Mixed halide bulk perovskite triplet sensitizers: Interplay between band alignment, mid-gap traps, and phonons

The steady increase in the power conversion efficiency (PCE) of lead halide perovskite photovoltaics (PV), going from 3.8% to 25.5% in 12 years, has piqued significant interest in large part due to their low cost and facile fabrication.^1,2 Key factors in this rapid PCE increase are the unique properties of perovskites, including defect tolerance, long charge carrier lifetime and diffusion lengths, high absorption cross sections, and compositional bandgap and band energy tunability.^3,4

Due to the lower bandgap of iodide-based perovskites compared to other halide compositions, they are most commonly utilized for PV applications. However, the addition of small amounts of chloride or bromide into iodide-rich films has shown to increase the carrier lifetimes. The addition of bromide, in particular, has shown to be a viable route to fabricating pinhole-free and smooth perovskite thin films, improving the long-term stability, increasing carrier transport and diffusion lengths, and decreasing nonradiative recombination.^5–7 Although large amounts of bromide addition result in a significant blue shift of the emission and increase in the bandgap, modest bromide addition of up to 15% only slightly shifts the absorption onset.

Recently, in addition to their function as the absorber layer in perovskite PV, bulk lead iodide perovskites consisting of various amounts of methylammonium (MA) and formamidinium (FA) as the A-site cation have shown great promise as triplet sensitizers for photon upconversion (UC) in rubrene through triplet–triplet annihilation (TTA). The same properties, which make perovskite materials uniquely qualified as the active layer in a PV, enable triplet sensitization through a free-charge carrier process.^8–15 

Photon UC, in general, describes the process of increasing the emitted photon energy upon photon absorption.^16–20 To comply with energy conservation laws, this process requires the combination of the energy of two absorbed photons to generate a single high energy excited state.^21,22 In TTA, this process occurs through the combination of long-lived triplet states, which enable the storage of the energy prior to the TTA-UC process.^18,22,23

In bulk perovskite-sensitized UC, the triplet state is accessed by an asynchronous hole transfer from the valence band (VB) of the perovskite to the highest occupied molecular orbital (HOMO) of rubrene, coupled to an electron transfer from the conduction band (CB) of the perovskite to form the bound triplet exciton.^8,10 In the current iodide-only perovskite UC systems, the band alignment is such that hole transfer is strongly energetically favored, while electron transfer is nearly isoenergetic. The exact energy alignment depends on the MA/FA ratio used, which tunes the doping of the film (n-type vs p-type)²⁴ and the absolute energy levels of the VB/CB. Generally, a lower MA content results in shallower VB and CB levels, which can potentially increase the energetic driving force for electron transfer.²⁵ However, this is counteracted by a higher absorption of the FA-rich perovskites, which we have previously shown to exhibit higher UC photoluminescence (PL) intensities than their MA-rich counterparts. Other important factors to consider are the native PL lifetimes and resulting carrier diffusion lengths,¹⁴ Fermi levels,²⁴ and interfacial band bending,^11,26 which can all influence the number of charges extracted at the perovskite/rubrene interface. We have previously shown that one of the current limitations of the perovskite/rubrene UC system is incomplete charge extraction, which can be attributed to the lack of driving force for electron transfer, a limited driving force for charge transport to the interface, localized trap states, or charge transfer states mediating the charge transfer.^9,11,15,26

In this study, we investigate the effect of bromide addition up to 15% on the TTA-UC process in perovskite/rubrene systems. The bilayer UC devices consist of an ∼100 nm thin perovskite film based on FA_0.85MA_0.15Pb(I_1−xBr_x)₃ (x = 0–0.15) onto which the rubrene annihilator layer doped with ∼1% dibenzotetraphenylperiflanthene (DBP) is spin-coated.²⁷ To avoid quenching of the generated triplet states by molecular oxygen, the devices are encapsulated in a nitrogen environment with a two-part epoxy.²⁸ We find that small amounts of bromide increase the native perovskite PL quantum yield (QY) and lifetime, which results in an increased UC yield due to the competition between interfacial carrier trapping and charge transfer to rubrene.²⁶ Interestingly, we find that even sub-bandgap photons can be absorbed and subsequently upconverted, indicating the possible role of mid-gap trap states and thermal excitation to the band edge by the rich phonon bath in perovskites in the UC process.^29–31 

