Main

Photochemical upconversion (PUC)1 can enable photon-driven processes that would otherwise require photons of much higher energies2. Photovoltaics3, photocatalysis4,5 and 3D-nanoprinting6 are three such examples. With triplet-sensitizing dyes and triplet fusion (TF) chromophores in solution, impressive upconversion efficiencies have been demonstrated which approach the fundamental limit of 50% quantum yield7,8,9. Rational achievement of high efficiencies in the solid state has, however, been largely elusive10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33.

To absorb as much light as possible in a thin layer, a high concentration of sensitizers is required. The higher the sensitizer concentration, the higher the triplet concentration under the same illumination and the more efficient is TF. However, high concentrations of dyes or nanocrystals can lead to aggregation and subsequent quenching that reduces efficiency34,35.

Recently, we reported that a nanostructured alumina support could be used to immobilize sensitizer molecules, mitigating their mutual interaction while maintaining strong light absorption and sensitizer–annihilator proximity33. Injection of triplet states into the interstitial solution, containing emitter molecules, resulted in TF upconversion at high efficiencies. For diphenylanthracene (DPA) emitter, which is intrinsically limited to about 25% upconversion quantum yield on account of a mediocre spin-statistical factor (only about half of the TF events generate an excited singlet), we achieved an internal upconversion quantum yield of 9.4%, which is 40% of the maximum for diphenylanthracene.

Solutions have the advantage that the upconverted singlet state has the same mobility as the sensitized triplet states: they simply diffuse with the molecule36. However, in the absence of solvent, in a neat chromophore, triplet states may diffuse by short-range hopping (Dexter mechanism37), whereas excited singlet states may rapidly migrate by the longer-range Förster mechanism38. Although this behaviour is the essence of organic photovoltaics39, where singlet states can migrate 100 nm to a heterojunction to undergo charge separation, in PUC it is catastrophic. If the upconverted singlet state can migrate to within the Förster radius of a sensitizer, it will be lost as the sensitizer efficiently converts the excited singlet state back into a lower-energy triplet state. As such, the efficiency will be impaired. Rational design of efficient, nanoscale PUC devices therefore requires a mechanism to localize the upconverted singlet state, or else separate it from the sensitizer16,40. One approach is to dope the TF medium with species that exhibit higher triplet energies, but lower singlet energies, than the bulk. This is the approach taken to mitigate singlet fission in rubrene, which efficiently competes with photon emission22,31,41,42. However, at the nanoscale such high concentrations are required that phase separation between the TF medium and the dopant is difficult to control.

In this Article, we demonstrate that the structure of the TF medium can influence exciton mobility by trapping the upconverted state at excimer-forming sites. We invoke a neat chromophore, 9,10-bis(n-octyl-diisopropylsilylethynyl)anthracene (NODIPS-An, Fig. 1a) as the TF medium that fills the pores of nanostructured alumina to which bespoke sensitizer molecules are adsorbed. The upconverted light is excimeric in character, emanating from localized states. We demonstrate a photon generation upconversion quantum yield of 8.2(4)%. Taking into account the fluorescence quantum yield of the neat chromophore, this corresponds to an excited-state upconversion yield of 19.0(8)%. We attribute the majority of efficiency losses to the sensitization step, and propose a pathway to the efficiency limit in solid-state, nanostructured PUC devices to overcome these limitations.

Fig. 1: Photophysics of NODIPS-anthracene and PdPQ4NACA.
Fig. 1: Photophysics of NODIPS-anthracene and PdPQ4NACA.The alternative text for this image may have been generated using AI.
Full size image

a, Structure of NODIPS-anthracene. b, Steady-state absorption (dashed) and PL (solid) of neat (green) and dilute (blue) NODIPS-anthracene. Inset top left: photograph showing blue emission in the dilute solution. Inset bottom right: photograph showing the greenish emission in neat NODIPS-An. c, TCSPC of neat (green) and dilute (blue) NODIPS-anthracene. Solid lines are monoexponential fits of the data with the lifetimes reported in the figure. d, Structure of PdPQ4NACA. e, Steady-state absorption of PdPQ4NACA. f, Process of PUC for PdPQ4NACA and NODIPS-anthracene. The process consists of six photophysical steps: (1) photon absorption by the PdPQ4NACA sensitizer; (2) ISC; (3) TET; (4) TTA; TF; (5) FRET; and (6) photon emission. Ex1, excimer.

