Spin- and Voltage-dependent emission from Intra- and Intermolecular TADF OLEDs

Organic light emitting diodes (OLEDs) based on thermally activated delayed fluorescence (TADF) utilize molecular systems with a small energy splitting between singlet and triplet states. This can either be realized in intramolecular charge transfer states of molecules with near-orthogonal donor and acceptor moieties or in intermolecular exciplex states formed between a suitable combination of individual donor and acceptor materials. Here, we investigate 4,4'-(9H,9'H-[3,3'-bicarbazole]-9,9'-diyl)bis(3-(trifluoromethyl) benzonitrile) (pCNBCzoCF3), which shows intramolecular TADF but can also form exciplex states in combination with 4,4',4''-tris[phenyl(m-tolyl)amino]triphenylamine (m-MTDATA). We observed that orange emitting exciplex-based OLEDs also generate a sky-blue emission from the intramolecular emitter with an intensity that can be voltage-controlled. The main drawback of all-optical methods and electro-optical device characterization is their lack of sensitivity for the spin degree of freedom as long as no magnetic field is applied. In this work, we use electroluminescence detected magnetic resonance (ELDMR) to study the thermally activated spin-dependent triplet to singlet up-conversion in operating devices. Thereby, we can investigate intermediate excited states involved in OLED operation and derive the corresponding activation energy for both, intra- and intermolecular based TADF. Furthermore, we can give a lower estimate for the extent of the triplet wavefunction to be>1.2 nm. Photoluminescence detected magnetic resonance (PLDMR) reveals the population of molecular triplets in optically excited thin films. Overall, our findings allow us to draw a comprehensive picture of the spin-dependent emission from intra- and intermolecular TADF OLEDs.


I. Introduction
Organic light emitting diodes (OLEDs) represent a promising alternative to conventional LEDs for display applications and room lighting. One of the major challenges in the development of OLED technologies has been the improvement of efficiencies since only 25% of injected charge carriers form emissive singlet states, while 75% form non-radiative triplet states (Brown1992, Rothberg1996, Bruetting2012). However, thermally activated delayed fluorescence (TADF) can be induced, if molecules exhibit a small energy splitting ∆EST between singlet and triplet states (Endo2009, Uoyama2012, Goushi2012). An enhanced reverse intersystem crossing (RISC) enables harvesting of triplets thus dramatically increases the efficiency of TADF based devices.
∆EST is predominantly determined by the orbital overlap between the highest occupied molecular orbital (HOMO) and the lowest unoccupied molecular orbital (LUMO) of electrons and holes forming excitons (Endo2009). Therefore, the strategy to achieve small ∆EST is to design molecules which possess spatially separated HOMO and LUMO levels.
This has either been achieved in molecules with twisted donor and acceptor moieties forming intramolecular charge transfer (CT) states (Dias2013, Zhang2014) or in suitable combinations of separate donor and acceptor molecules forming intermolecular exciplex states (Goushi2012, Hung2013, Li2014). Both approaches have been implemented successfully and high external quantum efficiencies (EQE) in the range of 20% have been demonstrated (Zhang2014, Liu2016).
One of the remaining challenges is to tune the emission color of the used molecules for the desired application while maintaining high efficiency. An advantage of OLEDs for lighting applications is their surface-emission characteristic since conventional LEDs emit as potentially glaring point sources. Consequently, it is desirable to build efficient OLEDs which produce warm white light. The corresponding broad emission spectrum is often realized by blending the spectra of two single complementary emitters. One approach is the combination of a blue and an orange light source which has been realized by a blue fluorescent and an orange phosphorescent emitter (Schwartz2006, Qin2005).