The base perovskite composition used is FA_0.85MA_0.15Pb(I_1−x,Br_x)₃, where x is varied from 0 to 0.15. To elucidate the role of bromide addition on the TTA-UC process, the bilayer UC devices are fabricated by spin coating a 10 mg/ml rubrene solution doped with 1% DBP onto the base perovskites. A postfabrication thermal annealing step is utilized to increase the UC performance and to drive off excess solvent.³² 

As expected, we find an increase in the bandgap upon Br addition with a corresponding blue shift of the absorption onset and emission peak wavelength from 800 nm for the iodide-only perovskite FA_0.85MA_0.15PbI₃ to 760 nm for FA_0.85MA_0.15Pb(I_0.85,Br_0.15)₃, as shown in Fig. 1(a). There are no additional shifts in the absorption onset or emission wavelength upon rubrene deposition, and both the rubrene/DBP and perovskite emissions are clearly observed under 405 nm excitation [Fig. 1(b)]. For comparison, the absorption and emission spectra of a rubrene thin film are shown in Fig. S1.

A strong increase in the absolute PL intensity is seen upon bromide addition (Fig. 2), highlighting the expected increase in radiative recombination caused by a reduced number of nonradiative carrier traps. Due to photodarkening of the perovskite under continuous illumination and halide segregation occurring (vide infra), the absolute PL intensity, however, must be taken with a grain of salt as it just depicts a snapshot of the PL at a certain time after illumination. Halide segregation and the resulting red shift of the PL peak intensity are a known issue in mixed halide perovskite thin films.^33,34 Halide migration under illumination results in phase-segregated bromide-rich and iodide-rich regions, of which the latter can act as charge carrier recombination centers, hindering the effective flow of charge carriers to the interface at which they are extracted. To infer whether halide migration has an influence on the UC intensity, we first investigate the effect of continuous illumination on the mixed halide perovskite compositions and corresponding UC devices presented here. Figures 2(a) and 2(b) depict the perovskite emission under 405 nm excitation over the course of 120 s for the perovskite and the bilayer UC devices, respectively.

We observe the characteristic red shift of the PL, indicating halide segregation for the mixed halide perovskites.^35–37 As expected, with an increase in the bromide concentration, the magnitude of the observed red shift increases: Δλ = 5 nm, Δλ = 15 nm, and Δλ = 35 nm for 5%, 10%, and 15% bromide, respectively. A smaller red shift can be observed in the mixed halide perovskite PL intensity under 780 nm excitation (Fig. S2), highlighting the expected effect of excitation photon energy on the halide migration process. In addition, both photobrightening and photodarkening of the perovskite PL have previously been reported,^38,39 as well as photodarkening followed by photobrightening.²⁶ These effects are strongly dependent on the photon fluence, the excitation wavelength, and the local perovskite environment.^38,39 Due to spot-to-spot variations in the relative PL behavior due to local inhomogeneity,^26,40,41 the change in the absolute intensity of the perovskite PL is not considered here. No changes in the rubrene/DBP emission are found under direct excitation, indicating that the emissive dye is stable and does not degrade under these conditions.

Since the triplet sensitization is thought to occur through an asynchronous free-charge carrier transfer from the perovskite to the triplet state of rubrene,¹⁰ the band alignment is the key driving force for charge transfer to the triplet state of rubrene, and the composition-dependent changes in the absolute VB and CB of the mixed halide perovskites are critical in this study. To first approximation, the addition of bromide is expected to shift the VB to a deeper level, and the CB to a shallower level, effectively increasing the driving force for both electron and hole transfer (Fig. 3).^25,33

A recent study also reports shallower energy levels for a related mixed halide perovskite: MAPb(I_0.875,Br_0.125).⁴² The 200 meV shift of the CB, VB, and Fermi levels observed in this system would also result in an increase in the driving force for electron transfer to the rubrene triplet, however, at the expense of a reduction in the driving force for hole transfer, which would still remain energetically favored. The predicted increase in the driving force for charge transfer to the triplet state upon bromide addition, however, comes at the cost of a wider bandgap, thus a reduced absorption in the near-infrared spectral region.^34,42

