Manipulatable Cascade Pumping Single-Photon Upconversion Authors: Qi Xiao, Wen Xu, Xiumei Yin, Na Zhou, Xinyao Dong, Ge Zhu, Xixian Luo, Yinglin Song, Bin Dong Published: 2026 ABSTRACT AbstractPhoton upconversion (UC), while promising for infrared photonics, is fundamentally constrained by limited spectral response range, low efficiency, and slow response speeds. Here, we present a machine learning-guided single-photon UC strategy based on cascade pumping that implements a “LEGO-inspired photon stacking” mechanism, in which intermediate state of lanthanide ions (Ln3+) becomes a “virtual ground state” for direct single-photon pumping to target energy levels. As a proof-of-concept, the NaYS2:Ho3+ UC emissions are selectively enhanced by 2-3 orders of magnitude via precise population control. This mechanism extends efficient UC response to ~2100 nm and reduces response time from 30 ms to 54 μs. The approach generalizes to other Ln3+ (Tm3+/Pr3+/Er3+), and energy-transfer optimization in Ho3+-sensitized systems yields near-pure RGB emission. We further demonstrate the high-sensitivity and rapid-response UC narrowband photodetection, enabling low-threshold CO2 sensing with a sensitivity of 6.4×10−4 ppm−1. Our work offers strategy for developing single-photon UC and infrared photodetection technologies.Subject terms: Optical materials and structures, Applied optics, Optical physics, Materials for optics Introduction Photon responsive devices, including solar cells, photodetectors (PDs), and phototransistors, serve as essential components in optoelectronic systems for applications such as solar energy harvesting, optical communications, and sensing1–4. The spectral responsive range of these devices determined by the band gap or absorption represents a crucial parameter that directly influences the spectral conversion capacity and efficiency of the devices. Photon upconversion (UC) offers a promising pathway to bypass these constraints by converting low-energy infrared photons into utilizable high-energy emissions5–13, thereby unlocking otherwise inaccessible spectral regions14–16. Lanthanide ions (Ln3+) doped luminescent materials stand as premier UC candidates due to their unique 4 fn electron configuration with abundant intra-4f infrared absorption transitions (e.g., Yb3+: 2F7/2-2F5/2 980 nm, Er3+: 4I15/2−4I13/2 1530 nm, Tm3+: 3H6-3F4 1870 nm, Ho3+: 5I8-5I7 2100 nm, etc.)17–19. These materials have been extensively employed to broaden the infrared response of optoelectronic devices20, with theoretical models projecting remarkable potential such as 47.6% maximum efficiency for UC-enhanced solar cells21. Despite seminal advances from the discovery of efficient β-NaYF4:Yb3+,Er3+ UC22 to early device demonstrations by Gibart et al.23, progress in enhancing photon-responsive devices through UC implementation has remained disappointingly limited. Practical Ln3+-doped UC remains constrained by three interlinked limitations. Firstly, severely restricted excitation bandwidths with only three effective wavelengths (980, 808, and 1530 nm) established in five decades24–26. The progress began with the discovery of 980 nm Yb3+, Er3+ UC in the 1970s22, followed by the development of 808 nm Nd3+-Yb3+-Er3+ UC in 201327, and most recently the emergence of 1530 nm Er3+ self-sensitized UC28. It stems from insufficient energy-level matching across conventional UC mechanisms (energy transfer, migration, self-sensitization). Second, UC suffers intrinsically low efficiency due to large multiphoton energy losses (>97%) during Ln3+-to-Ln3+ transfers, compounded by narrow 4 fn absorption cross-sections demanding high-power and wavelength-precise excitation29. Third, response speed is sluggish as UC emission relies on sequential energy transfers/migrations through multiple slow 4f-4f transitions, creating long-lived intermediate states (usually millisecond-scale). While plasmon-enhanced approaches show promise for accelerating transitions and transfer rates30,31, research remains confined to single nanoparticles. Consequently, developing UC systems with an extended excitation band, high efficiency, and rapid response is imperative for next-generation photon-responsive devices that require wide spectral operation, high sensitivity, and microsecond-speed functionality. Conventional UC relies on single wavelength excitation from ground state to target excited state of Ln3+, which requires perfectly matched energy levels with single pumping band for UC at each step, thereby inherently constraining both the tunability of responsive wavelengths and the overall efficiency of UC17. The ladder-like energy levels of Ln3+ enable a stepwise “photon LEGO” mechanism where electrons accumulate through sequential transitions. Theoretically, independent population manipulation permits direct pumping from any state (ground or excited) to