Geminate and Nongeminate Pathways for Triplet Exciton Formation in Organic Solar Cells

Organic solar cells (OSCs) have recently shown a rapid improvement in their performance, bringing power conversion efficiencies to above 18%. However, the open‐circuit voltage of OSCs remains low relative to their optical gap and this currently limits efficiency. Recombination to spin‐triplet excitons is a key contributing factor, and is widely, but not universally, observed in donor–acceptor blends using both fullerene and nonfullerenes as electron acceptors. Here, an experimental framework that combines time‐resolved optical and magnetic resonance spectroscopies to detect triplet excitons and identify their formation mechanisms, is reported. The methodology is applied to two well‐studied polymer:fullerene systems, PM6:PC60BM and PTB7‐Th:PC60BM. In contrast to the more efficient nonfullerene acceptor systems that show only triplet states formed via nongeminate recombination, the fullerene systems also show significant triplet formation via geminate processes. This requires that geminate electron–hole pairs be trapped long enough to allow intersystem crossing. It is proposed that this is a general feature of fullerene acceptor systems, where isolated fullerenes are known to intercalate within the alkyl sidechains of the donor polymers. Thus, the study demonstrates that engineering good donor and acceptor domain purity is key for suppressing losses via triplet excitons in OSCs.


DOI: 10.1002/aenm.202103944
now exceed 18%. [1][2][3][4][5] However, the photon energy loss in OSCs, defined as the difference between the optical gap (E g ) and the energy of the extracted charges (qV OC ), [6] where q is the elementary charge and V OC is the open-circuit voltage, remains high and is now the main factor limiting OSC performance. [7,8] The primary cause is excessive nonradiative recombination in OSCs, [9] which impacts the V OC of the solar cell by reducing the charge carrier lifetime from the intrinsic radiative limit. [10][11][12] This non radiative voltage loss (ΔV nr ) can be directly calculated from the electroluminescence external quantum efficiency (EQE EL ) of the solar cell run at moderate forward bias as a light-emitting diode ( ) where k B is the Boltzmann constant and T is temperature. Here, the EQE EL can be further separated into the different contributions [13] γ χη = Φ EQE EL PL out (2) where γ is the charge balance factor, Φ PL is the photoluminescence quantum efficiency, χ is the fraction of recombination Organic solar cells (OSCs) have recently shown a rapid improvement in their performance, bringing power conversion efficiencies to above 18%. However, the open-circuit voltage of OSCs remains low relative to their optical gap and this currently limits efficiency. Recombination to spin-triplet excitons is a key contributing factor, and is widely, but not universally, observed in donoracceptor blends using both fullerene and nonfullerenes as electron acceptors.
Here, an experimental framework that combines time-resolved optical and magnetic resonance spectroscopies to detect triplet excitons and identify their formation mechanisms, is reported. The methodology is applied to two wellstudied polymer:fullerene systems, PM6:PC 60 BM and PTB7-Th:PC 60 BM. In contrast to the more efficient nonfullerene acceptor systems that show only triplet states formed via nongeminate recombination, the fullerene systems also show significant triplet formation via geminate processes. This requires that geminate electron-hole pairs be trapped long enough to allow intersystem crossing. It is proposed that this is a general feature of fullerene acceptor systems, where isolated fullerenes are known to intercalate within the alkyl sidechains of the donor polymers. Thus, the study demonstrates that engineering good donor and acceptor domain purity is key for suppressing losses via triplet excitons in OSCs.

