Generating Light from Upper Excited Triplet States: A Contribution to the Indirect Singlet Yield of a Polymer OLED, Helping to Exceed the 25% Singlet Exciton Limit

The mechanisms by which light is generated in an organic light emitting diode have slowly been elucidated over the last ten years. The role of triplet annihilation has demonstrated how the “spin statistical limit” can be surpassed, but it cannot account for all light produced in the most efficient devices. Here, a further mechanism is demonstrated by which upper excited triplet states can also contribute to indirect singlet production and delayed fluorescence. Since in a device the population of these TN states is large, this indirect radiative decay channel can contribute a sizeable fraction of the total emission measured from a device. The role of intra‐ and interchain charge transfer states is critical in underpinning this mechanism.


DOI: 10.1002/advs.201500221
directly measured, especially in the luminescent materials used in OLEDs. Given that in an OLED very high triplet populations are generated in relatively well confi ned, thin recombination layers, typical 5-10 nm, [ 7 ] the triplet density will be very high and TTA very effi cient. Thus, TF will contribute to the number of singlets produced in the device, and this contribution will depend critically on the TF branching ratio in TTA. Jortner et al. [ 8 ] fi rst proposed that quantum spin statistics controls the process of singlet formation arising from the annihilation of two triplet excitons 3 A* + 3 A* 5/9 5 (AA)** 3/9 3 (AA)** 1/9 1 (AA)** 3 A*+ 1 A 0 or 3 A*+ 3 A* where 1 A * , 3 A * , and 5 A * are excited singlet, triplet, and quintuplet states and 1 (AA) * , 3 (AA) *,5 , (AA) * are interaction pairs with singlet, triplet, and quintuplet spin multiplicities. The subsequent decay of the triplet pairs yields an excited triplet exciton and a ground state molecule. In this scenario only 5.5% singlet states will be produced. However, evidence from OLED devices and TTA up-conversion measurements suggest that this value is incorrect. For example, Cheng et al. [ 9 ] have estimated a 0.125 singlet yield in their solution measurements. Kondakov et al. in detailed measurements of delayed electroluminescence from OLED's, [ 10 ] indirectly implied that the contribution to the overall electroluminescence yield of their devices from TF can be 20% [ 11 ] or 50% [ 12 ] depending on the emissive material. This would require a far higher TF singlet yield than 0.055.
OLEDs are very complicated systems where charge transport, exciton migration, exciton polaron quenching and other processes take place. This makes estimation of TF effi ciency from OLED measurements very diffi cult. Kondakov et al rationalized the 20% delayed fl uorescence (DF) contribution from TF by pointing out that quintuplet interaction pairs cannot participate in TTA because they are energetically untenable, the energy of two triplet excitons is not enough to produce a quintuplet state, [ 13 ] hence changing the dynamics described in Scheme I, and yielding a total singlet yield of 0.2, i.e., into account triplet recycling, that is the T N states produced by the triplet annihilation channel relax to T 1 states, that can then The mechanisms by which light is generated in an organic light emitting diode have slowly been elucidated over the last ten years. The role of triplet annihilation has demonstrated how the "spin statistical limit" can be surpassed, but it cannot account for all light produced in the most effi cient devices. Here, a further mechanism is demonstrated by which upper excited triplet states can also contribute to indirect singlet production and delayed fl uorescence. Since in a device the population of these T N states is large, this indirect radiative decay channel can contribute a sizeable fraction of the total emission measured from a device. The role of intra-and interchain charge transfer states is critical in underpinning this mechanism.