target energy levels using precisely matched excitation energies32,33. This concept was previously validated in molecular systems through visible/mid-infrared co-excitation for achieving mid-infrared-to-visible UC and molecular vibration detection34,35. Implementing such a LEGO-inspired process, where stable intermediate states serve as “virtual ground states” for single-photon transitions, would enable arbitrary-wavelength access, near-lossless population control, and microsecond-scale kinetics, thereby overcoming the core limitations of UC. In this work, we develop single-photon Ln3+ UC systems using machine learning-optimized cascade pumping with dual-wavelength that implements a LEGO-like photon stacking mechanism. This approach enables direct single-photon pumping from intermediate “ground states” to target energy levels. As demonstrated in NaYS2:Ho3+ (or NaYS2:Tm3+/Pr3+/Er3+), strategically chosen excitation wavelengths enhance selective UC emissions by 2-3 orders of magnitude while extending effective response to ~2100 nm. The single-photon process simultaneously accelerates response times and realizes near-pure RGB emission through directed population engineering. Furthermore, we achieve high-sensitivity, rapid-response narrowband photodetection at ~2100 nm (1.83 A/W responsivity, 1.02 × 1011 Jones detectivity), enabling targeted CO2 gas sensing. Results As exhibited in Fig. 1a, we demonstrate a schematic representation of the LEGO brick stacking mechanism. LEGO stacking enables efficient transport of toy figures from ground level to an intermediate platform. Selective brick stacking bridges this gap, achieving migration to the target platform. This conceptual framework provides an analogy for photon UC in RE3+ (Fig. 1b), where platforms are energy levels of RE3+, toy figures present electrons, and LEGO blocks represent excitation photons. Owing to energy level mismatch, continuous single-wavelength pumping struggles to populate the target emitting state. This results in electrons accumulated at intermediate energy levels that cannot be further excited, thereby causing energy loss. However, the intermediate energy level populated by an auxiliary wavelength (λ1) can be regarded as a metastable “virtual ground state”. Subsequent cascade pumping by a photon (λ2) drives UC emission, which depends on the excitation photon density, absorption cross-sections and lifetimes of the energy levels (Supplementary Note 1), constituting a single-photon UC mechanism based on LEGO-like stacking. Selective emission enhancement and excitation wavelength extension based on single-photon UC by LEGO-inspired photon stacking. a The stacking diagram of LEGO bricks. b The single-photon UC systems under photo-assisted cascade pumping of Ln3+ based on the LEGO-inspired stacking effect. c Overview of the architecture for predicting the LEGO-inspired stacking UC under dual-wavelength cascade pumping through a machine learning algorithm. d Selected optimal parameters. Rare earth ion: Ho3+; Host material: ternary sulfide NaYS2; Target energy levels: 5F3, 5F4, and 5F5 for red, green, and blue emitting levels; Energy level gaps: 5I8 + 5064 cm–1 → 5I7 + 10392 cm–1 → 5F5, 5I8 + 5064 cm–1 → 5I7 + 13217 cm–1 → 5F4, 5I8 + 8578 cm–1 → 5I6 + 12198 cm–1 → 5F3. e–g The UC emission spectra of NaYS2:Ho3+ (1 mol%) under various single/dual-wavelength excitation, accompanied with a simple transition mechanism of single-photon infrared UC by LEGO-inspired photon stacking. And the excitation power is summarized as (e) 2096 nm: 250 mW and 980 nm: 100 mW; (f) 2096 nm: 250 mW and 760 nm: 100 mW; (g) 1208 nm: 250 mW and 808 nm: 200 mW. To validate our proposed concept, we employed a machine learning (ML) algorithm to predict the LEGO-like stacking UC behavior under cascade pumping of dual-wavelength (Fig. 1c, Supplementary Note 2). A comprehensive database is constructed by inputting multiple critical parameters into the algorithm36, including rare earth ions, energy level gaps, target excited states, and characteristics of different host materials (lifetimes and phonon energy). To achieve red, green, and blue (RGB) emissions from a single rare earth ion system, we incorporated the target emission wavelengths (red: ~650 nm, green: ~550 nm, blue: ~480 nm) as inputs. As demonstrated in Fig. 1d, the ML algorithm identified the optimal solution for attaining multiple emissions of red, green, and blue is to employ dual-wavelength cascade pumping to accomplish the population of Ho3+ target emitting energy levels in ternary sulfide NaYS2 materials (R: 5I8 + 5064 cm–1 → 5I7 + 10392 cm–1 → 5F5; G: 5I8 + 5064 cm–1 → 5I7 + 13217 cm–1 → 5F4; B: 5I8 + 8578 cm–1 → 5I6 + 12198 cm–1 → 5F3). First, NaYS2:Ho3+ UC phosphors synthesized through a gas-solid reaction route exhibit micrometer-sized crystals with 0.28 nm (104 plane) interplanar spacing, and