Introduction
Driven by the recent development of nonfullerene electron acceptor (NFA) materials, the power conversion efficiencies (PCEs) of organic solar cells (OSCs) have rapidly increased, and events that can decay radiatively (spin-singlet excitations) and η out is the photon out-coupling efficiency. To address the low EQE EL of OSCs, recent studies have focused on improving Φ PL of spin-singlet excitations in the donor:acceptor blend. [7,[14][15][16] However, the optimum optical gap for a single junction solar cell lies in the near-infrared spectral region [10,17] where most OSCs show stronger multi-phonon nonradiative recombination rates, termed the energy gap law. [9,18] As such, this may limit the scope for increasing Φ PL of spin-singlet states in such systems. [19] Recently, the recombination of charge carriers via spin-triplet excitons has been identified as another significant nonradiative voltage loss pathway in both fullerene and NFA OSCs. [20][21][22][23] For example, in the benchmark PM6:Y6 system, [24] the fraction of charge carriers that recombine via the triplet exciton (T 1 ) of the low E g component, Y6, is ≈90%; [20] comparable T 1 recombination fractions have also been reported in fullerene acceptor OSCs. [25,26] As a result, χ in Equation 2 is limited to 0.1 and the EQE EL of the PM6:Y6 blend is reduced by a factor of 10, lowering the V OC by ≈60 mV. We note that to achieve a V OC gain comparable to eliminating recombination via T 1 , Φ PL of the low E g component, which is considered to provide the limit for Φ PL of the blend when recombination can proceed via the lowest energy singlet exciton (S 1 ), [7,15] should be raised by a factor of 10. To explore the feasibility of such an improvement in luminescence efficiency, we have measured Φ PL for a neat film of the widely used NFA Y6. Here, we obtain Φ PL = 2%, which would necessitate an increase in Φ PL to ≈20%. A Φ PL of this magnitude would be unprecedented among fluorescent organic small molecules with an emission peak around 950 nm, [19] moreover in a molecule that can also operate as an electron acceptor in a highly efficient OSC. Thus, while enhancing Φ PL can provide incremental improvements in device performance, it is unlikely to alone yield the step-change in V OC required for PCEs of >20% to be realized in single junction OSCs. Furthermore, triplet states have also been implicated in the degradation of both fullerene and NFA OSC blends, potentially presenting a fundamental barrier to commercial applications which require photovoltaic modules with long term stability. [27][28][29][30] Therefore, eliminating recombination via T 1 should now be a key focus for further improving the V OC and operational lifetimes of OSCs. However, we note this remains an understudied topic in the field. [23,[31][32][33] This is likely due to the difficulty in detecting and characterizing spin-triplet states in organic semiconductors, as they are generally optically dark and, in the context of OSCs, often short lived due to the presence of rapid annihilation processes. [20]

Probing Recombination via Spin-Triplet Excitons in Organic Solar Cells
We present here an experimental framework for probing triplet excitons in OSCs to assist with the task of engineering out recombination via T 1 . We consider that there are three main ways in which T 1 can be created in OSCs: 1) direct intersystem crossing (ISC) from un-dissociated S 1 states (Figure 1a); 2) back charge transfer (BCT) from geminate spin-triplet charge transfer ( 3 CT) states ( Figure 1b); 3) BCT from 3 CT states formed via nongeminate recombination (Figure 1c). In general, any T 1 states formed will ultimately relax to the lowest energy T 1 in the system. But, if molecular T 1 states are energetically higher than the 3 CT states, T 1 will not be formed. [34,35] As 3 CT states are readily converted to spin-singlet CT states ( 1 CT) through the hyperfine interaction Triplet formation pathways and organic solar cell materials studied. a) Schematic of the direct intersystem crossing pathway for T 1 formation in organic solar cells. 1) After optical excitation into S 1 , charge transfer to the 1 CT state does not occur and 2) T 1 is formed by intersystem crossing instead. b) Schematic of the geminate BCT pathway for T 1 formation in organic solar cells. 1) After optical excitation into S 1 , 3) charge transfer to the 1 CT state successfully occurs. 4) However, the 1 CT state does not separate into free charges (FC) and instead undergoes spin-mixing to form 3 CT states. These 3 CT states then undergo a spin-allowed back charge transfer process to form molecular T 1 states. c) Schematic of the nongeminate BCT pathway for T 1 formation in organic solar cells. 1) After optical excitation into S 1 , 3) charge transfer to the 1 CT state successfully occurs. 5) 1 CT then separates into free charges. 6) Spin-statistical nongeminate recombination of free charges leads to the formation of 3 CT states, which can then undergo back charge transfer to form molecular T 1 states. d) The chemical structures of the two donor polymers and the fullerene acceptor material used in this study. The structure of the nonfullerene acceptor Y6 is included for reference in the discussion of the PLQE results.
(HFI), [36] their presence is not expected to significantly impact device performance. [20] Therefore, T 1 states become a problem if they are energetically below the CT states. However, as the offset between CT states and the lowest energy S 1 is typically small (<0.2 eV) to reduce energy losses associated with charge generation, [7,9,14] the molecular T 1 states almost always lie below the CT states due to the large S 1 -T 1 energy gap in most organic semiconductors with localized molecular excitons. [37][38][39] Furthermore, as most OSC blends comprise at least one donor and one acceptor component, with ternary systems containing an additional donor or acceptor, [2][3][4][5]40,41] there is the potential for T 1 states to be formed on any of these materials through each of the three pathways presented above. Thus, fully understanding these complex and overlapping mechanisms will require the application of multiple experimental techniques, each targeting a specific subset of the possible T 1 formation pathways.
To achieve this, we propose the combination of optical and magnetic resonance spectroscopies, which have previously been utilized for investigating T 1 states in organic semiconductors. [20,25,[42][43][44][45] Specifically, we present a framework to investigate recombination via T 1 through three complementary methods: transient absorption (TA), time-resolved electron paramagnetic resonance (trEPR), and photoluminescence detected magnetic resonance (PLDMR) spectroscopies. In Table S1 in the Supporting Information, we present a summary of the three different techniques, as well as the T 1 recombination pathways that they can detect; a more detailed discussion on each of these techniques is provided with the corresponding experimental results (vide infra) and in the Supporting Information.
To develop our experimental framework, we have chosen to examine two model fullerene OSC systems: PM6:PC 60 BM and PTB7-Th:PC 60 BM (chemical structures in Figure 1d). These blends give PCEs of 7.4% and 7.5%, respectively; further information is given in Figure S1 in the Supporting Information. The polymers PM6 and PTB7-Th have been chosen as they are commonly used donor materials in efficient fullerene and NFA OSCs. [3,24,40,41,[46][47][48][49] We have opted to use fullerene acceptors for two key reasons. First, unlike NFAs which exhibit strong spectral features in TA, [20,50,51] the fullerene component does not make any significant contribution to the observed TA spectrum in the visible and near infrared probe regions. Thus, the use of fullerene blends avoids the complex superposition of the polymer and NFA spectral features and dynamics, simplifying the data interpretation. Second, in many NFA blends, the geminate BCT pathway to T 1 states is not observed in trEPR. [20] Conversely, fullerene blends often show geminate BCT T 1 formation. [23,43,52,53] Thus, fullerene acceptor blends are the ideal model systems to demonstrate how it is possible to probe the three main T 1 formation mechanisms, clarifying the strength of our approach and the complementarity of optical and magnetic resonance techniques.

