and layer-tunable electronic structures. This, together with the surface atomic flatness and close contact with acceptor molecules, enables efficient and robust UC at a modest excitation power in an ultrathin bilayer heterojunction device. The combined ultrathinness and the facile large-scale production render it to be readily integrated into optoelectronic and photonic devices. In addition to VIB-VIA TMDs (e.g., MoSe₂ and WSe₂ used in this study) with relatively weak interlayer interaction and slow out-of-plane transport, 2D layered materials with strong interlayer interaction (such as black phosphorus) have fast out-of-plane transport and layer-tunable bandgap from visible to IR. In view of the whole library of 2D layered semiconductors with a wide range of bandgap energies and broadly multicomponent 2D heterostructures as sensitizers and molecular annihilators with different energies covering the ultraviolet-to-IR range, we anticipate that our findings will open a new research field for solid-state UC and enable new applications of 2D semiconductors.

MATERIALS AND METHODS

Sample preparation

Fused silica substrates were cleaned by sequential sonication in detergent solution, deionized water, acetone, and isopropyl alcohol. Monolayers of MoSe₂ were mechanically exfoliated onto gel films (Gel-Pak) from bulk crystals (HQ Graphene) and then transferred on substrates. The CVD-grown MoSe₂ samples (continuous film on sapphire and isolated triangles on sapphire) were purchased from Shanghai OnWay Technology Co. Ltd. Monolayers were identified with an optical microscope and confirmed by their PL spectra.

Rubrene (99% gas chromatography; TCI) and DBP (98% high-performance liquid chromatography; Sigma-Aldrich) were purchased and used without further purification. The following procedure was performed in a nitrogen-filled glovebox to avoid oxygen. First, rubrene was dissolved in anhydrous toluene (Sigma-Aldrich) at 8 mg/ml and then mixed with DBP at a 0.25% molar ratio from a stock solution of 0.12 mg/ml. Then, the mixed solution was spin-coated on MoSe₂ at 6000 rpm for 20 s. Afterward, the bilayer device was annealed at 80°C for 10 min.

Steady-state micro-PL spectroscopy

To avoid the potential effect from oxygen, samples were sealed in a homemade nitrogen cell with thin transparent windows on both sides. Absorption and PL measurements were performed on a home-built microscope setup. The absorption spectra were measured using a white light supercontinuum laser (NKT Photonics, SuperK COMPACT) by normalizing the transmitted light from the sample on the substrate to that from the bare substrate. The transmitted light was collected and analyzed by a spectrograph (Princeton Instruments, SP2300) coupled with an electron-multiplying charge-coupled device (Princeton Instruments, ProEM1600). PL spectra were taken by the same spectrometer under CW laser excitation. The whole PL detection system was calibrated with both wavelength and intensity calibration (IntelliCal, Princeton Instruments). Specifically, the spectrometer wavelength was calibrated using the atomic emission lines from Hg and Ne/Ar lamps (Princeton Instruments, HGNE 2022), and the relative sensitivity versus wavelength was calibrated by an intensity calibration lamp (Princeton Instruments, LSVN-0358). In such a way, system response is removed, and the sample PL profile and intensity are recovered. PL spectra were taken under 772-nm CW laser excitation (with linewidth ~ 1 nm). The laser beam was focused on the sample through a 50× objective with a numerical aperture (NA) of 0.65. To take the full PL spectrum, the linear excitation light was filtered by a linear polarizer with a high extinction ratio (~100,000:1). To calculate incident power density, the photos of excitation laser beam on samples were taken with a camera, and the beam size is determined by 1/e² definition (width between the two points where the intensity is 1/e² of the peak value), from which we can calculate power density. The power density for steady-state PL measurement was controlled between 50 and 500 mW cm⁻².

TRPL measurements

The TRPL of samples was recorded by the TCSPC module (SPC-130) and a single-photon avalanche photodiode (SPAD). To measure the TRPL from MoSe₂, samples were excited at λ = 700 nm with 10-MHz train pulses (YSL, SC-OEM), and the NIR PL (750 to 850 nm) was collected by SPAD (MPD, PD-050-CTE-FC). The TRPL for UC emission was collected for 550 to 650 nm at 200 kHz. The excitation power was adjusted to obtain a ≤5% counting rate in each measurement to avoid pileup artifacts in the detector. The IRF of the TRPL setup was obtained by directly collecting the decay curve of laser pulse. The full width at half maximum of IRF is ~100 ps, which implies that any process faster than ~20 ps (~IRF/5) can be indistinguishable and show an identical curve to IRF. The power density for TCSPC was controlled to be 10 W/cm⁻² for measuring UC PL and ~100 W/cm⁻² for measuring MoSe₂ PL.

PL imaging

The photon UC imaging of the sample was performed on a home-built laser scanning microscope. The samples were placed in an upright microscope frame (Olympus, BX51WI), equipped with a 2D laser scanning galvanometer (Thorlabs, GVS002). The 2-ps, 785-nm laser with a repetition rate of 80 MHz (Applied Physics & Electronics, picoEmerald) was used with a power of ~300 μW at the sample. The microscope objective was a high-NA water-immersion lens (Olympus, 60×, NA = 1.2). The forward-going signal photons were collected by a high-NA oil-immersion condenser (Olympus, NA = 1.4). A short-pass dichroic mirror (Thorlabs, DMSP 650) and two optical band-pass filters (Thorlabs, FELH 550, FESH 750) were used to clean up the signal photons, which were then detected by a photomultiplier tube (Hamamatsu, H7422-40). The extracted signal was then sent to a data acquisition unit. The final images were assembled in a home-built LabVIEW program during laser scanning. The image pixel dwell time was 10 μs per pixel. The power density for PL imaging was controlled to be ~1000 W/cm⁻².

Supplementary Materials

This PDF file includes:

Section S1. Sample preparation

Section S2. Optical measurements

Section S3. Calculation on solar irradiation and excitation density

Section S4. Determining upconversion efficiency

Table S1

Figs. S1 to S11