Meanwhile, it is still an important point of discussion how the spin degree of freedom influences the first order forbidden RISC rate in operational TADF OLEDs. Since pCNBCzoCF3 shows both intra-and intermolecular CT states it is an interesting model system to employ inherently spin sensitive methods that just recently were applied for the first time to donor:acceptor based intermolecular TADF OLEDs (Vaeth2017, Bunzmann2019), but not yet to intramolecular emitters or their combination. In this study, we investigate the TADF characteristics of the building blocks of warm white OLEDs based on the intramolecular CT emission from pristine pCNBCzoCF3 and devices based on emission from exciplex states formed between pCNBCzoCF3 and m-MTDATA. Moreover, we elucidate properties of the involved triplet states that are involved in the RISC mechanism for both device types by using electroluminescence detected magnetic resonance (ELDMR) for electrical generation and photoluminescence detected magnetic resonance (PLDMR) for optical excitation.  ITO was used as an anode for all devices and Ca/Al as a cathode. PEDOT:PSS and BCP were used as hole and electron transport layers respectively. Two types of devices were built with different choice of materials for the emissive layer.

II. Materials and Devices
The first device type is based on TCTA and pCNBCzoCF3 in the emissive layer (left part of Figure 1b). In this case, excitons form on pCNBCzoCF3 giving rise to sky-blue emission as shown in the corresponding EL spectrum (Figure   1c). This EL spectrum and the PL spectrum of pristine pCNBCzoCF3 (Figure S1), are almost identical showing that there is no exciplex formation between TCTA and pCNBCzoCF3. Instead, TCTA functions as an additional transport layer increasing the efficiency of the device.
For the second type of device, the emissive layer consists of m-MTDATA and pCNBCzoCF3 (right part of Figure   1b). Here, a clear redshift between the EL spectrum of the device and the PL spectrum of pristine m-MTDATA and the EL spectrum of the pCNBCzoCF3-based device is recognizable (Figure 1c). This proves the formation of an exciplex state which gives rise to orange emission. There is a shoulder at 420-500 nm which can be assigned to additional emission from pristine pCNBCzoCF3 as shown in Figure 2b. Here, holes overcome the energetic barrier at the m-MTDATA:pCNBCzoCF3 interface and form excitons with electrons in the pCNBCzoCF3 layer, giving rise to the observed shoulder in the EL spectrum. This effect however does not need to be considered a deficiency since the use of an intramolecular TADF emitter as one of the constituents for an exciplex emitter can be beneficial for device efficiency (Liu2016). In such a case, triplets can undergo RISC on a TADF molecule, whereas they could be lost due to non-radiative decay on a conventional emitter molecule. As a result, a potential loss channel is averted. With a combination of the sky-blue emission of pristine pCNBCzoCF3 and the orange emission of the exciplex, a warm white OLED can be realized (Kaminskiene2018, Sych2020). and an EL spectrum of the exciplex-based device. The shoulder at 420-500 nm can be assigned to emission from pristine pCNBCzoCF3. (c) CIE1976 chromaticity coordinates of the EL spectra presented in (a). From 6 V to 10 V the color temperature shifts from 2866 K to 3231 K.
Remarkably, the relative intensity of the pCNBCzoCF3 emission at the shoulder increases with increased driving voltage, as shown in Figure 2a. Between 6 V to 10 V the color temperature of the respective EL spectra shifts from 2870 K to 3230 K (Figure 2c). A device, that is optimized to exhibit emission from both pristine pCNBCzoCF3 and the exciplex state, could be used to tune the emission in between warm and cold white by setting a different driving voltage. Meanwhile the brightness could be controlled independently by pulse width modulation. This concept is desirable as modern smartphone displays already employ a change in color temperature (so called "night shift") and this is also in demand for future room lighting applications. So far this can only be implemented by incorporating several OLEDs of different color and not within just one device.
If color purity is desired from the orange exciplex-based OLEDs, the corresponding devices can be realized with a 1:1 blended layer of m-MTDATA: pCNBCzoCF3 in the emission layer. In this case the EL shoulder at around 450 nm vanishes ( Figure S2). spectra of a pCNBCzoCF3-based device and an exciplex-based device, recorded at 245 K. The magnetic field axis is shifted, such that resonance peaks at B=B0 are centered around B-B0. The scaling of left and right axis was chosen such that the shape of both spectra can be compared properly. A Lorentzian fit to the spectra is shown in black.