Source data

Results

Molecular dynamics simulations

Molecular dynamics (MD) simulations were performed on NODIPS-An, with the details reported in the Methods and Supplementary Section 1. Radial distribution functions (RDFs) and angular-radial distribution functions (ARDFs) at several temperatures down to 298 K are reported in Supplementary Figs. 1–3. The structural decorrelation time for neat NODIPS-An at 298 K is around 10 µs (Supplementary Fig. 4). There is strong ordering of the π-stacked nearest neighbours indicated by the sharp first peak in the RDF (Supplementary Fig. 1). Despite the notable peak broadening in the RDF and ARDFs with increasing temperature, the lack of change in the peak positions indicates that the local structure of NODIPS-An is largely preserved as the temperature increases, albeit with increasing disorder. The ARDF shows that the nearest neighbours are separated by 3.8–4.0 Å and consist of π-stacked pairs in which the angle between the backbones varies between ~90° and 55° (Supplementary Fig. 2). Neat NODIPS-An is essentially solid on the timescales of the experiments presented below. On the longest timescales measured spectroscopically, ~50 µs, the structure decorrelates, but molecular diffusion is minimal. As such, triplet exciton transport in NODIPS-An must occur by hopping. Similarly, excimer formation, which occurs on a picosecond timescale, must occur by Förster resonance energy transfer (FRET) to a pair of chromophores predisposed to excimer formation, which are likely to be H-aggregated in the ground state.

Spectroscopy of NODIPS-An and PdPQ4NACA

The synthesis of NODIPS-An is reported in Supplementary Section 7.2. Figure 1a shows the structure of NODIPS-An and the absorption and emission spectra of both neat and dilute NODIPS-An are shown in Fig. 1b. The dilute spectra are nearly identical to those of TIPS-anthracene, a room-temperature solid, showing that the addition of NODIPS groups has a similar effect to that of the electronically similar TIPS groups on the anthracene chromophore43. As a neat room-temperature chromophore, the absorption spectrum exhibits a long tail, demonstrating excitonic coupling that is evident in the excimeric emission spectrum.

The S1 state energy is estimated to be 2.82 eV, from the 440-nm intercept of absorption and emission spectra in Fig. 1b. Our work on TIPS-anthracene suggests that the T1 state is about 1.4 eV (ref. 44). The magnitude of the Stokes shift of the excimer emission suggests that its energy is about 2.58 eV, assuming similar curvatures on the S0 and excimer potential energy surfaces.

Figure 1c shows the time-resolved emission of the dilute and neat chromophore. The MD results and the absorption spectrum indicate that there is a range of excitonic environments in the material. However, the time-resolved emission tracked at the peak of the excimeric emission (532 nm) is single-exponential on the timescale and resolution of the time-correlated single photon counting (TCSPC) (instrument response function ~500 ps). The observed lifetime is 23.6 ns, which is longer than the observed 5.2 ns singlet state lifetime of the monomer, indicating a reduction in the radiative rate. Formation of an excimer state must therefore occur on a sub-500-ps timescale in the neat chromophore.

The fluorescence quantum yield of neat NODIPS-An was determined to be 43%, using our previously reported instrument33. With an observed lifetime of 23.6 ns, this implies an intrinsic radiative lifetime, τr, of 55 ns, which is several times what is determined in the dilute solution, 5.2 ns (τr = 6.7 ns, quantum yield 78%). The increase in radiative lifetime is associated with a cancellation of the transition moment, as in H-aggregates45. This difference cannot be accounted for by the emission wavelengths alone.

The structure of the sensitizer, palladium(II) amino-nitro-tetrakisquinoxalino-porphinecarboxylic acid (PdPQ4NACA), and its absorption spectrum are shown in Fig. 1d,e, respectively. Its spectrum is similar to the previously reported PdPQ4NA, which has a broader Q-band than PdPQ4 (ref. 46). As explained in Supplementary Section 7.3, the PdPQ4NACA is a mixture of inseparable isomers, relating to the relative positions of the carboxyl and nitro/amino groups. PdPQ4NACA, like the structurally related PdPQ4NA, exhibits minimal fluorescence. Strong phosphorescence is observed in the 850-nm region, indicating a significant intersystem crossing (ISC) yield.

