Fast Spin-Flip Enables Efficient and Stable Organic Electroluminescence from Charge-Transfer States
Lin-Song Cui1,8,*, Alexander J. Gillett1,8, Shou-Feng Zhang2,3,8, Hao Ye4, Yuan Liu5, Xian-Kai Chen3,*, Ze-Sen Lin4, Emrys W. Evans1, William K. Myers6, Tanya K. Ronson7, Hajime Nakanotani4, Sebastian Reineke5, Jean-Luc Bredas3, Chihaya Adachi 4,*, Richard H. Friend1,*

1Cavendish Laboratory, University of Cambridge, JJ Thomson Avenue, Cambridge, CB3 0HE, United Kingdom.
2Department of Electronic Engineering, Guangxi University of Science and Technology, Liuzhou 545006, China
3School of Chemistry and Biochemistry, Center for Organic Photonics and Electronics, Georgia Institute of Technology, Atlanta, GA, 30332-0400, USA
4Center for Organic Photonics and Electronics Research (OPERA), Kyushu University, 744 Motooka, Nishi, Fukuoka 819-0395, Japan.
5Dresden Integrated Center for Applied Physics and Photonic Materials (IAPP), Technische Universität Dresden, Dresden, 01069, Germany.
6Centre for Advanced Electron Spin Resonance (CAESR), University of Oxford, South Parks Road, Oxford, OX1 3QR, UK.
7Department of Chemistry, University of Cambridge, Lensfield Road, Cambridge, CB1EW, United Kingdom.
8These authors contributed equally to this work: Lin-Song Cui, Alexander J. Gillett, Shou-Feng Zhang 
*Corresponding author. Lin-Song Cui: E-mail: lc724@cam.ac.uk; Xian-Kai Chen: E-mail: chenxiankai@email.arizona.edu; Chihaya Adachi: E-mail: adachi@cstf.kyushu-u.ac.jp; Richard H. Friend: E-mail: rhf10@cam.ac.uk.
A spin-flip from a triplet to a singlet excited state, that is, reverse intersystem crossing (RISC), is an attractive route for improving light emission in organic light-emitting diodes (OLEDs), as shown by devices using thermally activated delayed fluorescence (TADF). However, device stability and efficiency roll-off remain challenging issues that originate from a slow RISC rate (kRISC). Here, we report a TADF molecule with multiple donor units that form charge-resonance-type hybrid triplet states leading to a small singlet-triplet energy splitting, large spin-orbit couplings, and a dense manifold of triplet states energetically close to the singlets. The kRISC in our TADF molecule is as fast as 1.5×107 s−1, a value some two orders of magnitude higher than typical TADF emitters. OLEDs based on this molecule exhibit good stability (estimated T90 about 600 hours for 1,000 cd m−2), high maximum external quantum efficiency (＞ 29.3%), and low efficiency roll-off (＜ 2.3% at 1,000 cd m−2).










Main
The spin-triplet excited states of organic π-conjugated molecules play an essential role in organic optoelectronic devices, such as organic light-emitting diodes (OLEDs), organic solar cells, and organic semiconductor lasers.1,2,3,4 Efficiently harvesting triplet excitons is critical in OLEDs, as spin-statistical charge recombination leads to the formation of triplet to singlet excitons in a 3:1 ratio.5,6,7,8,9,10,11,12 Recent developments have shown that purely organic emitters exploiting thermally activated delayed fluorescence (TADF) in OLEDs can achieve nearly 100% internal quantum efficiencies (IQE).13,14,15, However, severe issues still affect TADF-based OLEDs: for example, operational stability and efficiency roll-off remain unsatisfactory, which largely limits their potential for commercialization.16,17,18,19 
	In TADF emitters, a fast reverse intersystem crossing (RISC) process from the triplet to the singlet manifold is crucial for the efficient utilisation of triplet excitons. Previously, RISC was considered to be a process involving simply the lowest triplet (T1) and singlet (S1) excited state; the degree () of mixing of these states can be expressed in the framework of first-order perturbation theory as , where  is the spin-orbit coupling (SOC) between the S1 and T1 states and  is their energy gap.20 This relationship suggests that small  and large  values are required for efficient RISC. Over the past decade, small (< 0.1 eV) values have been reported in a broad series of simple donor-acceptor (D-A)-type molecules; in such instances, the spatial separation between the highest occupied and lowest unoccupied molecular orbitals (HOMO and LUMO) leads to an extremely small electron-exchange energy (equal to  when the main electronic configuration describing both the S1 and T1 states simply corresponds to a HOMO-to-LUMO transition).21,22,23,24,25,26,27,28 In such D-A-type molecules, however, the S1 and T1 states often have a dominant charge-transfer (CT) -exciton character, which results in a vanishingly small SOC (< 0.1 cm-1) between the two CT states.29 The consequence of this negligible SOC is that RISC from a CT-type T1 to a CT-type S1 is inefficient even when  is small, in most reported TADF emitters, RISC rate constants (kRISC) are typically < 106 s−1.30,31,32,33 A slow RISC process results in long triplet exciton lifetimes, leading to bimolecular exciton annihilation processes that are responsible for the severe efficiency roll-offs and material degradation occurring in TADF-based OLEDs.34,35,36,37,38 Thus, designing new TADF molecules with fast RISC rates is of utmost importance to improve OLED stability and efficiency.