high purity and crystallinity (Supplementary Figs. 1–3, Supplementary Note 3). Optical characterization reveals Ho3+ shows two strong NIR absorption bands centered at ~1200 nm (5I8 → 5I7) and ~2000 nm (5I8 → 5I6) (Supplementary Fig. 4). These intermediate excited states (5I7 and 5I6) can be further triggered as the new “ground state” to target upper energy levels via photon stacking. Figure 1e–g present the UC emission spectra of NaYS2:Ho3+ under both single-wavelength and dual-wavelength cascade pumping. Single-wavelength excitation (2096 nm, 1208 nm, 980 nm, 808 nm, or 760 nm) produces much weaker green (~555 nm, 5S2,5F4 → 5I8) and red (~658 nm, 5F5 → 5I8) emissions. This inefficiency arises from energy mismatches between the excitation wavelength and upper level gaps of Ho3+, hindering sequential electron transitions and reducing UC response (Supplementary Fig. 5). Strikingly, dual-wavelength cascade pumping (2096 nm & 980 nm) enhances the red UC emission (5F5 → 5I8) by ~4220-fold compared to the sum of the two single excitation intensities (Fig. 1e, Supplementary Fig. 6). This selective enhancement persists across various Ho3+ doping concentrations (Supplementary Fig. 7). The UC mechanism involves the initial population of electrons in the 5I7 via 2096 nm absorption (5I8 → 5I7), followed by their subsequent promotion to the 5F5 via 980 nm photon re-absorption (5I7 → 5F5). This pathway achieves efficient single-band red emission, increasing the red/green emission intensity ratio (IR/IG) from 2.2/16.9 (single-wavelength excitation) to 606.3 (dual-wavelength excitation) (Supplementary Fig. 8). This direct electron pumping to target levels reduces energy losses from multi-step excitation, evidenced by a 66% decrease in the 5I6 → 5I8 transition (Supplementary Fig. 9). We further confirm this significant red UC enhancement in classical Ho3+ hosts, including fluoride (NaYF4) and oxides (Y2O3 and La2Mo2O9) (Supplementary Fig. 10-12 and Supplementary Note 3). The low phonon energy, unique layered structure, and long excited-state lifetimes render NaYS2 a candidate host for efficient cascade pumping single-photon UC37. Similarly, selective single-band UC green emission (5F4/5S2 → 5I8) in Ho3+ is achieved by using cascade pumping at 2096 nm and 760 nm (5I8 → 5I7 → 5F4/5S2), resulting in a 219-fold enhancement in luminescence intensity (Fig. 1f). Concurrently, an 8-fold enhancement in red emission can be observed, attributed to non-radiative relaxation from 5S2/5F4 to 5F5. This strategy effectively shifts the dominant emission from red (under single-wavelength excitation) to overwhelmingly green, reflected in a significantly reduced IR/IG ratio of 0.1 (Supplementary Fig. 13). Remarkably, dual-wavelength cascade pumping at 1208 nm and 808 nm (5I8 → 5I6 → 5F3) produces an intense, rarely observed blue UC emission (5F3 → 5I8) in Ho3+, yielding an extraordinary ~1023 fold selective enhancement in blue emission. Minor enhancements in green and red emissions also occur due to non-radiative relaxation (5F3 → 5F4 → 5F5) (Fig. 1g). The resulting spectrum demonstrates dominant blue UC emission with residual green and red components, likely due to the small energy gaps between the 5F3, 5F4, and 5F5 levels. Conversely, the large energy gaps leads to the reduced enhancement (Supplementary Fig. 14). These findings collectively demonstrate that efficient selective UC emissions from specific energy levels in Ho3+-doped systems can be effectively realized through the “photon LEGO stacking” effect enabled by cascade pumping. Precise color tuning of UC can be achieved by utilizing the LEGO-inspired stacking effect of different wavelength combinations (Fig. 2a). Under cascade pumping at 2096 nm and 980 nm, the sample exhibits single-band red emission with IR/IG of 606.3. Introducing an additional 760 nm source activates green emission, whose intensity gradually increases with rising excitation power, accompanied by a decrease in the IR/IG from 606.3 to 0.3. In this scheme, the 2096 nm laser populates the 5I7 level as a “virtual ground state”, while the competitive single-photon absorption between the 760 nm and 980 nm excitation enable switchable UC luminescence colors (Fig. 2b). Power-dependent UC emissions using Ho3+ red emission as a representative case were investigated, with independently tunable power at 2096 nm and 980 nm (Fig. 2c and Supplementary Fig. 15). When increasing 980 nm power from 92 mW to 561 mW at constant 2096 nm power (250 mW), red UC intensity enhances from 6670-fold to 34700-fold compared to single-wavelength 2096 nm excitation (Fig. 2c). This trend is consistently observed when varying 2096 nm power at fixed 980 nm power (Supplementary Fig. 16). The significant UC enhancement through power modulation should be assigned to the direct pumping of electrons from 5I7 to the target red emission