A mixture of these chromophores is capable of undergoing PUC. In this system, PUC consists of six photophysical steps, shown in the simplified Jablonski diagram in Fig. 1f. We fabricated PUC devices where the sensitizer molecules were fixed to a nanostructured alumina scaffold, with NODIPS-An filling the interstitial pores (NODIPS-An/sens/scaffold). Scanning electron microscopy (SEM) images and the reflectance spectrum of the dyed-alumina scaffold are shown in Supplementary Figs. 7 and 8. This device architecture follows our previous work33, but with solvent replaced by the neat, room-temperature liquid NODIPS-An chromophore. A schematic of the device is shown in Fig. 2a, along with the upconverted emission in Fig. 2b after 670-nm excitation. The difference in appearance of this spectrum compared with that in Fig. 2b is due to the array detector used for measuring upconversion.

Fig. 2: PL measurements and the upconversion device.
Fig. 2: PL measurements and the upconversion device.The alternative text for this image may have been generated using AI.
Full size image

a, Cartoon depiction of the NODIPS-An scaffold device with the upconversion process depicted in the device enlargement; the processes depicted are photon absorption and emission (hν) TET, TTA and FRET. b, Steady-state upconversion spectra with excitation at 670 nm (650-nm short-pass filter). c, Integrated magneto-PL exciting at 670 nm and detecting the upconverted emission. d, Heat map of delayed NODIPS-An fluorescence sensitized by PdPQ4NACA on alumina. e, Integrated emission from d, fitted to the square of the difference of two exponentials, which is the expected analytic form for a homogeneous sample in the absence of strong second-order decay. The rise time is 520 ns, with a 37 μs decay.

Source data

Figure 2c shows the magnetic field effect on the upconverted photoluminescence (PL). It is well established that triplet fusion is hindered by magnetic fields, owing to the details of the spin-character of weakly exchange-coupled triplet pairs47,48,49,50,51. At low field, for materials that do not rotationally diffuse on the timescale of TF, the PL is increased. This gives rise to a characteristic curve where the TF PL yield initially increases and then falls. This is the opposite effect to that observed in solid-state singlet fission systems52. In this case, the effect is one of the largest observed, with the PL diminished by about 8% at high field and enhanced by 2% at 25 mT. These results confirm that TF is effectively taking place in an annihilator material that is solid on excitonic timescales.

Time-resolved PL of upconversion in devices

Triplet energy transfer (TET) from the sensitizer to NODIPS-An was investigated by time-resolved photoluminescence (TRPL)53. The TRPL of upconverted fluorescence is shown in Fig. 2d,e. After a short rise time, the upconverted emission around 550 nm decays with a lifetime of 37 µs, commensurate with the expected lifetime of triplet excitons. A fit to the square of the difference of two exponentials is shown in Fig. 2e, which is the expected analytical form for upconverted emission if the decay is dominated by first-order processes54. The discrepancy of the fit is likely to be due to inhomogeneity, or a significant second-order component. No better fit was obtained by attempting to take account of second-order decay. The rise time of 0.52 µs is surprisingly long. A mixture of the sensitizer and NODIPS-An exhibited a rise time of 320 ns (Supplementary Fig. 10), so the long triplet transfer lifetime is not due to a hindered interaction between the species in the device but is intrinsic to this pair of chromophores.

TA spectroscopy

To further characterize NODIPS-An and the dynamics of triplet–triplet annihilation (TTA) within the device, we turned to femtosecond transient absorption (TA) spectroscopy. Figure 3a displays the TA spectra of neat NODIPS-An following excitation at 410 nm. There is evident stimulated emission (SE) and excited-state absorption (ESA) at 470 nm and 525 nm, respectively. The transient spectra appear initially similar to those for free chromophores in solution (Supplementary Fig. 15) but rapidly degrade to a featureless ESA with no apparent SE or ground-state bleach (GSB). This indicates that the initial excitation has moved from (relatively) excitonically uncoupled chromophores with structured absorption spectra to the tightly coupled chromophores evidenced in the absorption spectrum in Fig. 1b and revealed by MD simulations.