	Our earlier theoretical investigations on Cz-TRZ (9-(4-(4,6-diphenyl-1,3,5-triazine-2-yl)phenyl)-9H-carbazole) and its derivatives indicated that the T1-state wavefunctions in these molecules are confined to the carbazole-phenylene-triazine fragment.39 It was predicted that the introduction of additional carbazole donors (which have a high-energy triplet state) would decrease the S1‐state energies but would have little impact on the T1‐state energies and wavefunctions: the consequence is not only a reduced  value but also S1 and T1 states with different excitation characters. Additionally, we have shown that in D-A-type TADF molecules, the introduction of multiple donor moieties results in the formation of charge-resonance-type hybrid triplet states; this leads to the appearance of a dense manifold of triplet states. This opens up additional RISC transition channels from higher-lying triplet (for example, T2) states to S1 (via a second-order spin-vibronic mechanism), which can facilitate the spin-flip processes.40,41 
	Based on these earlier findings,39,40,41 we have designed and synthesized a TADF emitter consisting of five carbazole (Cz) donors and one triazine (TRZ) acceptor, namely 5Cz-TRZ (9,9',9'',9''',9''''-(6-(4,6-diphenyl-1,3,5-triazine-2-yl)benzene-1,2,3,4,5-pentayl)pentakis(9H-carbazole)) (see Fig. 1a). For comparison, we have also investigated three other molecules consisting of the same D and A moieties with 5Cz-TRZ: mCz-TRZ (9-(4-(4,6-diphenyl-1,3,5-triazin-2-yl)phenyl)-3,6-dimethyl-9H-carbazole), TmCz-TRZ (9,9′,9″-(5-(4,6-diphenyl-1,3,5-triazine-2-yl)benzene-1,2,3-triyl) tris(3,6-dimethyl-9H-carbazole)), and DACT-II (9-[4-(4,6-diphenyl-1,3,5-triazine-2-yl)phenyl]-N,N,N',N'-tetraphenyl-9H-carbazole- 3,6-diamine)) (see Fig. 1a). Here, we demonstrate that in addition to the presence of a dense triplet-state manifold, a small  and a large spin-orbit coupling are also simultaneously achieved in 5Cz-TRZ. As a result, 5Cz-TRZ shows an ultra-fast kRISC, ~ 1.5 × 107 s−1: a value two orders magnitude faster than those typically found in the most other TADF molecules.17,19,25 Such a fast RISC process reduces the triplet exciton concentration and the number of bimolecular annihilation events in the emission layer, which in turn increases the lifetimes of the TADF-based OLEDs more than tenfold and simultaneously overcomes the efficiency roll-off issue. Furthermore, full-colour hyperfluorescent OLEDs exploiting 5Cz-TRZ as a sensitizer also exhibit very high external quantum efficiency (EQE) values of over 24.9%. Our achievement of a fast kRISC (> 107 s−1) in purely organic molecules opens the door to highly efficient and stable TADF-based OLEDs.
Results and discussion
Computational Results
As shown by the Natural Transition Orbital (NTO) analysis in Fig. 1b, the S1 state of 5Cz-TRZ shows a pronounced CT-excitation character; the hole and electron wavefunctions are localized at the ortho- and meta-carbazole donors and the central (para-carbazole)-phenylene-triazine, respectively, with a small spatial overlap (Oh/e) of ~ 0.3. In contrast, the T1 state exhibits a hybrid character as it consists of mixed local and CT excitations with a large Oh/e of ~ 0.7. (see the “Computational Details” section for more details on the calculations). The clear difference in the excitation characters of the S1 and T1 states results in a high  value of ~ 0.4 cm-1. Intriguingly, the calculations predict a  value as small as ~ 0.02 eV, which is usually obtained in case where S1 and T1 both have CT character. The central (para-carbazole)-phenylene-triazine present in 5Cz-TRZ has a low-energy triplet state at ~ 2.92 eV due to the large electronic coupling between the para-carbazole and phenylene-triazine fragments (see Fig. 1c). The additional ortho- and meta-carbazole donors in 5Cz-TRZ possess higher-energy triplet states (~ 3.61 eV), which have little effect on the T1-state energy (~ 2.95 eV at the T1-state equilibrium geometry) of the whole 5Cz-TRZ molecule (Fig. 1c), a result consistent with our earlier calculations.39 In contrast, the additional donors effectively stabilize the S1-state energy of 5Cz-TRZ due to electronic delocalization and polarization effects. The consequence of these factors is the realization of both a large  and a small  value in 5Cz-TRZ. In addition, as demonstrated in our earlier investigations,41 the Cz donors connected to the phenylene-triazine acceptor leads to a dense manifold of triplet states (with a small energy gap between T1 and T2, ~ 0.24 eV), due to the formation of charge-resonance-type hybrid triplet states (see Fig. 1b). Another feature is that the  (SOC between the T2 and S1 states, ~ 1.3 cm-1) is significantly larger than , a consequence of the differences in the spatial distributions of the T1 and T2 hole wavefunctions (Fig. 1b). In the framework of Marcus electron-transfer theory, the rate constants from the T1 and T2 to S1 ( and ) are evaluated to be 4.1 × 106 and 1.4 × 109 s-1, respectively. Importantly, the small energy gap between the T1 and T2 states opens a spin-flip channel from T2 to S1 in addition to the T1-to-S1 channel, which implies that the RISC rate in 5Cz-TRZ must be larger than 4.1 × 106 s-1.
	The electronic structure and related properties of TADF molecules usually depend on the selection of donor and acceptor moieties. For example, the  (~ 0.17 eV) of 5CzBN (2,3,4,5,6-penta-(9H-carbazol-9-yl)benzonitrile) with cyano group as acceptor is much larger than that of 5Cz-TRZ.41 As a result, the kRISC of 5Cz-TRZ should be much faster than that of 5CzBN, although the dense manifold of triplet states also appears in 5CzBN.