Introduction
From the outset of research into organic light emitting diodes (OLEDs) it has been assumed that the process of charge recombination which generates excitons is controlled by random spin statistics with singlet and triplet excitons formed in the ratio 1:3. Wohlgenannt et al. questioned this and by measurements of singlet and triplet formation cross sections concluded that the recombination process was in fact spin dependent in polymers. [ 1 ] This fi rst result stimulated many experimental and theoretical studies of possible physical mechanisms that would give rise to spin dependent recombination. [ 2 ] However, more recent studies have shown that the process of triplet-triplet annihilation (TTA) and the production of singlets from TTA, by triplet fusion (TF), [3][4][5][6] contributes to the secondary production of singlet excitons. The initial recombination process still obeys spin statistics whilst the total singlet production yield can exceed 25%. TTA is key to our understanding of triplet exciton dynamics in organic materials and therefore essential for a proper understanding of the use of organic materials in many applications. However, the effi ciency of TF has not been Indeed, in the seventies this had been shown experimentally in small organic molecules, for example, Groff et al. indirectly determined a 0.22 yield of TF to account for the delayed fl uorescence in anthracene [ 14 ] while Ern et al. found a very similar TF yield of 0.19 for pyrene dimers. [ 15 ] In our direct measurement of the singlet to triplet exciton yield in OLED devices based on the polyspirobifl uorene (PSBF) polymer, structure shown in Figure 1 , a total singlet yield of 0.44 was obtained. [ 16 ] Further, King et al. have also shown unambiguously that DF accounts for up to 25% of the total EL output in other polymer devices [ 17 ] from measurements of delayed electroluminescence, fi rst observed by Sinha et al. [ 10 ] Secondary singlet states generated by TF thus give rise to total singlet yields which exceed the classical spin statistical limit of 0.25 in both polymer and small molecule OLED devices, implying that charge recombination does not have to violate spin statistics in order to exceed an internal quantum effi ciency of 25%. [ 18 ] In order to understand the total secondary singlet production mechanism we set out to measure directly the effi ciency that singlets are produced from triplets in PSBF. We subsequently discovered that two triplet channels are operative. As well as the normal channel of TF, a second channel is present. When triplets annihilate to generate the upper triplet excited state, T N , these upper triplet states can also generate singlet states by decaying via a charge transfer state. The impact this mechanism has on the overall effi ciency of light generation is profound and explains the high effi ciency of many polymer and small molecule fl uorescent OLED emitters.

Results
To measure TF yield in fi lms directly, we have developed a method using nanosecond gated time resolved spectroscopy similar to that which we use to study triplet exciton dynamics in polymer fi lms, see refs. [ 4 ] and [ 6 ] for details.
The key element in these measurements is the ability to capture both prompt and delayed fl uorescence simultaneously (one curve) as shown in Figure 1 . We ascribe the decay up to ≈20 ns to prompt fl uorescence (PF). This is confi rmed by TCSPC measurements on the same fi lms giving an average fl uorescence lifetime of PSBF of ≈3 ns. [ 19,20 ] More detailed analysis reveals a triexpontential decay with two major emitting species, the initially photogenerated local exciton, 1 LE (≈1 ns), and a slightly lower energy 1 CT state (≈5 ns). [ 21 ] The decay tail appearing from ≈200 ns to 1 s as a power law is assigned to DF. This DF appears in PSBF as a result of TF, as clearly shown in Figure 2 a-the DF intensity follows predominantly a quadratic dependence with increasing laser pulse intensity.
Given that PF follows an exponential decay law and DF a power law at 20 K, we can fi t the luminescence decay curve using the equation More than one exponential term is needed, as PF decay in PSBF fi lms at 20 K is not mono exponential as shown in previous detailed fl uorescence lifetime studies. [ 19,20 ] Integration of the exponential terms gives the PF fraction of the total emission and integration of the power law term gives the DF contributions, ϕ DF of the total sample emission as in Equation ( 3) , see the Supporting Information for details.
Using Equation ( 3) , we determined ϕ DF at 20 K to be 0.33 ± 0.08 and at room temperature to be 0.14 ± 0.04, see Supporting Information for details. At room temperature, and at intermediate times (≈20-200 ns, Figure 1 b) the DF intensity makes a larger contribution in comparison with the DF at 20 K over the same time period, indicating an increase in the thermally activated DF process. However at ≈800 ns the DF undergoes a rapid intensity roll-off and overall DF efficiency decreases substantially. This DF intensity roll-off is due   ( 2) . The curve was obtained by combining decays recorded using nanosecond gated time resolved spectroscopy (from 1 ns to 1 s, at 15 µJ per pulse excitation) and singlet photon counting techniques (from 3 ps to 10 ns, at <1 nJ per pulse excitation). b) Emission decay of PSBF spincoated fi lm recorded at 20 K (black circles) and 296 K (red circles).
to a change from dispersive to nondispersive triplet exciton migration and has been observed previously in both small [ 6 ] and polymer (polyfl uorene) [ 3,4 ] fi lms. At room temperature, nonradiative decay becomes much stronger (in polymers) and so the NR quenching of the 3LE triplet states competes with TF reducing the overall amount of singlets generated by TF and so the overall DF yield reduces. Turning to our previous PSBF OLED device data with measured singlet yield at 20 K of 0.44 ± 0.04. [ 16 ] Taking ϕ DF = 0.33 at 20 K, and a spin independent charge recombination triplet yield of 75%, then the contribution of DF to the total singlet yield would be 0.33 × 0.75 = 0.25 ± 0.06. 0.25 ± 0.06 added to the directly created singlet excitons (with 0.25 yield) gives a total singlet production yield of 0.50 ± 0.06 which is within error limits is in excellent agreement with the measured 0.44 ± 0.04. The agreement is striking especially bearing in mind that very different experimental methods were used to determine them.