Fig. 3: TA spectroscopy.
Fig. 3: TA spectroscopy.The alternative text for this image may have been generated using AI.
Full size image

a, Time-slices of neat NODIPS-An with a pump excitation at 410 nm (ΔmOD, change in thousandths of optical density). b, Kinetic plots of neat NODIPS-An GSB (465–480 nm) and ESA (520–540 nm). The kinetic trace is fitted with a Gaussian convoluted with a biexponential function (red line). c, Time-slices of a device with pump excitation at 670 nm. d, Kinetic trace of the device ESA (550–580 nm). The trace is fitted with Gaussian convoluted biexponential and an offset. Shaded areas in a and d correspond to the integrated areas tracked in the kinetic plots. The fitting results along with errors and relative contribution are displayed in the kinetic plots.

Source data

Figure 3b tracks the evolution of the SE and the structured ESA as a function of pump–probe time delay. The sharp SE decays in just ~10 ps indicating rapid singlet exciton migration55. The chromophores barely move on this timescale, and because the SE disappears, the excited, relatively uncoupled, chromophore must have returned to the ground state. The excitation has moved in this time to chromophores without a sharp absorption spectrum. In the fit in Fig. 3b, the final decay constant has been fixed at 23.6 ns to match the PL lifetime. There is an intermediate time constant which depends on the spectral region being fit, which probably reflects the diversity of tightly coupled chromophore pairs that can accept the excitation. These tightly coupled excited dimers are the excimers evidenced by the emission spectrum in Fig. 1b. After about 1 ns, there is no further spectral evolution, indicating that the exciton is trapped at the excimer-forming site.

The femtosecond TA of the device in Fig. 3c shows the dynamics of the porphyrin sensitizer in the first 8 ns. There is no evidence of energy transfer to NODIPS-An within 8 ns, as expected from the TRPL in Fig. 2e. The porphyrin exhibits rapid thermalization on a 2-ps timescale, followed by loss of the stimulated emission feature around 700 nm on a subnanosecond timescale. Preservation of the Q-band GSB on the same timescale indicates quantitative ISC to the long-lived T1 triplet state. We thus conclude that \({\varPhi}_{\mathrm{ISC}} \approx 1\). There is some fast decay of the bleach signal around 500 nm, which is investigated further by nanosecond TA.

To further characterize the dynamics of the device, we performed nanosecond TA on a control sample without NODIPS-An (sens/scaffold) and compared it with the PUC device. The heat maps for devices with and without NODIPS-An are shown in Fig. 4a,b. The sample with sensitizer only (Fig. 4a) shows the characteristic transient spectrum observed in Fig. 3c. There is an evident ESA rise in the PUC device in Fig. 4b in the region of 500–520 nm that is not apparent in Fig. 4a. To track the bleach decay of the control in Fig. 4c, three exponentials are required, with the longest timescale being 2 µs. Three exponentials are still required for the PUC device with longest timescale shortening from 2 µs to 500 ns, in line with the observed rise in the upconverted emission (Fig. 2e). The intermediate timescale is roughly a quarter of that without NODIPS-An present. Both with and without NODIPS-An, there is a fast decay with a timescale of about 5 ns. This accounts for about half of the decay, and as such we contend that this represents excitons that decay without achieving TET.

Fig. 4: Nanosecond TA spectroscopy.
Fig. 4: Nanosecond TA spectroscopy.The alternative text for this image may have been generated using AI.
Full size image

a,b, Heat maps of a device containing only the sensitizer PdPQ4NACA (a) and a device with NODIPS-An (b). Reference lines indicate the areas integrated for the kinetic traces. c, Kinetic trace for the region of 470–490 nm tracking the GSB for the sensitizer with and without NODIPS-An. Solid lines are triexponential fits of the kinetic traces with the resulting lifetimes and weights reported in the figure. d, Kinetic traces of PdPQ4NACA phosphorescence.