	It is clear from the above discussion that our molecular-design strategy for 5Cz-TRZ is based on two ideas. First, while the T1 exciton is confined in the (para-carbazole)-phenylene-triazine segment and has a low triplet energy, the introduction of additional Cz donors in the ortho- and meta-positions significantly decreases the S1-state energy, which ultimately reduces the  value. Additionally, since we have confined the T1 state, it now possesses a local-excitation (LE) character, whilst the S1 state has a CT character; this results in strong spin-orbit coupling between the LE-dominated T1 state and the CT-dominated S1 state. Second, the introduction of the additional Cz donors in the ortho- and meta-positions leads to another positive characteristic: a dense manifold of triplet states due to the formation of charge-resonance-type hybrid triplet states. This opens up the T2-to-S1 RISC channel, in addition to the T1-to-S1 channel.
The data for the other three molecules (mCz-TRZ, TmCz-TRZ, and DACT-II) that we investigated to provide a point of comparison are given in Supplementary Fig. S1 and Table S1. In these molecules, the absence of multiple donor groups that could form charge-resonance-type hybrid triplet states results in RISC rates that are 1 to 4 orders of magnitude slower than in 5Cz-TRZ; this hinders their electroluminescent performance and device stability.
Photophysical Properties
The photophysical properties of 5Cz-TRZ were analysed using ultraviolet‐visible (UV/Vis) and photoluminescence (PL) spectroscopies. As shown in Fig. 2a, the absorption spectrum of 5Cz-TRZ is formed from the combination of the Cz donor and TRZ acceptor absorptions, with an additional weak absorption band between 380 – 420 nm that is attributed to the intramolecular CT transition from the Cz donors to the TRZ acceptor. With increasing solvent polarity, the emission spectrum of 5Cz-TRZ broadens and red‐shifts from 470 nm (cyclohexane) to 550 nm (dimethyl formamide), consistent with the strong CT character of the first singlet excited state. Fig. 2b displays the PL and phosphorescence (Phos) spectrum of 5Cz-TRZ in a frozen toluene matrix at 77 K. As expected, the Phos spectrum of 5Cz-TRZ is well resolved and shows a characteristic vibrational structure, indicating that the T1 state must have significant local-excitation character. In addition, the transient electron spin resonance (TrESR) measurements for 5Cz-TRZ reveal relatively broad, polarized spin resonance signals that are indicative of T1 states formed following spin-orbit-coupling-mediated intersystem crossing42 (see Fig. 2c and Supplementary Table S2 for more detailed discussion of the data). The S1 and T1 state energies of 5Cz-TRZ are estimated to be 2.85 and 2.79 eV from the onset of low-temperature PL and the first peak of Phos spectra, respectively, resulting in a value of 0.06 eV . These experimental data thus point both to a small  and to S1 and T1 states with different excitation characters, consistent with our theoretical results.
	To evaluate the delayed fluorescence behaviour, the transient PL characteristics were analysed in degassed solutions at room temperature (Fig. 2d). 5Cz-TRZ exhibits clear prompt and delayed fluorescence components with a delayed exciton lifetime as short as 0.8 μs in dimethyl formamide. The kRISC value was estimated from the PL quantum efficiencies (PLQE) and the exciton lifetime of the prompt and delayed components (a detailed description of the calculations is provided in the Supplementary Information). 5Cz-TRZ shows an extremely fast kRISC of ~ 1.5 × 107 s−1 in toluene, which is some two orders of magnitude higher than those of conventional TADF molecules such as TmCz-TRZ and DACT-II (Supplementary Fig. S6 and Table S3). We also investigated the excited-state dynamics of 5Cz-TRZ in toluene using pump-probe transient absorption (TA) spectroscopy (Fig. 2e). The picosecond TA measurement spectra exhibit broad photoinduced absorption (PIA) signals over the entire spectral region (500 – 1025 nm); the signals undergo rapid spectral shifts within the first 10 ps, which can be assigned to structural and solvent reorganization processes43 (more detailed discussions of the dynamical processes can be found in the Supplementary Information). Nanosecond TA was also performed to investigate the timescales of intersystem crossing (ISC) and RISC, and to characterize triplet excited-state features. As illustrated by Figs. 2f, g, 5Cz-TRZ undergoes ISC from the 1CT state to the triplet manifold with a time constant of 4.5 ns, shorter than the ISC timescales of TmCz-TRZ and DACT-II (Supplementary Figs. S7 and 8). The combination of the nearly degenerate singlet and triplet states and the large spin-orbit couplings are responsible for the rapid ISC process observed in 5Cz-TRZ. Temperature-dependent time-resolved PL decay data for 15wt%-doped films of 5Cz-TRZ in 3,3′-di(9H-carbazol-9-yl)-1,1′-biphenyl (mCBP) were also collected to investigate the excited-state decay dynamics under device-relevant conditions (Fig. 2h). The RISC rates are observed to increase with increasing temperature from 200 to 300 K (Fig. 2i), reaching 1.37 × 107 s−1 at 300 K. The  value is estimated to be 0.03 eV from the Arrhenius plots of the kRISC values in 5Cz-TRZ (Supplementary Fig. S5).