However, as pointed out above, for PSBF it is not possible to achieve ϕ DF > 0.2 from TF alone. Clearly, in PSBF 2 E T > E Tn ( E T = 2.22 eV, E Tn = 3.77 eV) [ 16 ] so that the maximum ϕ DF can only be 0.2. As our measurements require that ϕ DF > 0.2 we conclude that there must be a further contribution to DF. One possible process could be upper excited state reverse intersystem crossing from the T N state (uRISC), i.e., from T N → S 1 . This process would be in competition with internal conversion of T N to lower energy triplet states. However, it has never before been observed in luminescent polymers, but T N → S N transitions have been reported in a few organic molecules. [23][24][25] If uRISC where to be this source of the extra singlets, then it would require a contribution of 0.13 to the total DF emission. This requires an overall triplet to singlet yield (including T N → S 1 uRISC) of 0.33, that is 80% of triplets annihilate via the triplet channel producing T N states with 0.4 yield (two T 1 states required to produce one T N state), and therefore, 0.13/0.4 = 0.33 triplets need to decay by uRISK to generate the required DF singlets. There are reports of uRISC yields of 0.17 in dibromoanthracene, [ 26 ] ≈0.70 in cyanine dyes, [ 27 ] 0.80 in rose bengal dye [ 23 ] and 0.90 in tetraphenylphorphyrin. [ 23 ] However, we do not believe this to be the case here from the dynamics of the process we subsequently measure.
Careful analysis of the DF spectra, Figure 3 , clearly shows that the delayed emission is complex and evolves with time. From the initial S 1 1 (π, π*) decay, i.e., vibronically resolved emission measured in the fi rst nanosecond, we observe a spectral red shift with time and loss of vibronic resolution. By 50 ns, this red-shifted unstructured emission dominates. We have previously shown that this arises from decay of intramolecular CT states on the PSBF chain. The PSBF 1 CT state is ≈0.1 eV lower in energy than the S 1 . [ 21,22 ] After 200 ns, and on into the microsecond regime we again observed the re-emergence of structured, delayed S 1 1 (π, π*), ascribed to TF. As has recently been shown, triplet states can be effi ciently harvested if they couple to energetically close lying triplet states, yielding thermally activated delayed fl uorescence (TADF). A very small 1 CT-3 CT gap occurs in charge transfer states in spiro materials because of the orthogonal nature of highest occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) states, where negligible orbital overlap yields very small electron exchange energy. [ 28 ] This mechanism can be 100% effi cient with a small enough CT splitting [ 29 ] and in OLEDs produces very effi cient devices through this TADF triplet harvesting mechanism. [ 30,31 ] Therefore, in PSBF at intermediate times, 1 CT emission arising from both initially created 1 CT states (via electron transfer from the excitonic S 1 excited state) and TADF could be expected. The back electron transfer rate 3 CT → 3 LE to the very low lying PSBF 3 (π, π*) 3 LE state is typically relatively slow (≈5 × 10 6 s −1 ) [ 23 ] which will allow 3 CT harvesting to compete with this IC channel, giving rise to the observed 1 CT emission at intermediate times but only TF at longer times. The increase observed in the PSBF CT emission contribution at room temperature compared to 20 K [ 21 ]  dependence on excitation energy from PSBF fi lm recorded at 20 K. The slope of the DF power dependence being <2 indicates mixed linear and quandratic contributions to the total decay. b) Time evolution of emission from the PSBF spincoated fi lm at 20 K. Initial vibronically resolved S 1 1 (π, π*) emission is observed in the fi rst nanosecond (black trace), spectral diffusion of singlet excitons red shifts this S 1 emission over ≈10 ns (red trace) and concomitant with the appearance of increased red edge emission. By 50 ns, this red unstructured emission dominates ascribed to intramolecular CT states. [ 22 ] After 200 ns, and on into the microsecond regime delayed S 1 1 (π, π*) again observed ascribed to TF. The CT state is ≈0.1 eV lower in energy than the S 1 .
indicates that the TADF recycling [ 28 ] of 3 CT to 1 CT is active in the PSBF. Because TF does not involve any electron transfer, TF must regenerate 1 LE not 1 CT singlet states.