III. Magnetic Resonance
(b) Illustration of the origin of the magnetic resonance spectra. The Zeeman energy levels of a CT triplet are split in an external magnetic field B and transitions that are induced by microwave photons (hν MW ) can be detected in relative electroluminescence change (∆EL/EL) as one signal of two overlapping contributions centered at B=B0 with its width determined by the dipolar interaction D. (c) OLED emission intensity (IP) in dependence of device current density (j) for different temperatures. The inset shows a schematic illustration of the used cryostat setup. The OLED is driven by a current j while the emitted light induces a photocurrent IP in a photodetector placed in front of the OLED. The slope ∆IP/∆j is proportional to the EQE. For a TADF based device the EQE depends exponentially on the activation energy EA and on the inverse thermal energy 1/kT. (d) Arrhenius plot of the slopes of the IP-j curves and the integrated ELDMR spectra. The activation energy EA is derived from the slope of linear fits for each data set.
Kaminskiene et al. showed the potential of pCNBCzoCF3 for building warm white OLEDs (Kaminskiene2018). The focus of their work was to explore the spectral properties and performance characteristics of the corresponding devices.
However, a detailed investigation of the TADF characteristics of pCNBCzoCF3 as well as spin sensitive measurements, such as ELDMR, are still missing. These measurements elucidate properties of triplet states which are involved in the light generation mechanisms of the devices and the temperature dependence reveals the TADF activation energy (Vaeth2017, Bunzmann2020). The idea of ELDMR experiments is to probe EL, while applying a static magnetic field that induces a Zeeman splitting of triplet states. By applying resonant microwaves, the following resonance condition is fulfilled (Weil2007): Here ℎ is the Plank constant, MW is the microwave frequency, = 2.002 is the g-factor of the spin system, / is the Bohr magneton, ∆ s = 1 is the allowed change of the magnetic quantum number, D is the dipolar interaction and is the angle between the direction of the external magnetic field and the vector connecting the two spins of the triplet state. During an ELDMR measurement microwaves of fixed frequency MW are applied while the magnetic field is swept. During this sweep the microwave-induced change of electroluminescence ∆EL is detected.
ELDMR spectra for OLEDs based on pCNBCzoCF3 emission and on exciplex emission (1:1 mixed layer) are shown in Figure 3a. In both cases a change of EL arises in resonance. The intensity of the signals is different but the line shape is almost identical. The origin of these signals is illustrated in Figure 3b. Under resonant conditions rate equations that describe the interplay between triplet sublevels and the singlet state are altered, causing a change of the RISC rate and consequently a change in EL. The dipolar interaction D lifts the degeneracy between triplet states with s = 0 and s = ± 1 at =0, which results in two resonant transitions at different magnetic fields. If D is small, these two magnetic fields are very close. If the molecules are also randomly oriented the respective resonance curves overlap and appear as a single broadened peak. Accordingly, the corresponding linewidth is a measure for the strength of D. The exact value can, however, not solely be derived from the linewidth because other broadening mechanisms like unresolved hyperfine interactions with nearby nuclei are contributing as well. Instead, the full width at half maximum (FWHM) can be considered as an upper limit for 2D. This estimation allows the calculation of a lower boundary for the extent of the triplet wave function, i.e. the distance BCD between electron and hole forming the triplet state. The following approximation can be made (Jeschke2002): Here, one obtains BCD in units of nm by using in units of mT. From the FWHM = 3 mT of the ELDMR spectra shown in Figure 3a one finds 2 ≤ 3 mT resulting in BCD ≥ 1.2 nm. Such numbers fit to delocalized triplet states, where the electron-hole distance is large. In contrast, strongly localized molecular triplet excitons, that are being discussed to be involved in TADF emission (Dias2013, Santos2016), exhibit distinct broader spectra because of close electron-hole distance and thus strong dipolar interaction (Vaeth2016, Bunzmann2020). Such molecular triplets are, however, not observed in the studied material systems. Consequently, the narrow ELDMR linewidth is consistent with expectations for CT triplet states. For the device based on pure pCNBCzoCF3 emission, this corresponds to the 3 CT triplet, delocalized between the accepting and the donating moieties of pCNBCzoCF3. For the exciplex-based device, it corresponds to the 3 Exc triplet, formed between an electron located on pCNBCzoCF3 and a hole located on m-MTDATA. The striking similarity of the two signals is evidence for almost identical wavefunction delocalization in both cases: a delocalized triplet is probed, that extents over different moieties of a single molecule or even two adjacent but separate molecules.