Device characterization
Given the promising photophysical properties of 5Cz-TRZ, which include a fast kRISC and a high PLQE, its electroluminescence (EL) performance was evaluated. The optimized device configuration shown in Fig. 3 has the following architecture: ITO/HAT-CN (10 nm)/α-NPD (30 nm)/Tris-PCz (10 nm)/mCBP (6 nm)/15 wt% TADF emitter: mCBP (20 nm)/CF3-TRZ (10 nm)/30 wt% Liq: BPPB (45 nm)/Liq (2 nm)/Al (120 nm), where HAT-CN, α-NPD, Tris-PCz, CF3-TRZ, BPPB and Liq are 1,4,5,8,9,11-hexaazatriphenylene hexacarbonitrile, N,N′-diphenyl-N,N′-bis(1-naphthyl)−1,10-biphenyl-4,4′-diamine, 9,9′,9″-triphenyl-9H,9′H,9″H−3,3′:6′3″-tercarbazole, 6,6'-((perfluoropropane-2,2-diyl)bis(4,1-phenylene))bis(2,4-diphenyl-1,3,5-triazine), 1,3-bis(9-phenyl-1,10-phenanthrolin-2-yl)benzene and 8-hydroxyquinoline lithium, respectively; 15wt% of 5Cz-TRZ doped in mCBP (device A) served as the emitting layer (EML). For comparison, the control devices (devices B and C) were fabricated with the emitters TmCz-TRZ and DACT-II. The materials used and the device structures are given in Supplementary Fig. S10.
	Devices A and B show sky-blue emission with EL peaks at 486 and 496 nm, respectively, whereas device C exhibits pure green emission with the emission peak at 530 nm (Fig. 4b). As illustrated in Fig. 4a and Supplementary Fig. S11, high and stable efficiencies are achieved with device A. The maximum EQE of device A is 29.3%, a value much higher than those of device B (19.7%) and device C (24.3%). It is especially noteworthy that device A, which has 5Cz-TRZ as dopant, exhibits an extremely low efficiency roll‐off; the efficiencies still remain 28.6% at 1,000 cd m−2 and 27.0% at 5,000 cd m−2, corresponding to a 2.3% and 7.8% decrease in EQE at these brightness points. In contrast, devices B and C display severe efficiency roll-off effects, with much lower EQEs of 8.6% and 16.2% at 5,000 cd m−2; this corresponds to 56.3% and 33.3% decreases in EQE, respectively. 
	The transient EL decay characteristics at different luminance levels were used to further confirm the low efficiency roll‐off of the 5Cz-TRZ-based device (Fig. 4c), with very little reduction in the excition lifetime observed at high luminance levels. With this behaviour confirmed, we then analysed the operational stability of the devices (Fig. 4d), showing that the T90 (time to reach 90% of initial luminance) values of devices A, B, and C are 35.4, 3.2, and 5.1 h for an initial brightness (L0) of 5,000 cd m−2. Device A thus exhibited by far the longest operational stability, with device lifetime over ten times longer than that of device B. The T90 of device A is predicted to be ~ 600 h for an initial brightness of 1,000 cd m−2 according to the formula LT(L1) = LT(L0) × (L0/L1)n, where LT is lifetime and L1 is the desired initial luminance of 1,000 cd m−2 (n=1.75, which is obtained by fitting T90 versus L0 in Supplementary Fig. S14).44 This device lifetime is longer than those previously reported TADF-based OLEDs (Supplementary Table S5). 
	Hole-only and electron-only devices based on the structures of devices A, B, and C were also fabricated to characterize the carrier-transport properties. All devices exhibited similar properties (Supplementary Fig. S15), which implies that charge-carrier transport is not the main factor determining the differences in efficiency roll-off and operational stability. Thus, the reduced efficiency roll-off and improved operational stability of device A is primarily determined by the fast kRISC; this limits the amount of bimolecular exciton annihilation processes taking place in the devices. 
	To better understand the factors behind the high EQE of device A, we investigated the emitting-dipole orientations in these devices using variable-angle PL measurements. As shown in Supplementary Fig. S16, the orientation factor (θ) of 5Cz-TRZ is estimated to be 0.17 (we recall that θ = 0 for completely horizontal alignment and 0.33 for isotropic distribution).45 Using this value, the optical simulation results indicate that device A can achieve a maximum EQE of 30.9%, which is consistent with our experimental data. 
	The fast kRISC of 5Cz-TRZ inspired us to further investigate the performance of hyperfluorescent OLEDs with 5Cz-TRZ as a sensitizer (Supplementary Fig. S17). 2,5,8,11-tetra-tert-butylperylene (TBPe), 5,12-Dibutyl-1,3,8,10-tetramethylquinacridone (TMDQA), 2,8-di-tert-butyl-5,11-bis(4-tert-butylphenyl)-6,12-diphenyltetracene (TBRb) and 4-(Dicyanomethylene)-2-tert-butyl-6-(1,1,7,7-tetramethyljulolidin-4-yl-vinyl)-4H-pyran (DCJTB) were selected as sky-blue, green, yellow, and red fluorescent emitter dopants, respectively. For the same device architecture as the TADF-emitter-based OLEDs, the hyperfluorescent OLEDs using 5Cz-TRZ as sensitizer show EQEs as high as 24.0% (device D; sky bule), 22.3% (device E; green), 24.9% (device F; yellow) and 23.5% (device G; red). The slightly lower EQEs and larger efficiency roll‐offs compared with the 5Cz-TRZ-only device are probably due to hole trapping by the fluorescent molecules, as the HOMO levels of the fluorescent materials are shallower than that of 5Cz-TRZ. However, the operational lifetimes of the OLEDs based on 5Cz-TRZ as the sensitizer were longer than that of the OLED with 5Cz-TRZ as the emitter dopant (see Fig. 4d). We do note that the EL spectrum of device D is different form its PL spectrum in the co-deposited film (Supplementary Fig. S18), which is probably induced by the microcavity effect of the device rather than an incomplete energy transfer.19 
	Finally, we constructed a white OLED (device H) by using 5Cz-TRZ as the sky-blue emitter, complemented by TBRb and DCJTB as yellow and red emitters, respectively. A high EQE of 21.8 % was achieved with Commission Internationale de I'Eclairage (CIE) coordinates of (0.43, 0.45) and colour rendering index (CRI) of 81 (Fig. 4b). 