energy level or increased 5I7 population through improved excitation efficiency (Fig. 1e). The absolutely dominance of red UC emissions demonstrates the absence of competing energy transfer or cross-relaxation processes typically occurred in Ln3+ systems, highlighting uniqueness of this strategy. Photon-number analysis via log-log plots (Iup vs. P) shows that red emission follows a two-photon process under 980 nm, a three-photon process under 2096 nm, but crucially, a one-photon process under dual-wavelength excitation (Fig. 2d and Supplementary Fig. 17). This confirms each UC transition effectively becomes a single-photon process at cascade pumping, implying the enormous potential for highly efficient UC. Additionally, any intermediate state in Ln3+ can function as a new “virtual ground state”, enabling effective excitation at wavelengths up to 2100 nm by targeting gaps to upper energy levels. Theoretically, a virtual ground state requires a long excited-state lifetime, and Ln3+ naturally satisfied by the intrinsically long lifetimes (typically on the order of microseconds to milliseconds) of 4f-4f transitions (Supplementary Fig. 18 and Supplementary Note 1). Mechanism investigation of single-photon UC by LEGO-inspired photon stacking mode. a The UC emission spectra of NaYS2:Ho3+ obtained by modulation the competition between 980 nm and 760 nm under synergistic excitation at 2096 nm. b The transition mechanism of UC emission regulation for Ho3+. c Enhancement factors of typical red UC emission intensities from Ho3+ under dual-wavelength of (x)980 & 2096 nm excitation compared to under single-wavelength of 2096 nm excitation. d Double logarithmic dependence of red emission intensity on the laser powers of single-wavelength of 980 nm excitation or dual-wavelength cascade pumping with the fixed power of 2096 nm excitation. Note: The number of photons (n) can be calculated as follows: Iup∝Pn. e Time-dependent red (658 nm) emission profiles of NaYS2:Ho3+ under pulsed dual-wavelength excitation with a duration time (pulse width) of 30 ms. 2096 nm laser as C.W. beam; 980 nm laser as P.W. beam. The rise time of the UC process is the required time for the luminescence intensity ranges from 10% to 90%. f Ultrafast transient absorption kinetic curves of Ho3+ with a probe wavelength of 658 nm. g Mechanism model of energy levels for modulation of red UC luminescence upon dual-wavelength cascade pumping mode. Tailoring the 2096 nm NIR laser as continuous wave (C.W.) and synergistically combing with pulsed wave (P.W.) of 980 nm as input. h–j The emission spectra and cooperative luminescence mechanism of typical single-doped systems ((e) 808 nm: 268 mW and 1208 nm: 230 mW; (f) 808 nm: 268 mW and 980 nm: 361 mW; (g) 980 nm: 30 mW and 2096 nm: 20 mW) under single/dual-wavelength excitation. Time-resolved dynamic curves of Ho3+ red emission (5F5 → 5I8) require more than 30 ms to reach a steady state under pulsed 980 nm excitation alone (Fig. 2e). Remarkably, introducing 2096 nm continuous wave (C.W.) excitation dramatically reduces the rise time to ~54 μs with a 3-order-of-magnitude acceleration, particularly crucial for developing UC photon-responsive devices. Ground-state bleaching recovery probed at 658 nm further elucidates the mechanism (Fig. 2f). Under 980 nm excitation alone, electron transitions from ground state (5I8) to the red-emitting level (5F5) remain negligible. In contrast, upon introducing 2096 nm co-excitation, a pronounced decrease in probe absorption at 658 nm is observed, indicating efficient depletion of “virtual ground state” in 5I7 and subsequent population of the red-emitting level via single-photon absorption. The shortened ground-state bleaching lifetime (7.2 μs) compared to the luminescence decay time (1098 μs) demonstrates that a portion of electrons at intermediate state (5I5) from the 980 nm single-excitation return rapidly to “virtual ground state” via non-radiative relaxation. This also confirms that the excited electrons at the red-emitting level do not undergo additional energy transfer or cross-relaxation processes that could delay ground-state recovery. The relatively long ground-state bleaching lifetime of 5F4 green-emitting level (135 μs) can be attributed to increased non-radiative relaxation under 760 nm single-wavelength excitation (Supplementary Fig. 19). As illustrated in Fig. 2g, 2096 nm photons excite Ho3+ from the ground state (5I8) to 5I7, accumulating electrons in this intermediate state. Continuous 2096 nm excitation stabilizes the 5I7 population, creating a stationary “virtual ground state”. Subsequent 980 nm absorption then directly promotes electrons from 5I7 to 5F5, triggering intense red emission. This optimized pathway bypasses energy-loss processes (e.g., energy transfer, cross-relaxation), enabling ultrafast response. This phenomenon of shortened