To investigate the TADF character of both device types we measured temperature dependent ELDMR spectra ( Figure   S3). In both cases the signal shape stays identical, while the intensity decreases with decreasing temperature. This tendency is contrary to what is commonly observed in magnetic resonance experiments, where lower temperatures yield a higher signal intensity due to a higher spin polarization. Consequently, we assign the thermal response of the ELDMR spectra to the thermal activation of delayed fluorescence. A quantitative analysis of the signal intensities can be carried out via an Arrhenius plot, which allows us to derive an activation energy EA of the probed effect as shown in Figure 3d. For the device based on pure pCNBCzoCF3 we find an activation energy of 37.3±6.2 meV and for the exciplex-based OLED 33±10 meV. Both device types yield an activation energy in the range of thermal energy kBT which fits to TADF emitters. Hence, we consider these values to be an estimate for the singlet triplet splitting ∆EST.
The determination of ∆EST for pCNBCzoCF3 has shown considerable discrepancies in different previous reports. In (Kaminskiene2018) a ∆EST of 11 meV was reported, while (Cao2017) reported 190 meV. Both values were determined via the difference between emission peaks in PL and phosphorescence spectra of films at 77 K. This mismatch demonstrates that the determination of ∆EST via PL measurements is not unambiguous and has to be treated carefully.
Our results support the value of (Kaminskiene2018), which is in the range of thermal energy at room temperature, in line with reasonable values for TADF emitters.
As a comparative method, we measured temperature-dependent EQE of the exciplex-based device (1:1 mixed layer) to corroborate our findings. The exact temperature-dependence of a TADF OLED's EQE is non-trivial (Tao2014), however, in a first order approximation, one can assume: EQE ~ exp(-EA/kBT), with an activation energy EA. A schematic illustration of the corresponding setup is shown in the inset of Figure 3c. An OLED is driven by a current j and the emitted light is collected with a photodiode placed in front of the OLED measuring the photocurrent IP.
Temperature dependent current density (j) and electroluminescence (IP) to voltage (V) characteristics for an exciplexbased device are shown in Figure S4. In order to disentangle the effects of TADF and temperature dependent hopping transport in organic semiconductors the photocurrent IP is plotted vs. the current density j in Figure 3c. The resulting curves yield a linear dependency. The slope ∆IP/∆j of these lines is proportional to the EQE of the device. A quantitative analysis of the temperature dependence of these slopes is, therefore, equivalent to an analysis of the EQE itself. An Arrhenius plot allows the determination of the corresponding activation energy EA, which is included in Figure 3d. Here an activation energy of 44.5±6.5 meV is determined for the exciplex based device, which is slightly higher than the value of 33±10 meV derived via ELDMR, but still consistent within the experimental errors. Overall, a value in the range of thermal energy is obtained, confirming the TADF character of the exciplex state between pCNBCzoCF3. We performed PLDMR measurements in order to explore the differences between the generation of excitons via optical excitation or electrical injection. In PLDMR, the external magnetic field is swept and the change of PL from an optically excited film is probed while resonant microwaves are applied.