Source data

The TET efficiency after this initial decay is gauged by phosphorescence quenching. Figure 4d shows integrated phosphorescence between 820 nm and 860 nm as a function of time for devices with and without NODIPS-An. Without NODIPS-An, the phosphorescence after 10 ns decays with a long lifetime typical of molecular triplet states. With NODIPS-An, it decays on a 100-ns timescale, and thereafter on a tens of microseconds timescale. This longer lifetime, which represents only a few percent of the decay, is probably indicative of a small amount of reabsorption from upconverted light53. The TET efficiency in the period post 10 ns is estimated to be 98%. Considering the 48 ± 2% of excitons lost in the initial decay, we estimate the overall TET efficiency to be 49 ± 2%.

The upconversion quantum yield of the devices is given by the product of the quantum yields of the events illustrated in Fig. 1f,

$${\varPhi }_{{\mathrm{UC}}_{{\rm{g}}}}={\varPhi }_{\mathrm{ISC}}{\varPhi }_{\mathrm{TET}}{\varPhi }_{\mathrm{TTA}}{\varPhi }_{{\rm{f}}}\left(1-{\varPhi }_{\mathrm{ET}}\right)$$
(1)

Here, \({\varPhi}_{\mathrm{ET}}\) accounts for the undesired back transfer of energy from the upconverted singlet state to the sensitizers, which we aim to suppress. \({\varPhi }_{\mathrm{TTA}}\) has a maximum of 0.5, so the maximum quantum yield of PUC is 50%. Our previously reported apparatus33 compares measured upconversion power to scattered light from a Lambertian surface. We calculated the quantum yield at the point of isotropic emission, taking into account the refractive index of the emitting medium. In terms of the conventions supported by our group56, this yield corresponds to the generation quantum yield \({\varPhi }_{{\mathrm{UC}}_{{\rm{g}}}}\). This is the yield of photons generated by the upconversion process, before losses due to reabsorption and waveguiding. Reabsorption losses are estimated to be 14% by comparison of the emission spectra due to upconversion and direct excitation (Supplementary Fig. 9). Collection efficiency and waveguiding is accounted for by comparison with a fluorescence standard33. The champion device recorded \({\varPhi }_{{\mathrm{UC}}_{{\rm{g}}}}=8.2\left(4\right) \%\). Accounting for the quantum yield of the NODIPS-An (43%), we can estimate that the excited-state upconversion yield was \({\varPhi }_{{\mathrm{UC}}_{{\rm{s}}}}=19.0\left(8\right) \%\). This is about 40% of what is possible in a PUC system (\({\varPhi }_{{\mathrm{UC}}_{{\rm{s}}}}=50 \%\)). As we have determined that the TET efficiency is about 49%, this step is evidently responsible for the majority of efficiency losses—the product \({\varPhi }_{\mathrm{TTA}}\) \(\left(1-{\varPhi }_{\mathrm{ET}}\right)\) is determined to be 39%, which is the upper limit to a device where the identified losses are mitigated. Given that \({\varPhi }_{\mathrm{TTA}}\) < 0.5, we conclude that \({\varPhi}_{\mathrm{ET}}\)  < 0.22.

Discussion

The principal design issue with solid-state annihilators is that the excited state is hypermobile. The excited singlet state moves by the FRET mechanism and can diffuse on the order of tens of nanometres. As such, this ensures that the excited state can access the lower singlet level of the sensitizer, which results in an efficiency loss as two excitons become one recycled exciton. Here we have found that excimer formation in a nanostructured solid-state device guards against mobility by shifting the emission spectrum of the annihilator away from its absorption spectrum to attenuate self-FRET. In other words, the excimer acts as a means to trap the excitation on the annihilators, thereby reducing back transfer to the sensitizer. In the present case, excimer formation reduces the fluorescence quantum yield from 78% to 43%. Although this is a significant drop, it is not catastrophic. Pyrene excimers are found to radiate 7.5 times more slowly than monomers, indicating near cancellation of the transition moment in the π-stacked geometry57. Here the radiative lifetime is 8.2 times longer in the neat chromophore compared with solution. The fluorescence quantum yield of pyrene dimers in crystals was found to be 68%, suggesting that excimer formation itself should not catastrophically affect upconversion efficiency58. We posit that there is a ‘Goldilocks zone’, wherein there is sufficient excitonic coupling to reduce mobility, and transition moment cancellation is mitigated such that the fluorescence quantum yield remains high. It is possible that the introduction of chiral structures could mitigate transition moment cancellation by inducing a twist in chromophore alignment. Incorporating such annihilators into a device and improving the TET efficiency between the sensitizer and the upconverter would lead towards a device that approaches the thermodynamic limit. We propose to incorporate TF moieties into the sensitizer structure to achieve this, as has been successful with nanocrystals59,60 and dyes61.