To gain deeper insight into the effect of the band alignment on the triplet sensitization and quenching of the perovskite excited state, we turn to time-resolved PL spectroscopy. A slight decrease in the lifetime can be observed upon addition of 5% bromide, however, with a corresponding increase in the amount of radiative decay, as observed in the increased absolute PL intensity (see Fig. 2). With a further increase in the bromide content, the average PL lifetimes become longer [Fig. 4(a) and Table I]. In agreement with previous results, we find the amplitude of the rapid early time component to increase with an increase in the bromide concentration, while the lifetime at late time increases up to our highest bromide content of 15% [inset, Fig. 4(a)].³³ Due to the multiexponential behavior of the perovskite PL decay, we have fit each intensity decay to a triexponential as shown in the following equation:

All perovskite PL lifetimes are taken under 635 nm illumination and at a constant fluence of 2.8 W/cm² and, thus, result in different carrier concentrations in the perovskite thin films due to different absorption cross sections based on the bromide content. These changes in the native lifetime are expected due to the carrier density-dependent recombination dynamics observed in perovskite thin films.^43–45 With an increase in the carrier density, the early time component becomes slower due to reduced rapid nonradiative decay, while the late time dynamics show an increased rate of recombination due to a higher probability of recombination.

For all FA_0.85MA_0.15Pb(I_1−x,Br_x)₃/rubrene bilayer devices, consistent with our previous results, we find quenching of the perovskite PL dynamics under 635 nm excitation [Fig. 4(b)].^10,26,46 We observe similar early time dynamics for all compositions; however, the amplitude of this component shows spot-to-spot variations (Fig. S3). The decay rate of the late time component is clearly dependent on the underlying composition. However, care must be taken when analyzing these results. For the related MA-rich perovskite/rubrene UC devices, we have shown that there is no direct correlation between the extent of early time perovskite PL quenching and the UC PL intensity. Rather, we found that the lesser the observable early time quenching or the longer the average PL lifetime, the higher the UC PL intensity.²⁶ The early time component is directly dependent on the underlying trap density, and carrier trapping and triplet sensitization have been shown to be competing pathways.^14,26

Furthermore, perovskites exhibit power-dependent lifetimes: at lower carrier densities, the early time component becomes faster, while the long-lived tail of the perovskite emission becomes longer.^10,15 This general behavior is in agreement with our perovskite emission decays: once rubrene is added, the early time is quenched, while the lifetime at long times is elongated due to a lower carrier density, indicating efficient charge extraction.

In addition, the physical process of charge transfer to the rubrene triplet state is expected to be an ultrafast process on the sub-nanosecond timescale,^47,48 which cannot be captured within the time frame accessible by the time-correlated single-photon counting used here. Therefore, the detected early time component is more likely a convolution of “slow” carrier migration to the interface and subsequent carrier extraction on the ultrafast timescale, as well as residual charge carrier recombination well within the bulk of the perovskite material.

Based on these results, we expect that all perovskite compositions are capable of sensitizing the rubrene triplet state due to the observed quenching of the PL lifetime, indicating that charge extraction is occurring. Interestingly, despite similar early time dynamics, we find the late time dynamics to become elongated with an increase in the bromide content, indicating a lower remaining carrier density and, thus, a lower recombination probability. Considering the longer native lifetime of the perovskite with an increase in the bromide content, similar residual decay dynamics in the bilayer devices still correspond to large increases in the charge transfer efficiency to the rubrene triplet state. This increase in the number of charges transferred can be attributed to the increased carrier lifetimes in the mixed halide perovskite, enabling a longer carrier diffusion length prior to recombination or charge trapping and the increased driving force for charge transfer at the interface due to the more favorable band alignment.