IV. Photoluminescence Detected Magnetic Resonance
First, we measured temperature-dependent PLDMR of a m-MTDATA:pCNBCzoCF3 blend which is presented in Alternatively to the CT process, ISC can facilitate the population of local triplets 3 LE on m-MTDATA or CT triplets 3 CT on pCNBCzoCF3 (iii). Electron and hole are delocalized over two molecules (exciplex) and for the pCNBCzoCF3 excited states they are delocalized between donor:acceptor moieties of the same molecule. The distance between them is therefore relatively large, resulting in a small dipolar interaction . As already discussed in Figure 3b, spin species with a small exhibit narrow signals in magnetic resonance experiments which is why the narrow component in PLDMR spectra of m-MTDATA:pCNBCzoCF3 blends fits to the exciplex and the CT triplet. The broad signal can be assigned to the local triplet on m-MTDATA where electron and hole are localized on one molecule leading to a stronger dipolar interaction and consequently to a broader spectrum (Bunzmann2020).
The vanishing of the broad component at higher temperatures can be explained by the temperature dependence of the CT process from m-MTDATA to pCNBCzoCF3. CT processes are mediated by molecular vibrations which are less activated at lower temperatures (Marcus1956). Consequently, the efficiency of the CT process decreases at low temperatures and the probability for ISC from m-MTDATA singlets to triplets increases which is why the broad signal is more pronounced at lower temperatures. Local m-MTDATA triplets can decay non-radiatively of alternatively still undergo CT to the exciplex triplet (iv) and thus contribute indirectly to delayed fluorescence from exciplexes.
Consequently, spin manipulation of m-MTDATA triplets causes a change in exciplex emission intensity as observed by PLDMR, which enables indirect detection of non-emissive local triplet states.
In the next step, we measured PLDMR of a TCTA:pCNBCzoCF3 blend which is presented in Figure 4c. Similarly, there is a superposition of a broad and a narrow signal where the broad component is predominantly pronounced at low temperatures and the narrow one at high temperatures. In this case the broad signal can be assigned to the local triplet 3 LE on TCTA and the narrow one to the CT triplet 3 CT on pCNBCzoCF3.
To elaborate these assignments, the excitation pathways of spin states in an optically excited TCTA:pCNBCzoCF3 blend are illustrated in Figure 4d. Via optical excitation either local singlets 1 LE on TCTA or CT singlets 1 CT on pCNBCzoCF3 are generated (i) -as both materials can be excited at 365 nm according to photoexcitation spectra (see Figure S5a for pCNBCzoCF3 and (Cao2018) for TCTA). The PL spectrum of a TCTA:pCNBCzoCF3 blend exhibits only emission from pCNBCzoCF3, while no emission from TCTA or red-shifted exciplex emission is observed ( Figure S5b). Excitations on TCTA must therefore be depopulated efficiently. The absence of emissive exciplex states indicates charge transfer is not expected. An alternative is singlet Förster transfer from TCTA to pCNBCzoCF3 (ii).
Singlets on both materials can form local triplets 3 LE on TCTA or CT triplets 3 CT on pCNBCzoCF3 via ISC (iii). The electron-hole separation of the local triplet on TCTA is small, whereas it is relatively large for the CT triplet on pCNBCzoCF3. This means a strong dipolar interaction D for the local TCTA triplet and a relatively small D for the CT triplet. Therefore, the local TCTA triplet is observed as a broad signal and the CT triplet as a narrow signal.
As the TCTA triplet is detected by monitoring pCNBCzoCF3 PL with PLDMR, subsequent Dexter triplet transfer between TCTA and pCNBCzoCF3 seems plausible (iv). Consequently, ISC on TCTA does not need to be considered as a loss mechanism, but just further delays emission. In the exciplex-based device, free charges populate singlets and triplets in the exciplex state but to some extent also on pCNBCzoCF3. Emission originates from the exciplex singlet via prompt fluorescence and via delayed fluorescence after RISC from the exciplex triplet. Singlets on pCNBCzoCF3 decay either radiatively or undergo CT to the exciplex singlet. Triplets on pCNBCzoCF3 undergo RISC to form singlets or undergo CT to exciplex triplets. Activation energies of RISC processes are determined from temperature dependent ELDMR spectra. Neither is emission from m-MTDATA observed nor are signatures of local m-MTDATA triplets found in ELDMR spectra.