UC rise time also occurs when a 2096 nm P.W. light source and a 980 nm C.W. light source are used (Supplementary Fig. 20). It is worth noting that variations in the excitation frequency do not alter the cascade pumping luminescence (Supplementary Fig. 21). Then, we apply this “photon LEGO” concept to other NaYS2:Ln3+ (e.g., Tm3+, Pr3+, and Er3+) systems under dual-wavelength cascade pumping (Fig. 2h–j). Similarly, compared to single-wavelength excitation, dual-wavelength pumping realizes initial electron accumulation in Tm3+: 3H4, Pr3+: 1G4, and Er3+: 4I11/2 using 808 nm or 980 nm excitation, establishing stable intermediate states. Simultaneously, complementary wavelengths excitation (1208 nm for Tm3+: 3H6 → 1G4, 808 nm for Pr3+: 1G4 → 3P0,1, and 2096 nm for Er3+: 4I11/2 → 4F9/2) precisely target energy gaps. This approach significantly enhances selective UC emission, achieving several tens to hundreds of times improvement for Tm3+: 1G4 → 3H6, Pr3+: 3P0,1 → 3H4, and Er3+: 4F9/2 → 4I15/2. The single-photon UC mechanism enabled by our “photon LEGO” strategy provides unprecedented selective emission control at specific energy levels, allowing for smart UC manipulation. To achieve pure single-band UC emission while suppressing unwanted transitions, we integrate conventional energy transfer mechanisms with this LEGO stacking approach to directionally control population pathways in Ln3+. Using the NaYS2:Ho3+ system under 1208/808 nm dual-wavelength excitation as an example, in which UC blue emission significantly enhances, green/red emissions persist (Fig. 1g). Yb3+ and Tm3+ are selected as energy transfer mediators codoped with Ho3+ in NaYS2 UC systems to accomplish single band green and blue UC emissions. In NaYS2:Ho3+, Yb3+ (Fig. 3a), Ho3+ independently absorbs 808 nm and 1208 nm photons, promoting electrons to the 5I5 and 5I6 energy levels. Subsequent energy transfer from Ho3+ (5I5) to Yb3+ (2F5/2) enables cross-relaxation between populated Ho3+ (5I6) and Yb3+ (2F5/2) states (2F5/2 + 5I6 → 2F7/2 + 5F4,5S2), selectively enhancing green emission (5S2,5F4 → 5I8), as shown in Fig. 3c. Similarly, in NaYS2:Ho3+, Tm3+ (Fig. 3b, d), energy transfers from Ho3+ (5I6) to Tm3+ (3H5) after initial excitation, followed by 808 nm photon absorption promoting Tm3+ to 1G4 state, yielding ~1117-fold enhanced blue emission (1G4 → 3H6). Selectively tuning of photon UC via dual-wavelength energy transfer. Schematic of the proposed a Ho3+, Yb3+, and b Ho3+, Tm3+ codoped systems for investigating the directional control of electron population pathway based on the energy transfer under dual-wavelength excitation at 1208 nm and 808 nm. The emission spectra of c Ho3+ (5 mol%), Yb3+ (10 mol%), and d Ho3+ (5 mol%), Tm3+ (0.5 mol%) codoped NaYS2 under single/dual-wavelength excitation. e The normalized decay curves of 5I6 (Ho3+) energy levels with various components of Tm3+ in NaYS2:Ho3+,Tm3+. f Photographs of luminescence enhancement for Ho3+, Ho3+-Yb3+ and Ho3+-Tm3+ UC systems under single/dual-wavelength excitation (Ho3+: λ1: 980 nm, λ2: 2096 nm; Ho3+-Yb3+ and Ho3+-Tm3+: λ1: 808 nm, λ2: 1208 nm). Time-dependent corresponding emission profiles of g NaYS2:Ho3+, Yb3+ h NaYS2:Ho3+, Tm3+ under pulsed dual-wavelength excitation with a duration time (pulse width) of 60 ms. Under 808/1208 nm dual-wavelength excitation, NaYS2:Ho3+, Yb3+ and NaYS2:Ho3+, Tm3+ systems exhibit distinct single-band green and blue UC emissions, respectively, contrasting with the multiband emissions observed under single-wavelength excitation (Fig. 3c, d). The energy transfer from Ho3+ to Yb3+/Tm3+ plays a critical role in the modulation of UC emission, facilitated by minimal energy gaps between Ho3+: 5I5 and Yb3+: 2F5/2 in the Yb3+ codoped system (Ho3+: 5I6 and Tm3+: 3H5 in the Tm3+ codoped system). This enables efficient energy transfer that produces single-band emissions, confirmed by the concentration-dependent lifetime shortening of 5I6 → 5I8 transition for Ho3+ (78.6% transfer efficiency to Tm3+, Supplementary Note 4; Fig. 3e). Combined with red emission in NaYS2:Ho3+, this strategy achieves all three primary RGB UC colors (Ho3+: R/G = 606.3, R/B = 567.2; Ho3+-Yb3+: G/R = 3.4, G/B = 14.3; Ho3+-Tm3+: B/R = 18.1, B/G = 7.8) in Ho3+-based systems through tailored dual-wavelength excitation and energy transfer pathways (Fig. 3f). This function permits precise regulate over wide-gamut UC luminescence, with the resulting color gamut covering a region equivalent to 116.6% of the standard sRGB gamut (Supplementary Fig. 22). Time-resolved studies of green/blue emissions under 808 nm pulsed and 808/1208 nm dual-wavelength excitation (Fig. 3g, h) reveal significantly accelerated rise/decay kinetics under dual-wavelength conditions, validating the simplified UC population processes that enable efficient specific level emission (Supplementary Fig. 23). Accurate