Conclusion
In summary, we have designed a sky-blue metal-free organic emitter (5Cz-TRZ) with a RISC rate as fast as 1.5 × 107 s−1, some two orders of magnitude higher than the typical values for TADF emitters. The introduction of multiple carbazole donors in 5Cz-TRZ allows for the formation of charge-resonance-type hybrid triplet states; this simultaneously results in a small single-triplet energy splitting, large spin-orbit coupling, and dense triplet-state manifold, which all contribute to the fast RISC. In combination with the near-100% PLQE and a strong horizontal dipole orientation in the doped film, state-of-the-art OLED devices using 5Cz-TRZ as emitter show a maximum EQE of 29.3%, a low efficiency roll-off at high luminance (a mere 2.3% roll-off of the EQE at 1,000 cd m−2), and an estimated T90 value of ~ 600 h for an initial brightness of 1,000 cd m−2. This lifetime is tenfold longer than that of the reference device based on the TmCz-TRZ emitter, which has a slower kRISC. Additionally, full-colour hyperfluorescent OLEDs exploiting 5Cz-TRZ as a sensitizer reach high EQE values of over 22.0%. Thus, we consider that our new design strategy represents an important step towards achieving a fast kRISC in TADF molecules, paving the way to efficient and stable TADF and hyperfluorescent OLEDs.
References
1. Baldo, M. A. et al. Highly efficient phosphorescent emission from organic electroluminescent devices. Nature 395, 151–154 (1998). 
2. Einzinger, M. et al. Sensitization of silicon by singlet exciton fission in tetracene. Nature 571, 90–94 (2019). 
3. Samuel, I. D. W. & Turnbull, G. A. Organic semiconductor Lasers. Chem. Rev. 107, 1272–1295 (2007).
4. Kohler, A. & Bassler, H. Triplet states in organic semiconductors. Mater. Sci. Eng. R 66, 71–109 (2009).
5. Adachi, C., Baldo, M. A., Thompson, M. E. & Forrest, S. R. Nearly 100% internal phosphorescence efficiency in an organic light emitting device. J. Appl. Phys. 90, 5048–5051 (2001).
6. Ma, Y., Zhang, H., Shen, J. & Che, C. Electroluminescence from triplet metal-ligand charge-transfer excited state of transition metal complexes. Synth. Met. 94, 245–248, (1998).
7. Köhler, A. & Beljonne, D. The singlet-triplet exchange energy in conjugated polymers. Adv. Funct. Mater. 14, 11–18 (2004).
8. Kondakov, D. Y. Characterization of triplet-triplet annihilation in organic light-emitting diodes based on anthracene derivatives. J. Appl. Phys. 102, 114504 (2007).
9. Ly, K. T. et al. Near-infrared organic light-emitting diodes with very high external quantum efficiency and radiance. Nat. Photon. 11, 63–68 (2017).
10. Li, L. K. et al. Strategies towards rational design of gold(iii) complexes for high-performance organic light-emitting devices. Nat. Photon. 13, 185–191 (2019).
11. Hamze, R. et al. Eliminating nonradiative decay in Cu(I) emitters: >99% quantum efficiency and microsecond lifetime. Science 363, 601–606 (2019).
12. Ai, X. et al. Efficient radical-based light-emitting diodes with doublet emission. Nature 563, 536–540 (2018).
13. Uoyama, H., Goushi, K., Shizu, K., Nomura, H. & Adachi, C. Highly efficient organic light-emitting diodes from delayed fluorescence. Nature 492, 234–238 (2012).
14. Zhang, Q. et al. Efficient blue organic light-emitting diodes employing thermally activated delayed fluorescence. Nat. Photon. 8, 326–332 (2014).
15. Liu, Y., Li, C., Ren, Z., Yan, S. & Bryce, M. R. All-organic thermally activated delayed fluorescence materials for organic light-emitting diodes. Nat. Rev. Mater. 3, 18020 (2018).
16. Cui, L.-S. et al. Controlling synergistic oxidation processes for efficient and stable blue thermally activated delayed fluorescence devices. Adv. Mater. 28, 7620–7625 (2016).
17. Noda, H., Nakanotani, H. & Adachi, C. Excited state engineering for efficient reverse intersystem crossing. Sci. Adv. 4, eaao6910 (2018).
18. Cui, L.-S. et al. Long-lived efficient delayed fluorescence organic light-emitting diodes using n-type hosts. Nat. Commun. 8, 2250 (2017).
19. Chan, C.-Y. et al. Efficient and stable sky-blue delayed fluorescence organic light-emitting diodes with CIEy below 0.4. Nat. Commun. 9, 5036 (2018).
20. Turro, N. J., Ramamurthy, V. & Scaiano, J. C. in Principle of Molecular Photochemistry: An Introduction Ch. 3, 113–118 (University Science Books, 2009).
21. Kondo1, Y. et al. Narrowband deep-blue organic light-emitting diode featuring an organoboron-based emitter. Nat. Photon. 13, 678–682 (2019).
22. Kotadiya, N., Blom, P. W. M. & Wetzelaer, G.-J. A. H. Efficient and stable single-layer organic light-emitting diodes based on thermally activated delayed fluorescence. Nat. Photon. 13, 765–769 (2019).
23. Ahn, D. H. et al. Highly efficient blue thermally activated delayed fluorescence emitters based on symmetrical and rigid oxygen-bridged boron acceptors. Nat. Photon. 13, 540–546 (2019)
24. Wu, T.-L. et al. Diboron compound-based organic light-emitting diodes with high efficiency and reduced efficiency roll-off. Nat. Photon. 12, 235–240 (2018).