IV. Discussion
The relative positions of HOMO and LUMO levels in blends of pCNBCzoCF3 with m-MTDATA or TCTA are similar yet in one case an emissive exciplex state forms while in the other one only pCNBCzoCF3 emits light. By performing both PLDMR and ELDMR, we were able to reveal the characteristic magnetic resonance signatures of triplet states in these systems and determined the respective activation energies for light emission. This information allows us to draw a comprehensive picture of the light generation mechanisms in pCNBCzoCF3-based and in exciplex-based OLEDs.
The most important processes for the pCNBCzoCF3-based device are summarized in Figure 5a. Injected charges populate CT singlet and CT triplet states with a ratio of 1:3. Singlets decay radiatively as prompt fluorescence while triplets undergo RISC giving rise to delayed fluorescence. Both processes yield the sky-blue emission spectrum of pCNBCzoCF3. The triplet that is involved in RISC is found to be the CT triplet via ELDMR. Local triplets which are characterized by broad magnetic resonance spectra, as found in PLDMR, do not appear in ELDMR, and therefore do not play a role in the case of electrical injection. The activation energy of the RISC process is in the range of thermal energy proving that pCNBCzoCF3 exhibits TADF.
All involved processes for the exciplex-based device are shown in Figure 5b. The majority of charges populates the singlet and triplet of the exciplex state. Singlets decay radiatively via prompt fluorescence while triplets undergo RISC giving rise to delayed fluorescence. These processes are responsible for the orange emission. The triplet which is involved in RISC was identified as the exciplex triplet via ELDMR. The local triplet of m-MTDATA was only found in PLMDR but not in ELDMR. The activation energy of the RISC process is close to thermal energy which shows that the exciplex exhibits TADF as well as pCNBCzoCF3 itself. The EL spectrum of the exciplex-based device exhibits a shoulder which indicates additional population of CT singlets and triplets of pCNBCzoCF3. Singlets can decay radiatively giving rise to the observed shoulder in the EL spectrum but can also undergo a CT to the exciplex singlet.
Triplets on pCNBCzoCF3 can undergo a CT to the exciplex triplet or undergo RISC to the CT singlet of pCNBCzoCF3.

V. Conclusion
The inherently spin-sensitive techniques of electroluminescence and photoluminescence detected magnetic resonance (ELDMR, PLDMR) were employed to study efficient OLEDs based on intra-and intermolecular TADF effects. From the results we can draw the following conclusions. The electrically generated intermediate triplet states that are responsible for light generation are broadly delocalized (≥ 1.2 nm) over either one molecule that features intramolecular TADF or over two adjacent molecules which exhibit exciplex type emission. Aside from that, no strongly localized triplet excitons were observed for electrically driven devices with ELDMR. However, upon optical excitation, such localized molecular triplet excitons on the donor materials are generated at low temperatures as evident by PLDMR. By performing temperature-dependent EQE and spin-resonance measurements on both types of devices, we derived singlet-triplet gap ∆EST in the range of 33-45 meV which is in the order of thermal energy and hence in line with efficient TADF at ambient temperatures. Based on these studies, we propose an energy scheme of the recombination paths for the respective carrier generation modes. Finally, we observed that for orange emitting exciplex-based devices a residual sky-blue emission from the used intra-molecular TADF emitter is generated, whose intensity can be voltage controlled. This opens up the intriguing technical application of tuning both the OLED emission color by voltage adjustments and its brightness by pulse width modulation.