gas detection is critical for industrial safety. Laser absorption spectroscopy (LAS) technology offers high sensitivity and resolution for gas sensing38,39, with PDs acting as key components for specific wavelength reception. However, since most gases exhibit absorption bands in the infrared region beyond 2000 nm, conventional detectors for such wavelengths are often expensive and complex. UC-based PDs present a promising low-cost, high-efficiency alternative for LAS gas sensing. We propose a LAS gas sensing scheme using single-photon UC narrowband NIR detection, as illustrated in Fig. 4a. The setup employs two beams: a detection beam (λ1) passes through a gas cell containing the target analyte. According to the Beer-Lambert law, gas molecules absorb λ1 photons, attenuating the beam. The attenuated λ1 beam then combines with the reference beam (λ2) to jointly excite the UC layer of the PDs (Inset of Fig. 4a). The PDs then convert the gas-modulated UC signal into a measurable photocurrent variation, enabling precise concentration quantification. Figure 4b shows the fabricated narrowband NIR PDs (Supplementary Fig. 24). Under dual-wavelength cascade pumping, NIR photons are converted into visible light via UC (e.g., NaYS2:Ho3+), which is subsequently absorbed by perovskite MAPbI3 PDs to generate photocurrent, thereby enabling infrared photodetection. The PDs feature a vertical PEDOT-MAPbI3-PCBM structure for efficient carrier transport. A spin-coated NaYS2:Ho3+ film effectively absorbs both 2096 nm (detection light) and 980 nm (reference light), emitting visible photons. The absorption spectrum of the photoactive layer (MAPbI3, Supplementary Fig. 25) in PDs entirely covers and effectively absorbs the visible UC emission of NaYS2:Ho3+, or Ho3+/Yb3+(Tm3+) films, while remaining distinct from the excitation wavelengths of both the reference and detection lights. Figure 4c displays the time-dependent on/off photocurrent curves of NaYS2:Ho3+ based PD at 0 V bias. Under separate 980 nm or 2096 nm illumination (20 mW/cm2), negligible photocurrent is observed. However, simultaneous co-excitation generates 0.07 μA photocurrent. This enhancement stems from efficient UC under dual-wavelength excitation. Narrowband NIR light photodetection utilizing dual-wavelength cascade pumping towards gas sensing. a LAS gas sensing based on dual-wavelength cascade pumping, narrowband NIR optical detection. b Schematic illustration of photodetector device configuration under the coupling of detection light and reference light. e–: electron, h+: hole. c On/off time-dependent photocurrent curves of the photodetector under independent excitation of 2096 nm or 980 nm and cooperative excitation of 2096 nm and 980 nm. d Power density dependence of detectivity for 2096 nm NIR light detection with or without 980 nm light assistance. e The detectivity comparison of NaYS2:Ho3+-based PDs with other infrared detectors, including organics, quantum dots, UC materials, and 2D materials. f Response time of the NaYS2:Ho3+-based PDs for 2096 nm NIR light detection under single/dual-wavelength excitation. g Absorption of NaYS2:Ho3+ and CO2 gas ranging from 1800 nm to 2200 nm. h On/off photocurrent curves of fabricated PDs. i Normalized difference photocurrent ∆P/P0 of UC-based PDs and commercial PDs. The fitted linear slope is the sensitivity of CO2 gas detection. Three key metrics of PDs, photoresponsivity (R), detectivity (D*), and external quantum efficiency (EQE) were evaluated (Fig. 4d and Supplementary Fig. 26). Under sole 2096 nm, the values of R, D* and EQE are recorded to be 1.19 ×10–3 A/W, 5.94 × 107 Jones, and 0.07% with irradiation power density of 180 mW/cm2. Strikingly, with 980 nm assistance (20 mW/cm2), the values of R, D* and EQE for 2096 nm NIR detection -reach 1.83 A/W, 1.02 × 1011 Jones and 108.54%, achieving significant enhancements of ~1534 fold, ~1721 fold and ~1533 fold, respectively. The detection threshold dropped below 0.5 mW/cm2 with auxiliary 980 nm excitation, three orders of magnitude lower than that of single-wavelength excitation (Supplementary Fig. 27). Similar 1208 nm detection using Ho3+-Tm3+/Yb3+ with 808 nm reference light is shown in Supplementary Fig. 28, 29. Notably, prior Ln3+-based UC PDs are limited to short-wavelength NIR detection (≤1500 nm) due to UC system constraints, while this work is the demonstration of Ln3+ UC PDs achieving 2000 nm photodetection. The detectivity performance of NaYS2:Ho3+-based PDs is not only comparable to short-wavelength NIR detection based on other materials, including organics40,41, quantum dots42,43, UC materials28,44–48, and 2D materials49,50, but also significantly surpasses that of 2D material systems capable of detecting wavelengths >2000 nm51–56 (Fig. 4e). Response time, comprising rise time and decay time, is another critical parameter for evaluating