To further explore how the DF from the CT state arises, 0.01% PSBF matrix isolated in zeonex fi lms where studied at 20 K using a two pulse pump method (Figure 3 ). At this low polymer concentration, interchain TTA should be prevented and with a triplet yield of ϕ T = 0.12, [ 20 ] at low excitation intensities intrachain TTA should also be very low. Films were initially excited at 3.68 eV (S 0 to S 1 ) and the luminescence decay recorded (Figure 3 a full squares). 177 ns after the initial (3.68 eV) excitation we excite with a second 1.96 eV pulse (Figure 3 a empty triangles). 1.96 eV is not absorbed by the ground state [ 16 ] and neither S 1 states given that the lifetime of S 1 is ≈1 ns, [ 19 ] but rather excites the long lived T 1 excitons to the higher T N level, through an optically allowed transition. [ 20 ] It is evident from the data in Figure 3 a that after excitation with the 1.96 eV pulse, the DF intensity increases, by as much as 40%, Figure 3 b. Comparable results have also been obtained using 1.55 eV pulses which are resonant with the allowed T 1 to T N transition, onset at 1.35 eV. [ 20 ] Further, we were able to recorded time resolved emission spectra from 178 to 208 ns (relative to the initial 3.68 eV excitation) with three types of excitation. First, we excited with 1.96 eV (30-60 µJ per pulse) only to ascertain that there is no nonlinear two photon induced fl uorescence (Figure 3 b). We then excited with 3.68 eV (60-100 µJ per pulse) only to measure the base line delayed emission signal. Finally we excited with 3.68 and 1.96 eV at 177 ns delay to induce T 1 -T N transitions. In the latter case we observed the intensity increase of ≈40% which indicates a relatively effi cient increase in DF. We estimate from Figure 3 a that the lifetime of this induced delayed emission to be, ≈5-10 ns, which is considerably longer than the 1 ns 1.96 eV pump pulse and much longer than the expected T N state lifetime. This we believe rules out uRISK. Further, we clearly observe an increase in total DF, simply by exciting the T 1 states to the T N states, but the second excitation beam does not increase the total number of excited states, it only perturbs the triplet population (excites them). This second pump beam cannot induce extra TF, therefore a new physical decay channel must be accessed from the T N population, i.e., a further radiative decay channel becomes operative from the T N state.

Discussion
Given the lifetime and spectral shape of this induced delayed fl uorescence, we ascribe it to radiative decay of the 1 CT state. But, how does this state become populated from the T N ? The T N state will predominantly decay to the next lowest triplet state, not the lowest lying 3 (π, π*) 3 LE level but the (much closer in energy) 3 CT triplet state, thus a kinetic competition will take place. [ 32 ] A relatively long lifetime (>>100 ps) of the upper excited T N state is not unreasonable given that we know for 'polyfl uorene' backbone structures the T N state is orthogonal to the T 1 state [ 33 ] and further T N , is 1.55 eV above T 1 . Given that the highest energy vibronic mode for PSBF is the 180 meV, C=C bond stretch, T N,0 lies at least 8 vibrational quanta above T 1,0 giving rise to slow decay through the phonon bottle neck this causes. [ 34 ] Indeed such a large T N -T 1 energy gap can give rise to T N fl uorescence in the case where T N lies below S 1 , showing that upper excited state lifetimes can be relatively long, [ 35 ] at least on a par with S 1 which readily allows S 1 → 1 CT electron transfer. It can also be seen that the T N state is ≈0.7 eV above 3 CT giving a strong driving force for the electron transfer step.
Thus, a proportion of the T N states created by TTA will decay via electron transfer to the 3 CT state and via TADF generate the extra 1 CT DF signal we observe after the second pump pulse. The 1 CT-3 CT gap in PSBF can be estimated from the previous results of King et al., who observed a temperature activation energy for DF in PSBF fi lms of ≈14 meV. [ 36 ] This was attributed to the gap between the 1 CT state and the 1 (π, π*) S 1 state, however, this is too small for the gap measured spectroscopically, and in light of our recent fi ndings on TADF, we can now understand this data as being a measure of the exchange energy in the CT manifold. [ 29 ] The small exchange energy is comparable  with other CT systems where there is near orthogonally between HOMO and LUMO level and is the case for the PSBF where we have previously shown that the charge transfer occurs across the orthogonal spiro unit, mediate by spiro conjugation and is in line with the results of Mehes et al. on spiro charge transfer molecules. [ 28 ] This mechanism could readily give rise to an extra 0.13 contribution to the total DF. This then gives a new mechanism for generating DF from T N states. As TTA produces a predominance of T N , then the presence of intermediate CT states that give effi cient TADF gives effi cient singlet production by this induced TADF mechanism. These competing decay channels are shown schematically in Figure 4 .