Experimental
The materials m-MTDATA, TCTA and BCP were purchased from Sigma-Aldrich. pCNBCzoCF3 was synthesized as described elsewhere (Kaminskiene2018). All materials were used as received.
All OLED devices were fabricated on indium tin oxide (ITO) covered glass substrates (1 cm 2 ). First, poly(3,4ethylendioxythiophene):polystyrolsulfonate (PEDOT:PSS, 4083Ai) from Hereaus was spin coated with 3000 rpm for 1 minute, resulting in a 40 nm thick film. All further device fabrication steps were done inside a nitrogen glovebox to avoid degradation, starting with annealing of the PEDOT:PSS layer for 10 minutes at 130°C. Donor and acceptor layers were thermally evaporated in an evaporation chamber.
For devices based on pristine pCNBCzoCF3 the device stack was: 40 nm TCTA, 40 nm pCNBCzoCF3, 10 nm BCP.
For exciplex bilayer devices: 40 nm m-MTDATA, 40 nm pCNBCzoCF3, 10 nm BCP. For exciplex mixed layer devices: 30 nm m-MTDATA, 30 nm pCNBCzoCF3, with an additional 50 nm mixed layer (1:1) in between the pristine material layers. Finally, for all devices, the top electrode was (5 nm Ca / 120 nm Al), completing 8 OLEDs (3 mm 2 each) per substrate. For PL samples, the materials were evaporated onto glass substrates without the use of ITO, PEDOT:PSS or a metal cathode.
PL and photoexcitation spectra were measured with a calibrated fluorescence spectrometer FLS980-s (Edinburgh Instruments) equipped with a continuous broad-spectrum xenon lamp Xe1. EL spectra were recorded by biasing the OLED with an Agilent 4155C parameter analyzer in constant current mode and coupling the emitted light via light guides to an Acton Spectra SP-2356 spectrometer (Princeton Instruments) or a SPM002 spectrometer (Photon Control).
PLDMR samples were prepared from solutions of the materials in chlorobenzene. ~100 µl of the solutions were poured into EPR quartz tubes and the solvent was then evaporated by vacuum pumping. The sample tubes were subsequently sealed under inert helium atmosphere. Sample tubes were inserted into an EPR microwave cavity with optical access (Bruker ER4104OR) and an Oxford cryostat. Optical excitation was provided by a 365 nm UV LED behind a 400nm shortpass filter. The PL was detected by a silicon photodiode placed in front of the cavity behind a 409 nm longpass filter. Magnetic resonance measurements were done in two modified EPR spectrometers (Bruker E300, see Figure   S6, S7) equipped with continuous flow helium cryostats (Oxford). OLED devices were placed in contact to a microwave transmission line and a silicon photodiode was mounted on top for detection of EL. Forward bias was provided by a source-measure unit (Keithley 237). EL, PL and bias currents were connected to current-voltage transimpedance amplifiers (Femto). The signal change upon resonant microwave irradiation was then detected via a Lock-In-Amplifier (SR7230) with the on-off modulated microwave as reference.

Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jpcc.XXXXX.

Associated Content
(Dated: June 28, 2020) Figure S1. Comparison of the PL spectrum of pCNBCzoCF3 and the EL spectrum of a pCNBCzoCF3-based device.
The spectra are almost identical showing that emission in the pCNBCzoCF3-based device originates only from pCNBCzoCF3 and there is no exciplex formation between TCTA and pCNBCzoCF3.
PL pCNBCzoCF 3 EL pCNBCzoCF 3 OLED Figure S2. EL spectrum of an OLED with a 1:1 m-MTDATA:pCNBCzoCF3 mixed layer in the emission layer. Figure S3. Temperature-dependent ELDMR spectra of (a) a pCNBCzoCF3-based device (b) an exciplex-based device together with Lorentzian fits (black lines). For both systems the signal decreases with decreasing temperature, which is in line with thermal activation of delayed fluorescence. Therefore, the temperature behavior shows that the observed effect is of TADF nature.