25. Lee, D. R. et al. Design strategy for 25% external quantum efficiency in green and blue thermally activated delayed fluorescent devices. Adv. Mater. 27, 5861–5867 (2015).
26. Wu, K. et al. De novo design of excited-state intramolecular proton transfer emitters via a thermally activated delayed fluorescence channel. J. Am. Chem. Soc. 140, 8877– 8886, (2018).
27. Lin, T.-A. et al. Sky-blue organic light emitting diode with 37% external quantum efficiency using thermally activated delayed fluorescence from spiroacridine-triazine hybrid. Adv. Mater. 28, 6976–6983 (2016).
28. Zhang, Y. et al. Multi‐resonance induced thermally activated delayed fluorophores for narrowband green OLEDs. Angew. Chem. Int. Ed. 131, 1–7 (2019).
29. El-Sayed, M. A. The radiationless processes involving change of multiplicity in the diazenes. J. Chem. Phys. 36, 573–574 (1962).
30. Olivier, Y. et al. Nature of the singlet and triplet excitations mediating thermally activated delayed fluorescence. Phys. Rev. Mater. 1, 75602 (2017).
31. Marian, C. M. Spin-orbit coupling and intersystem crossing in molecules. WIREs Comput. Mol. Sci. 2, 187–203, (2012).
32. Etherington, M. K., Gibson, J., Higginbotham, H. F., Penfold, T. J. & Monkman, A. P. Revealing the spin-vibronic coupling mechanism of thermally activated delayed fluorescence. Nat. Commun. 7, 13680 (2016).
33. Penfold, T. J., Gindensperger, E., Daniel, C. & Marian, C. M. Spin-vibronic mechanism for intersystem crossing. Chem. Rev. 118, 6975–7025 (2018).
34. Murawski, C., Leo, K. & Gather, M. C. Efficiency roll-off in organic light-emitting diodes. Adv. Mater. 25, 6801–6827 (2013).
35. Einzinger, M. et al. Shorter exciton lifetimes via an external heavy-atom effect: alleviating the effects of bimolecular processes in organic light-emitting diodes. Adv. Mater. 29, 1701987 (2017).
36. Reineke, S., Schwartz, G., Walzer, K. & Leo, K. Reduced efficiency roll-off in phosphorescent organic light emitting diodes by suppression of triplet-triplet annihilation. Appl. Phys. Lett. 91, 123508 (2007).
37. Lee, J. et al. Hot excited state management for long-lived blue phosphorescent organic light-emitting diodes. Nat. Commun. 8, 1–9 (2017).
38. Munkhbat, B., Wersäll, M., Baranov, D. G., Antosiewicz, T. J. & Shegai, T. Suppression of photo-oxidation of organic chromophores by strong coupling to plasmonic nanoantennas. Sci. Adv. 4, eaas9552 (2018).
39. Chen, X.-K. et al. A new design strategy for efficient thermally activated delayed fluorescence organic emitters: from twisted to planar structures. Adv. Mater. 29, 1702767 (2017).
40. Chen, X. K., Zhang, S. F., Fan, J. X. & Ren, A. M. Nature of highly efficient thermally activated delayed fluorescence in organic light-emitting diode emitters: nonadiabatic effect between excited states. J. Phys. Chem. C 119, 9728–9733 (2015).
41. Noda, H., Chen, X. K. et al. Critical role of intermediate electronic states for spin-flip processes in charge-transfer-type organic molecules with multiple donors and acceptors. Nat. Mater. 18, 1084–1090 (2019).
42. Evans, E. W. et al. Vibrationally assisted intersystem crossing in benchmark thermally activated delayed fluorescence molecules. J. Phys. Chem. Lett. 9, 4053–4058 (2018).
43. Kuang, Z. R. et al. Conformational Relaxation and thermally activated delayed fluorescence in anthraquinone-based intramolecular charge-transfer compound. J. Phys. Chem. C 122, 3727–3737, (2018).
44. Zhu, Z.-Q., Klimes, K., Holloway, S. & Li, J. Efficient cyclometalated platinum(II) complex with superior operational stability. Adv. Mater. 29, 1605002 (2017).
45. Yokoyama, D. Molecular orientation in small-molecule organic light-emitting diodes. J. Mater. Chem. 21, 19187–19202 (2011).












Figures

Fig. 1 | Chemical structures and results of quantum-chemical calculations for 5Cz-TRZ, mCz-TRZ, TmCz-TRZ, and DACT-II. (a) Chemical structures of 5Cz-TRZ, mCz-TRZ, TmCz-TRZ, and DACT-II. (b) Natural Transition Orbitals describing the excitation characters of the S1, T1, and T2 states in 5Cz-TRZ; the weights of the hole–electron contributions to the excitations are included. (c) Left: Correlation diagram for the triplet states of the carbazole, phenylene-triazine, and carbazole-phenylene-triazine fragments, based on the T1 equilibrium geometry of 5Cz-TRZ. Right: Natural Transition Orbitals for the T1 states in the three fragments.



Fig. 2 | Photophysical characterization of 5Cz-TRZ. a, Absorption and PL spectra (300 K) of 5Cz-TRZ in dilute toluene solutions. b, Phosphorescent (Phos) and PL spectra of 5Cz-TRZ in toluene glass at 77 K. c, TrESR spectra (80 K) of 5Cz-TRZ in dilute toluene solutions. d, Transient PL spectra (300 K) of 5Cz-TRZ in oxygen-free dilute solution with different polarity. e, Transient absorption contour maps (300 K) of 5Cz-TRZ on nanosecond timescales in oxygen-free toluene solutions. Excitation wavelength: 355 nm, fluence: 5.49 μJ/cm2. f, Transient absorption spectra (300 K) of 5Cz-TRZ on nanosecond timescales in oxygen-free toluene solutions. g, Decay kinetics of the transient absorption taken at 750-760 nm and 970-980 nm. h, Temperature-dependent transient PL (200-300 K) decay curves of the 5Cz-TRZ doped film. i, Temperature-dependent reverse intersystem crossing (RISC) rate constants of the 5Cz-TRZ doped film.