Efficient solid-state upconversion remains a central challenge for TF-PUC systems. Here we have demonstrated that neat NODIPS-An, which behaves as a solid on experimental timescales, can capture the excitation produced during TTA as an excimer state. Here the utility of the chromophore’s liquid state lies in its ease of processing and rapid annealing, enabling it to infiltrate the pores of the scaffold. It would be straightforward to modify the material so that its melting point is still accessible, but above room temperature. We expect that low-melting-point, room-temperature solids may display similar behaviour. By trapping the excitation, the emission of NODIPS-An is shifted away from its own absorption band and its mobility is reduced, thereby suppressing back-energy transfer to the sensitizer, which is a major loss pathway in solid-state annihilators. This exciton-trapping strategy offers a promising route to mitigate back-transfer losses and to advance efficient solid-state upconversion.

Methods

Organic syntheses

NODIPS-An and PdPQ4NACA were synthesized according to the detailed procedures in Supplementary Section 7 and the references therein62.

Molecular dynamics

Simulations of NODIPS-An were carried out using the same methodology and parameters as those for NODIPS-Tc in ref. 52, with the only differences being the molecule type, system size and simulation durations. The initial configuration comprised 343 molecules on a cubic grid at very low density (box size (28 nm)3) to avoid overlapping molecules. The system energy was minimized and then a short (~100 ps) simulation was carried out at constant volume and temperature at a high temperature of 600 K, during which the simulation timestep was progressively increased to its final value of 1 fs. The simulation was then continued at constant pressure and temperature at 600 K and 100 atm for 1 ns to shrink the system to liquid density and to randomize the molecular positions and orientations. Owing to the low molecular mobility of pure NODIPS-An at room temperature, equilibration of this system at 1 atm was carried out in stages, with the temperature progressively decreased from 500 K to 450 K to 400 K to 350 K to 298 K, with sufficient time spent at each intermediate temperature to reach equilibrium. In addition, between the 350 K and 298 K simulations, the temperature was decreased at a constant rate over a period of 124.8 ns. Details of the simulations are given in Supplementary Table 1.

Sample preparation

Mesoporous Al2O3 films were deposited on 1-mm-thick, cleaned (acetone and isopropyl alcohol) microscopic glass slides (Livingstone) cut to 2 cm × 2 cm size. The films were fabricated on the glass slides by doctor blading an Al2O3 paste comprising 10-nm Al2O3 nanoparticles (Solaronix SA). Scotch tape (3M) was used to mask out an active area of 8 mm × 8 mm. The films were then sintered at 420 °C for 30 min in air to form the mesoporous scaffolding layer. Film thickness was measured using a Dektak profilometer (ANFF). The sensitizer solution (200 µM) was prepared by dissolving palladium(II) amino-nitro-tetrakisquinoxalino-porphinecarboxylic acid (PdPQ4NACA) dye at a concentration of 200 µM in 1:1 acetonitrile:2-butanol (Sigma-Aldrich). The Al2O3 films were soaked in the sensitizer solution in a closed glass jar at room temperature for 30 min. After soaking, the films were carefully removed from the solution and thoroughly rinsed several times in pure solvent (acetonitrile and 2-butanol) to remove excess sensitizer molecules, and then gently blow dried in N2. The films were then transferred into the glovebox (N2, MBraun) to carry out the remaining fabrication steps. Neat NODIPS-anthracene was dropcast onto the sensitized films and heated at 120 °C for 15 min. The films were then sealed using a Meltonix gasket sealing film (60 µm thick, Solaronix SA) and a microscopic glass slide was placed on top of the film (1 mm thick (Livingstone) cut to 2 cm × 2 cm size). Films for the low-temperature measurements were fabricated on sapphire substrates (Advaluetech).