Finally, to ensure that the dynamics of the rubrene layer are not affected by the addition of bromide, we report the dynamics of the upconverted emission [Fig. 4(c)]. Here, a characteristic rise and fall in the UC PL dynamics are observed, as the emissive singlet population is rate-limited by the diffusion-mediated TTA process. We have previously reported triplet population-dependent rise dynamics of the UC PL, a result of two varying types of TTA: (i) rapid interface-mediated TTA at the interface, which results in a rapid early rise time, and (ii) diffusion-mediated TTA far from the interface, which yields a slower rise time as it is rate-limited by the slow triplet diffusion process.⁴⁶ 

Interestingly, the brightest UC is found for FA_0.85MA_0.15Pb(I_0.95,Br_0.05)₃ (vide infra), which exhibits less rapid early time rise than FA_0.85MA_0.15PbI₃ (see Fig. 5). This suggests that the different interfacial band alignment upon bromide addition allows more carriers/triplets to diffuse away from the interface prior to TTA, indicating that the triplets are not as localized to the interface as in the iodide-only composition. We further find a correlation between the fraction of the rapid rise component and the bromide content: less rapid rise component is found for higher bromide concentrations. This observation can simply be traced back to the varying amounts of absorption at the excitation wavelength at 780 nm (see the supplementary material, Table S1). A lower absorption of the excitation laser results in less triplet states generated in rubrene, resulting in a long diffusion of triplets prior to TTA, yielding more diffusion-mediated TTA further from the interface.

To further investigate the UC properties of our bilayer devices, we determine the steady-state UC PL intensity under 780 nm excitation. The visible PL intensity under 780 nm excitation is related to the external UC efficiency, defined as the number of incident photons converted to UC photons. The external UC QY is dependent on the efficiency of triplet sensitization ϕ_TET, the efficiency of TTA ϕ_TTA, and the QY of the annihilator ϕ_ann. In contrast to previous excitonic systems,^18,49 the efficiency of intersystem crossing ϕ_ISC can be ignored as this process does not occur in the perovskite-based TTA-UC system presented here,

Both ϕ_TTA and ϕ_ann are specific to the annihilator, which is the same in all cases; hence, the differences in observed ϕ_UC can only be related to ϕ_TET. As we have previously shown, the absolute external UC quantum efficiency (ϕ_UC) is difficult to obtain for the perovskite-based UC systems due to the effects of perovskite photobrightening resulting from trap filling, particularly at low solar-relevant fluences.²⁶ As a result, we refrain from reporting absolute values and simply focus on relative PL intensities.

Under 780 nm excitation, to first approximation, we would expect only the compositions that have a bandgap larger than 1.59 eV (780 nm) to be capable of UC. However, we find that all compositions show a strong UC PL, as shown in Figs. 5(a) and S4. For additional statistics, two independent sample sets are compared. The inset in Fig. 5(a) shows a box plot of the integrated UC PL intensity (integrated from 500 to 630 nm) taken at 20 spots across each sample, highlighting the heterogeneity of the fabricated samples. We find the highest UC PL intensity for FA_0.85MA_0.15Pb(I_0.95,Br_0.05)₃, which can be attributed to a combination of the increased driving force for electron extraction and elongated charge carrier recombination dynamics with a strong underlying absorption at 780 nm. The second highest UC PL intensity is found for the iodide-only FA_0.85MA_0.15PbI₃. Beyond the 5% bromide content in FA_0.85MA_0.15Pb(I_0.95,Br_0.05)₃, we find an increase in the bromide concentration to yield a lower ϕ_UC, respectively.

To investigate the possible effect of halide migration on the UC PL, we track the UC PL intensity under continuous illumination for 120 s [Fig. S2(b)]. We do not observe a change in the UC PL intensity for the low bromide content of up to 15% based on halide migration. Rather, the previously observed dependence of the UC PL intensity on the underlying perovskite PL intensity is found; if the underlying perovskite photobrightens, the UC PL increases, while a photodarkening of the underlying perovskite PL results in a decrease in the UC PL intensity. This indicates that the slight effects of halide segregation under 780 nm excitation in the perovskite thin films including up to 15% bromide do not influence the charge transfer properties to the rubrene triplet level on the ensemble level investigated here. Rather, it is even possible that halide segregation can “concentrate” the charge carriers in iodide-rich domains, which are located at the interface, resulting in an increase in the probability of transfer to the triplet state of rubrene.