Fig. 3 | Device structures and materials. a, Architectures of the TADF and hyperfluorescent OLEDs. b, Chemical structures of the TADF and fluorescent emitters.







Fig. 4 | Device performance of OLEDs. a, EQE versus luminance characteristics. b, Normalized EL spectra. c, Transient EL decay characteristics of device A, B, C at 10, 1000 and 5000 cd m−2. d, Operational lifetimes of the TADF and hyperfluorescent OLEDs. 

















Methods
Synthesis and characterization. The general procedure for the synthesis of the TADF molecules and the characterization of their chemical properties are reported in the Supplementary Information.
Computational details. The initial ground-state geometries of the molecules were optimized with the long-range corrected functional ωB97XD (with the default range-separation parameter ω of 0.2 bohr-1) and the 6-31G(d,p) basis set.46 Then, following our earlier works,47 an iterative “gap-tuning” procedure was applied to obtain the optimal ω values for these geometries. The Tamm-Dancoff approximation (TDA) was employed in the framework of time-dependent density functional theory (TD-DFT) to study the excited-state properties; all the excited-state properties were examined at the TDA tuned-ωB97XD/6-31G(d,p) level combined with the polarizable continuum model (PCM; implicit solvent: toluene). Natural transition orbital analyses were also performed to examine the nature of the excited states; the spatial overlaps between the density distributions of the hole and electron wavefunctions in the excited states were estimated with the Multiwfn code.48 All quantum-chemical calculations were performed with the Gaussian 16 program.49 In addition, the spin-orbit couplings were estimated by employing the Breit-Pauli spin-orbit Hamiltonian with an effective charge approximation, as implemented in the PySOC code.50 
Absorption, steady-state PL spectra, PLQE and energy level measurements. Toluene solutions containing the TADF emitters (1 × 10−6 mol L−1) were prepared to investigate their absorption and photoluminescence characteristics in the solution state. Neat film samples were deposited on quartz glass substrates by vacuum evaporation to study their excitons confinement properties in the film state. UV-vis and PL spectra were recorded on a Perkin-Elmer Lambda 950 KPA spectrophotometer and JobinYvon FluoroMax-3 fluorospectrophotometer, respectively. Phos spectra were recorded on a JASCO FP-6500 fluorescence spectrophotometer at 77 K. Absolute PL quantum yields were measured on a Quantaurus-QY measurement system (C11347-11, Hamamatsu Photonics) under nitrogen flow and all samples were excited at 360 nm. The HOMO levels of neat films (100 nm) were measured by a Riken Keiki AC-3 photoelectron spectroscopy.
Thermal analysis. Thermal gravimetry-differential thermal analysis (TG-DTA) was performed by Bruker TG-DTA 2400SA with a heating rate of 10 °C min−1 under nitrogen atmosphere. Melting point was determined with a BUCHI Melting Point M‐565 instrument.
Electron paramagnetic resonance (EPR) measurements. Transient continuous-wave EPR measurements (trEPR) were performed in the Centre for Advanced ESR (CAESR) in the Department of Chemistry of the University of Oxford, using both Bruker BioSpin EleXSys I E680 and EleXSys II E580 spectrometers at X-band, around 9.75 GHz. The resonators used were Bruker EN 4118X-MD4W and ER 4118X-MD5W, with temperature controlled with an Oxford Instruments CF935O cryostat. The optical excitation of the samples employed use of an Opotek Opolete HE355 LASER with a 9 ns pulse length, synchronised to the spectrometer by a Stanford Research DG645 delay generator. EPR simulations were performed with the EasySpin toolbox in MatLab.51
Time-resolved spectroscopic measurements. All the solutions were deoxygenated with dry nitrogen gas to eliminate the deactivation of triplets by oxygen. The transient PL decay characteristics of solution samples were recorded using a Quantaurus-Tau fluorescence lifetime measurement system (C11367-03, Hamamatsu Photonics). Time‐resolved PL of the film samples were measured using an Andor electrically gated intensified charge‐coupled device and laser excitation at 400 nm. For low‐temperature measurements, an Oxford Instruments continuous flow cryostat was used with liquid helium as the coolant.
Transient absorption (TA) measurements. Sample photoexcitation in the nanosecond transient absorption experiments was achieved by the third harmonic (355 nm) of an electronically triggered Q-switched Nd:YVO4 laser (~1 ns pulse length, Advanced Optical Technologies Ltd AOT-YVO-25QSPX). For the picosecond transient absorption, ~100 fs excitation pulses at 400 nm, generated from the second harmonic of the 800 nm fundamental of the Ti:sapphire laser (Spectra Physics Solstice Ace), was used. For both temporal regions, the probe was generated by home-built broadband visible (500 – 770 nm) and NIR (830 – 1025 nm) non-collinear optical parametric amplifiers (NOPAs), pumped using the frequency-doubled output (400 nm) of the Ti:sapphire laser. The delay between the pump and probe pulses was varied using a Stanford DG645 delay generator for the nanosecond measurements, while a mechanical delay stage (Thorlabs DDS300-E/M) was used to delay the probe with respect to the pump for the picosecond measurements. The transmitted probe pulses were collected with a silicon dual-line array detector (Hamamatsu S8381-1024Q), which was driven and read out by a custom-built board from Stresing Entwicklungsbüro.