Steady-state absorption spectroscopy

Optical absorption measurements were undertaken on a Shimadzu UV-2600 Spectrophotometer.

Electron microscopy

Top-view SEM images of an Al2O3 film (fabricated on a glass substrate) were taken using Raith 150 TWO. For clearer imaging, samples were subjected to a thin Au sputter coating (HHV-TF600).

Temperature-dependent measurements

Temperature-dependent upconversion emission measurements were carried out in an optically accessible cryostat (Montana Instruments) with 670-nm continuous wave (CW) laser excitation (Thorlabs) with a 600-nm short-pass filter (Thorlabs FES600). Emission was coupled to a multimode fibre and passed into a spectrometer (Thorlabs CS200/M).

Quantum yield measurements

Quantum yield measurements and steady-state PL were undertaken on a homebuilt breadboard apparatus (detailed in ref. 33) with a 670-nm Thorlabs CW laser for excitation, and the emission was collected using a 650-nm short-pass filter to block the excitation laser. Direct excitation of NODIPS-An was carried out with a 405-nm Thorlabs CW laser and a 425-nm long-pass filter.

Magneto-PL

The sample was excited with a CW laser (Thorlabs) and the upconverted PL was collected with off-axis parabolic mirrors focused into an optical fibre leading to a spectrometer (Flame, Ocean Optics). An electromagnet (Magnetech MFG-6-24) controlled by a DC power supply (Keithley 2230G-30-1) provided the magnetic field, calibrated with a Gaussmeter (Lakeshore 475). Five measurements were done at each data point; four data points measured the field at zero, and one on-field data point was randomized within those five measurements. The step size was set to random. Fluctuations in laser power were accounted for by performing ten iterations.

Transient PL

Samples were excited with an optical parametric oscillator laser system (Opolette HE 355 LD, Opotek Inc., 670 nm, 20 Hz, 10 ns pulse width). Sample PL was collected with a 90° parabolic mirror and coupled into a spectrograph (Princeton Instruments 2300i) equipped with a 30 mm−1 grating blazed at 500 nm and an intensified time-gated camera (Princeton Instruments PM4256f-HR-FG- 18-P43-SM). A notch filter blocked the excitation beam before the camera, and line pass filters cleaned up the pump beam.

TCSPC measurements were performed on a commercial set-up (FS5, Edinburgh Instruments). A 375-nm laser diode (pulse width ~100 ps, f = 1 kHz) was used to excite the sample. The excitation density was kept low enough to maintain single photon statistics. The instrument response function was obtained by collecting the light scattered by a clean substrate placed in the sample region.

Femtosecond TA spectroscopy

The TA measurements were performed on a commercial set-up (Ultrafast Systems, Helios) pumped by a titanium:Sapphire (Ti:Saph) amplifier (Spectra Physics Solstice Ace) that generates an 800-nm pulse train (1 kHz, 100 fs). The 800-nm output was split with a beam splitter, and one part of the pulse was sent to an optical parametric amplifier (TOPAS Prime, Light Conversion) to generate the desired pump pulse (410 nm or 670 nm) by means of sum frequency generation. A grid polarizer and a half-wave plate set the pump beam to the magic angle (54.7°) relative to the probe. The second part of the 800-nm pulse was focused onto a sapphire crystal to generate a visible white light continuum probe. An 8-ns optical delay line with a minimum step size of 14 fs varied the pump–probe delay. The signal was detected with a silicon linear array detector.

Nanosecond TA spectroscopy

The nanosecond TA measurements were carried out using a commercial flash photolysis instrument (en-VISion, Magnitude Instruments). Here the 800-nm output from the Ti:Saph amplifier (Spectra Physics Solstice Ace) was sent to an optical parametric amplifier (TOPAS Prime, Light Conversion) to generate the 670-nm pump laser (500 Hz) and excite a (diameter = 5 mm) spot on the sample. A xenon lamp was used for the visible probe region, and overlapped on the sample at 90° to the pump. The probe transmitted through the sample was passed through a monochromator for wavelength separation, then onto fast Si or InGaAs photodiodes (for the visible and near-infrared probes, respectively) for time-resolved signal acquisition.