However, to clearly understand the effect of the underlying bromide content on the UC PL intensity, in particular, to understand the effect of the band alignment on ϕ_TET, the changes in the absorption at the excitation wavelength must be considered. To first approximation, based on our hypothesis that the band alignment is the critical factor in the charge extraction process, we would expect the highest bromide content to yield the highest UC efficiency relative to the number of absorbed photons. Normalizing the UC PL intensity by the overlap integral of the perovskite absorption and the excitation laser line shape J_overlap (see Table S1) yields the UC PL intensity relative to the absorbance of the perovskite thin film, a metric related to the internal UC quantum efficiency (ϕ_{UC_Int}),

Indeed, we find the perovskite composition with the highest bromide content FA_0.85MA_0.15Pb(I_0.85,Br_0.15)₃ to exhibit the highest ϕ_{UC_Int} [Fig. 5(b)]. This confirms our hypothesis that the underlying perovskite PL recombination dynamics and energetic driving force for concurrent electron and hole extraction are the key factors limiting the UC efficiency in the perovskite/rubrene UC devices.

Finally, we address the question how it is possible to excite the mixed halide perovskites using 780 nm excitation, which is slightly sub-bandgap in energy for the higher bromide concentrations of 10% and 15%, which emit at 775 and 760 nm, respectively. This observation is directly related to the defect tolerance of perovskite materials and the rich defect manifold available within the bandgap. Trap states in perovskites are classified as deep or shallow traps with the distinction that carriers caught in shallow traps can be de-trapped by the ambient thermal energy kT. In addition to trap states caused by atomic defects, the Urbach tails stemming from lattice disorder, doping, or thermal motion can result in shallow traps found within the bandgap.^50–53 Direct pumping of shallow traps has previously been reported in perovskites,⁵⁰ and the rich phonon bath present in the bulk perovskite enables rapid thermal excitation from the sub-bandgap trap states at ∼1.59 eV to the band edge at 1.63 eV for FA_0.85MA_0.15Pb(I_0.85,Br_0.15)₃.^29,30 The observation of the band edge FA_0.85MA_0.15Pb(I_0.85,Br_0.15)₃ PL under 780 nm pump is a clear indication that this thermal excitation is occurring (Fig. 5). To support our hypothesis that the band edge levels, rather than the sub-bandgap traps, are the mediating factors for charge transfer to the rubrene triplet state, we refer back to the PL dynamics of FA_0.85MA_0.15Pb(I_0.85,Br_0.15)₃ in the presence of rubrene (vide supra). If charges were not thermally excited to the band edge, we would not be able to track them via PL. In addition, if the charges were extracted to the triplet state of rubrene prior to their thermal excitation to the emissive band edge, we would not expect to see strong effects on the PL decay dynamics. Once the band edge is populated via thermal excitation, subsequent charge extraction enables the population of the triplet state of rubrene, which results in the observed TTA-UC process. A summary of the ongoing processes is shown in Fig. 6.

In summary, the effect of the halide composition on the TTA-UC process in rubrene has been investigated by changing the bromide fraction from x = 0 to 0.15. The increased carrier recombination lifetimes observed in mixed halide perovskite thin films with an increase in the bromide content combined with a more beneficial band alignment enable a more efficient charge extraction to the triplet state of rubrene, as observed in the higher internal UC efficiency ϕ_{UC_Int}.

From a practical device perspective, however, the external UC efficiency ϕ_UC is the relevant metric. Only absorbed photons can be upconverted through the TTA process. We find that slight blue shifts of the perovskite bandgap due to 5% bromide content are counteracted by the increased driving force for charge transfer to the rubrene triplet state, thus enabling a higher ϕ_UC than in the iodide-only counterpart. Despite more efficient charge extraction and a higher internal ϕ_{UC_Int}, higher bromide contents substantially reduce the absorption in the near-infrared spectral region, thus limiting the achievable apparent anti-Stokes shift in the TTA-UC process.

Halide segregation does not have a detectable influence on the UC properties investigated on the ensemble level here. However, further studies with a higher spatial resolution approaching the diffraction limit may yield more localized information on the role of phase segregation on the PL properties on the sub-micrometer scale.