On refl ection, this mechanism is the only way in which photo excitation of T N states (from T 1 states) can give rise to extra DF as observed; if they simply decay back to T 1 we would see no increase in DF after the second pulse. uRISK seems implausible as it would have been seem in many other, non-CT containing luminescent polymers before, given the effi ciencies required here and the lifetime of the T N state we measure here.

Conclusions
We present the fi rst direct evaluation of the total singlet exciton production yield from triplet excitons in a luminescent polymer, PSBF. We obtain a lower limit of ϕ DF = 0.33 for total delayed fl uorescence, which is clearly much greater than the "classical" TTA singlet TF yield of 0.055 and also above the expected 0.2 yield given that in the case of PSBF, only the quintuplet TTA channel is energetically unattainable. These results indicate that a further process must contribute to the effi ciency of delayed fl uorescence, which we identify as induced TADF via decay of T N states to the 3 CT state in competition with decay to the 3 (π, π*) state. From our combined TF and induced TADF yield of 0.33 for PSBF, we achieve agreement with the total singlet production yield of 0.44 ± 0.04 determined for PSBF devices without breaking spin statistics in both recombination and TTA.
Our results show that triplet annihilation can generate singlets in two ways. First, through triplet fusion, where the encounter complex gains suffi cient singlet wavefunction character that the two triplets become a singlet state. The second, distinct process requiring an intermediate charge transfer state. The normally wasted T N state formed via the TTA triplet channel can decay by electron transfer, to 3 CT which converts to 1 CT through the TADF process. This mechanism then also produces photons through the radiative decay of the 1 CT state.
More generally, many polymers either have intrachain CT states as is the case of PSBF, but also interchain CT states, or polaron pairs are also well known [ 37 ] which are effectively exciplex states that also give rise to effi cient TADF. [ 38,39 ] Thus, the T N decay channel via the 3 CT state can be an intrinsic mechanism for harvesting triplet states into singlets. [ 29 ] Also, many CT small molecules which give mixed TADF and TTA, when the lowest lying 3 LE state lies below the CT manifold still give good OLED performance. These molecules give a high triplet population leading to strong TTA and concomitant T N population which could again be harvested by this induced TADF mechanism giving enhance singlet yields, but not 100% singlet production that true TADF systems. [ 32,40 ]

Experimental Section
Solutions of PSBF where made in toluene at 10 mg mL −1 concentrations for spin coating. Guest host fi lms in zeonex were prepared by drop casting from 0.01% PSBF in zeonex by weight onto sapphire substrates. Absorption and emission spectra were collected using a UV-3600 double beam spectrophotometer (Shimadzu), and a Fluorolog fl uorescence spectrometer (Jobin Yvon).
Phosphorescence, prompt fl uorescence (PF), and delayed emission (DF) spectra and decays were recorded using nanosecond gated  Schematic energy level diagram for PSBF constructed from fl uorescence, phosphorescence, and photoinduced absorption measurements and delayed fl uorescence, see ref. [ 16 ] . From this Jablowski scheme the various decay channels for initial and photoinduced excited states can be seen in PSBF. The S 1 state will be quenched by the 1 CT state by electron transfer, 1 CT can interconvert slowly by hyperfi ne coupling to 3 CT. There is then a competition between TADF back to 1 CT and 3 CT quenching to the lowest energy 3 LE state of the system. luminescence and lifetime measurements (from 400 ps to 1 s) using either a high energy pulsed Nd:YAG laser emitting at 355 nm (EKSPLA) or a N2 laser emitting at 337 nm. Emission was focused onto a spectrograph and detected on a sensitive gated iCCD camera (Stanford Computer Optics) having sub-nanosecond resolution. PF/DF time resolved measurements were performed by exponentially increasing gate and delay times. Two pump measurements were made using the Nd:YAG laser as the master trigger source and via a delay generator a second diode laser was triggered after an appropriate time delay. Subsequent delayed emission was measured with the iCCD triggered by the master pulse with appropriate delay. Measurements were made at 20 K with samples in a displex cryostat.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.