the performance of PDs. Theoretically, the response time of PDs is primarily determined by the UC system due to their inherent population decay dynamics. As shown in Fig. 4f, NaYS2:Ho3+-based PDs exhibit 81.66 ms rise time and 10.54 ms decay time under sole 980 nm excitation. Remarkably, dual-wavelength cascade pumping (2096 nm + 980 nm) reduces these to microsecond-scale values: 595 μs rise time and 4.28 ms decay time. This acceleration stems from rapid single-photon UC processes, demonstrating dual-wavelength excitation fundamentally modifies UC kinetics to optimize PD response times. Furthermore, NaYS2:Ho3+-based PDs exhibit good stability and reproducibility (Supplementary Fig. 30). Finally, we apply this technology to CO2 sensing, a critical safety concern in industrial settings where leaks pose significant risks. Crucially, CO2 exhibits a strong ~2000 nm absorption band that overlaps with the NaYS2:Ho3+ absorption band (Fig. 4g), enabling dual-wavelength detection. In our demonstration, a 2004 nm C.W. laser (1 mW/cm2) served as detection light passing through a 76-m multipass cell containing CO2/N2 mixtures (N2 exhibits almost no absorption at 2004 nm; Supplementary Fig. 31) at varying concentrations, while a 980 nm P.W. laser (20 mW/cm2) acted as reference light. As expected, attenuation of the 2004 nm beam reduced Ho3+ red UC luminescence (Supplementary Fig. 32). Photocurrent decreases from ~0.1 μA to ~0.07 μA as CO2 concentration increasing from 0 to 5000 ppm (Fig. 4h), with effective detection below 500 ppm meeting industrial and mining requirements (≤5000 ppm). The values of ∆P/P0 for both UC-based PDs and commercial PDs exhibit excellent linear fitting, with R2 values exceeding 0.99. Sensitivity calculations from normalized ∆P/P0 show UC-PDs achieved 6.4 × 10–4 ppm–1, outperforming 5.9 × 10–4 ppm–1 of commercial 2004 nm PDs (Fig. 4i). These results demonstrate that UC-PDs serve as simpler, high-performance alternatives to conventional complex infrared detectors for CO2 monitoring. In conclusion, we propose Ln3+ single-photon LEGO-like stacking UC systems utilizing a machine learning algorithm under photon-assisted cascade pumping. The selection of appropriate excitation wavelengths based on the energy level gaps of Ln3+ enables directed population of excited states, thereby precisely controlling UC emission in NaYS2. The UC emissions of Ln3+ (Ho3+, Tm3+, Pr3+, Er3+) achieve a remarkable enhancement of two to three orders of magnitude under dual-wavelength cascade pumping. The UC effective response wavelength is extended to 2096 nm through photon-assisted cascade pumping. Simultaneously, this single-photon UC mechanism significantly simplifies the electron population process of excited states and drastically reduces the UC response time. Additionally, a strategic combination of energy transfer and LEGO photon-stacking effects is employed to facilitate internal electron-directed population and the selective enhancement of emission in Ho3+-sensitized UC systems. Consequently, we develop the narrowband NIR photodetection system for 2096 nm, assisted by 980 nm light, which demonstrates a relatively high responsivity of 1.83 A/W and a detectivity of 1.02×1011 Jones, respectively. Notably, this approach significantly lowers the detection threshold of PDs and enhances the response speed of NIR detection. Based on the LAS technology, the fabricated PDs can be targeted for CO2 gas sensing. These findings not only establish the single-photon LEGO-like stacking UC population mechanism, but also provide guidance for constructing NIR detection with high sensitivity and rapid response. Methods Materials Lanthanide oxides (Y2O3, Er2O3, Tm2O3, Ho2O3, Pr2O3, Yb2O3; 99.99%), and sodium carbonate (Na2CO3; A.R.) were used as the primary materials. Carbon disulfide (CS2, from Aladdin, ≥99%) was employed as the sulfur source. MAI (CH6IN, 99.99%), PbI2 (99.99%), DMSO (A.R.), C6H5Cl (A.R.), DMF (C3H7NO, 99.9%), the poly (3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS, abbreviated herein as PEDOT), [6,6]-Phenyl C 61 methyl butyrate (PCBM) were used for fabrication of the photodetectors. Synthesis of single-doped (Ho3+, Er3+, Tm3+, Pr3+) and co-doped (Ho3+-Yb3+, Ho3+-Tm3+) NaYS2 phosphors The single-doped (Ho3+, Er3+, Tm3+, Pr3+) and co-doped (Ho3+-Yb3+, Ho3+-Tm3+) NaYS2 phosphors were synthesized using a gas-solid reaction method. First, Ln2O3 and Na2CO3 were weighed according to the stoichiometric ratio, then the reactants were carefully mixed in an agate mortar for 30 min. The mixture was transferred to a quartz boat and placed in a tube furnace, and further heated under argon (Ar) atmosphere to 500 °C for 50 min. Subsequently, saturated vapor of CS2 (40 kPa) was pumped into the tube furnace at a flow rate of 50 mL/min in an Ar atmosphere. Finally, the samples were obtained after calcining