Angle-dependent PL measurements and optical simulation. The angle-dependent PL measurement was measured by a spectrometer (Ocean Optics USB 4000) together with a 405 nm continuous wave laser as the excitation source, a half cylindrical lens, a motorized rotation stage, and a polarizer to select p-polarized light. Optical simulation of present OLEDs was performed by an in-house developed tool which was described in detail by Furno et. al.52 The radiative dipole was considered as a forced damped harmonic oscillator and the transfer matrix method was adopted for calculation. The emission spectrum, PLQY and orientation factor of the emitter were taken from the experimental results. The electrical efficiency of the devices was set to be 100%.
Device fabrication and measurements. OLEDs were fabricated through vacuum deposition of the materials at ca. 6.0  10−8 Torr onto ITO-coated glass substrates with a sheet resistance of approximately 15 Ω sq−1. The ITO surface was sequentially cleaned ultrasonically with acetone, isopropanol, and deionized water, dried in an oven, and then exposed to UV/ozone for about 30 min. Organic layers were deposited at a rate of 2–3 Å s−1; Liq was subsequently deposited at 0.2 Å s−1 and then capped with Al (ca. 3 Å s−1). The devices were exposed once to nitrogen gas after the formation of the organic layers because a metal mask was included to define the cathode area. For all the OLEDs, the emitting area determined by the overlap of the two electrodes was 4.5 mm2. Current density-voltage-luminescence (J-V-L) characteristics were measured using a Keithley 2400 sourcemeter, Keithley 2000 multimeter and a calibrated silicon photodiode. Time‐resolved EL of the samples was recorded by the same ICCD spectrometer used in the time‐resolved PL measurements.
References accompanying methods section
46. Chen, X.-K., Kim, D. & Bredas, J.-L. Thermally activated delayed fluorescence (TADF) path toward efficient electroluminescence in purely organic materials: molecular level insight. Acc. Chem. Res. 51, 2215−2224 (2018).
47. Chen, X.-K. et al. Intramolecular noncovalent interactions facilitate thermally activated delayed fluorescence (TADF). J. Phys. Chem. Lett. 10, 3260−3268 (2019).
48. Lu T. & Chen F. Multiwfn: A multifunctional wavefunction analyser. J. Comput. Chem. 33, 580–592 (2012).
49. Frisch, M. J. et al. Gaussian 16, Revision C.01 (Gaussian Inc., 2016).
50. Gao, X. et al. Evaluation of spin-Orbit couplings with linear-response time-dependent density functional methods. J. Chem. Theory Comput. 13, 515−524 (2017).
51. Stoll, S. & Schweiger, A. EasySpin, a comprehensive software package for spectral simulation and analysis in EPR. J. Magn. Reson. 178, 42–55 (2006)
52. Furno, M., Meerheim, R., Hofmann, S., Lüssem, B. & Leo, K. Efficiency and rate of spontaneous emission in organic electroluminescent devices. Phys. Rev. B 85, 115205 (2012).

Acknowledgements
L.-S.C., A.J.G., E.W.E. and R.H.F. acknowledge the Engineering and Physical Sciences Research Council (EPSRC) for funding (EP/M01083X/1, EP/M005143/1). The Centre for Advanced ESR (CAESR) is supported by UK EPSRC (EP/L011972/1). We thank Diamond Light Source (UK) for synchrotron beamtime on I19 (CY21497). X.-K.C. and J.-L.B. acknowledge support from the Georgia Institute of Technology, Georgia Research Alliance, Vasser-Woolley Foundation, and Kyulux. H.Y., Z.-S.L., N.H. and C.A acknowledge the Japan Science and Technology Agency (JST), ERATO, Adachi Molecular Exciton Engineering Project for funding (JPMJER1305). Y.L acknowledges a stipend from the Chinese Scholarship Council (CSC). S.-F.Z. acknowledges financial support by Guangxi Department of Science and Technology (NO.AD19110030), Department of Education (NO.2019KY0394) and the start-up funds provided by Guangxi University of Science and Technology. E.W.E also thanks the Leverhulme Trust for funding (ECF-2019-054). We also thank Prof. Jun Yeob Lee for providing the control TADF materials and Prof. Chen Zhong for helpful discussions.
Author contributions
L.-S.C. designed the molecules and carried out device fabrication and measurements. A.J.G. conducted the transient absorption experiments. S.-F.Z. performed the theoretical calculations under the supervision of X.-K.C.. Y.L. performed the molecular orientation and optical simulations under the supervision of S. R.. H.Y. and Z.-S.L carried out steady-state and time-resolved photophysical properties under the supervision of C.A.. E.W.E. and W.K.M conducted the transient electron spin resonance measurements and analysed the results. T.K.R. performed single crystal X-ray diffraction and analysed the results. H.N. participated in the discussion of the photophysical mechanism. L.-S.C., A.J.G., X.-K.C., J.-L.B. and R.H.F. analysed the data and wrote the manuscript. All authors discussed the progress of research and reviewed the manuscript. 
Additional information
Supplementary information is available in the online version of the paper. Reprints and permissions information is available online at www.nauture.com/reprints. 
Correspondence and requests for materials should be addressed to L.-S. C and R.H.F
Data availability
The data that support the plots within this paper and other findings of this study are available in the University of Cambridge Repository (https://doi.org/10.17863/CAM.52923). Related research results are available from the corresponding authors upon reasonable request. Competing interests
The authors declare no competing interests.
20

image1.png

image2.png

image3.png

image4.png