More generally, the results presented here indicate that the main limiting factors in the efficiency of perovskite-sensitized TTA-UC are the energetic alignment between the perovskite sensitizer and the triplet acceptor and the defect density of the bulk perovskite thin film. To date, only the annihilator rubrene has successfully been utilized in perovskite-sensitized TTA-UC. This is in large part due to restrictions in annihilators capable of efficient solid-state UC and a lack of other known annihilators with a proper band alignment with the perovskite sensitizers. Future efforts should be focused on finding new sensitizer/annihilator pairs with a more beneficial band alignment while minimizing the energetic losses for interfacial charge extraction.

Glass substrates (University Wafer) were cleaned by sonication in solvents and UV–ozone treatment (Ossila) prior to perovskite film deposition. All precursor solutions were prepared from a 1.2 M PbI₂ (TCI) and 1.2 M PbBr₂ (TCI) stock solution in anhydrous DMF:DMSO (9:1, v:v, Sigma-Aldrich). 1.2 M MAI (Dyenamo), 1.2 M FAI (Dyenamo), and 1.2 M MABr (Dyenamo) were then mixed with the respective stock solution in a 1:1.09 ratio. To obtain the various mixed halide compositions, different ratios of these solutions were used. The films were spin-coated at 1000 rpm for 10 s and at 5000 rpm for 30 s. During the second spin-coating stage, chlorobenzene (Sigma-Aldrich) was used as an antisolvent. The films were subsequently annealed at 110 °C for 45 min.

Rubrene (99.99%) and DBP (98%) were purchased from Sigma-Aldrich and used as received without further purification. A solution of 10 mg/ml rubrene was prepared in anhydrous toluene (Sigma-Aldrich) and doped with DBP at a concentration of ∼1%. The rubrene/DBP solution was deposited by spin coating onto the perovskite thin film at 6000 rpm for 20 s. The resulting bilayers were annealed at 100 °C for 10 min. The upconversion devices were sealed with a cover slip using a 2-part epoxy (Devcon) under nitrogen atmosphere (<0.5 ppm O₂) prior to removal from the glovebox.

Absorbance spectra were collected using a Thermo Scientific Evolution 220 spectrophotometer. The steady-state and time-dependent emission spectra were collected using 405 nm (LDH-D-C-405, PicoQuant) continuous wave excitation at a power density of 92 W/cm² with a 425 nm long pass filter (Chroma Tech) to remove excess laser scatter. The upconverted emission and time-dependent emission spectra were collected under continuous wave excitation using a 780 nm laser (LDH-D-C-780, PicoQuant) at a power density of 143 W/cm². Excess laser scatter was removed with a 700 nm short pass filter (ThorLabs) or a 780 nm notch filter (ThorLabs). An Ocean Insight emission spectrometer (HR2000+ES) was used to record the emission. Time-resolved PL decays were measured using a 635 nm pulsed laser (LDH-P-C-635M, PicoQuant) at a repetition frequency of 125 kHz and an average power density of 2.8 mW/cm². A 633 nm notch filter (ThorLabs) was used to remove excess laser scatter, and a 700 nm long pass filter (ThorLabs) was used to isolate the perovskite emission. UC PL dynamics were measured with a pulsed 780 nm laser (LDH-D-C-780, PicoQuant) at a repetition frequency of 31.25 kHz and an average power density of 17 mW/cm². Photon arrival times were recorded using a MultiHarp 150 event timer (PicoQuant) connected to a silicon single-photon avalanche photodiode (Micro Photon Devices). The incident laser beam power was measured using a silicon power meter (ThorLabs PM100-D). Spot sizes were measured using the razor blade method (90:10).

See the supplementary material for additional figures and data as detailed in the text.

The authors acknowledge funding by Florida State University. Use of the Center for Nanoscale Materials, an Office of Science user facility, was supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under Contract No. DE-AC02-06CH11357.

FSU has filed a provisional application for a U.S. patent based on this technology that names L.N. and S.W. as inventors.

A.S.B. and Z.A.V. contributed equally to this work.

The data that support the findings of this study are available from the corresponding author upon reasonable request.