the mixture at 800 °C for two hours and washing with water and ethanol several times. Device fabrication The etched fluorine-doped tin oxide (FTO) glass substrate (2 cm × 2 cm) was first treated with plasma cleaning for 10 min. A PEDOT:PSS layer was then spin-coated onto the substrate surface at 4000 rpm for 30 s, followed by annealing in air at 150 °C for 10 min to form a uniform film. Subsequently, the substrate was transferred into a nitrogen-filled glove box for further processing. A perovskite precursor solution (MAPbI3) was spin-coated onto the PEDOT:PSS layer at 600 rpm, and 400 µL of chlorobenzene was added as an antisolvent during spinning at 4000 rpm to form the MAPbI3 layer. The spin-coated film was then annealed at 100 °C for 10 min. After cooling to room temperature, a PCBM solution (prepared in 1 mL of chlorobenzene) was spin-coated onto the MAPbI3 layer at 1000 rpm for 30 s, and the resulting PCBM film was annealed at 100 °C for 15 min to remove any residual solvent. Finally, the substrates were placed in an evaporation system, where a 90 nm thick Ag contact electrode was deposited on top of the PCBM layer to complete the device. Lanthanide UC films were prepared by a simple spin-coating method. Specifically, the UC phosphor was ground, dispersed in ethanol via ultrasonication, and spin-coated onto FTO at 1000 rpm for 10 s to form a uniform film. After multiple coating cycles, the film was dried at 80–100 °C to remove the solvent, yielding a dense layer ready for subsequent device integration. Structure and morphology characterization X-ray diffraction (XRD) was performed by using the Shimadzu-6000 X-ray diffractometer with Cu Kα radiation (λ = 0.15406 nm, 40 kV, 40 mA, 2θ = 10–70°). Scanning electron microscopic (SEM) images were acquired with a Hitachi S-4800 scanning electron microscope. Transmission electron microscopic (TEM) and high-resolution transmission electron microscopic (HRTEM) images of samples were measured by using a JEM-2100F transmission electron microscope. Optical properties measurement UC luminescence spectra and decay curves were measured by a Jobin Yvon iHR550 monochromator equipped with R928 and H10220B-75 photomultiplier tubes from Hamamatsu Photonics and a Tektronix DPO 5104 digital oscilloscope using 808 nm, 980 nm, 1208 nm, 1532 nm and 2096 nm laser diodes (maximum power of 1.0 W for the 1532 nm laser, maximum power of 2.0 W for 808 nm, 980 nm, and 2096 nm laser, maximum power of 3.0 W for the 1208 nm laser), and a pulsed work Horizon OPO laser, for signal collection from 400 nm to 1700 nm. In addition, a tunable Nd:YAG laser pumped OPO laser was used as the excitation source (e.g., 760 nm), with a pulse duration of 20 ns, repetition frequency of 5 Hz and a line width of 4–7 cm–1. The UC luminescence images were captured using a Canon EOS 5D Mark III digital camera (Tv=1:1250, Lens: EF 24–70 mm f/2.8 L II USM, Av=2.8, ISO = 12800). The 2004 nm detector chip, supplied by INP Photonics (Suzhou) Co., Ltd., is a photodiode designed for carbon dioxide sensing applications, model number INP-PCS-2004N-A09-01. Supplementary information Supplementary Information Transparent Peer Review file Source data Source data Author contributions B.D., W.X. and Y.S. conceived and supervised the project. Q.X. designed the experiment. X.Y. and N.Z. synthesized the samples. Q.X., X.D. and G.Z. performed the characterization and optical measurements. W.X., Q.X. and X.L. analyzed and interpreted the data. Q.X., W.X. and B.D. wrote the manuscript, with input from all authors. Peer review Peer review information Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available. Funding This work is supported by National Key Research and Development Program of China (2024YFA1409900), National Natural Science Foundation of China (U24A202439, U2441222, U24A2018, 62575408, 12574433), the Science and Technique Foundation of Dalian (2022JJ11CG003), Liaoning Xingliao Talent Plan (XLYC2203170), the Science and Technique Foundation of Liaoning Province (2023JH1/10400057, 2023JH1/10400080, 2024JH3/50100028, 2024JH2/102400021). Data availability The data that support the findings of this study are available within the paper, supplementary information and source data file. All data are available from the corresponding author upon request. Source data are provided with this paper. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Wen Xu, Email: xuwen@dlnu.edu.cn. Yinglin Song, Email: ylsong@hit.edu.cn. Bin Dong, Email: dong@dlnu.edu.cn. Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-75688-0. References Associated Data Data Availability Statement The data that support the findings of this study are available within the paper, supplementary information and source data file. All data are available from the corresponding author upon request. Source data are provided with this paper.