Using Transient Absorption to Detect Triplet States in Organic Solar Cells
We begin by using TA to explore T 1 formation. TA has been widely used to explore the photophysical processes occurring in OSCs, [20,26,50,51,54,55] as it is able to provide insights into the evolution of both optically bright and dark states on timescales spanning femtoseconds to milliseconds. Thus, TA is well suited to probing optically dark T 1 states in OSCs as the distinct T 1 photoinduced absorption (PIA) signatures, typically located in the near-infrared (NIR) spectral region, [20][21][22]25,26,56] are a clear fingerprint for the presence and molecular location of these states. Furthermore, TA can, in theory, distinguish between the presence of a monomolecular (direct ISC or geminate BCT) or bimolecular (nongeminate BCT) T 1 formation pathways through the fluence dependence of T 1 generation; monomolecular pathways show no fluence dependence, [56] while bimolecular events exhibit a strong fluence dependence. [20,25,26] However, in the case where significant bimolecular pathways are present, the fluence dependent behavior of T 1 formation will dominate, masking any underlying monomolecular processes. Thus, as many fullerene and NFA OSCs blends demonstrate nongeminate BCT T 1 formation, [20][21][22]25,26] TA can, in general, only be reliably used to detect the nongeminate pathway.

Transient Absorption of PM6:PC 60 BM
In Figure 2a, we present the TA from the NIR region of PM6:PC 60 BM. At 0.2-0.3 ps after photoexcitation at 600 nm, we observe the presence of a PIA centered around 1175 nm. Through comparison to the TA of a neat PM6 film ( Figure  S2, Supporting Information), we attribute this feature to the S 1 state of PM6. The PM6 S 1 is rapidly quenched within a picosecond, indicating ultrafast electron transfer to PC 60 BM. Subsequently, over timescales of hundreds of picoseconds, we notice the formation of a new PIA band peaking at the edge of our probe range around 1650 nm. As the T 1 PIA for PC 60 BM has previously been reported at 720 nm, [57,58] we attribute this new PIA to T 1 located on PM6. We also observe a strong fluence dependence for the formation of this new PIA (Figure 2b), indicating that the nongeminate BCT process plays a large part in PM6 T 1 formation. In the highest fluence measurement for PM6:PC 60 BM presented here (6 µJ cm −2 ), we note that the rise in the T 1 PIA begins to flatten out towards 2 ns; this can be attributed to the competing effects of triplet-charge annihilation (TCA). As the rate of TCA depends on the charge carrier density in the blend film (as well as the charge carrier mobility [25] ), it is expected to become more prominent on subnanosecond timescales under higher excitation fluences. [20,25,26] Indeed, TCA is the primary nonradiative quenching pathway of T 1 in OSCs and is therefore directly responsible for the increased nonradiative voltage losses in OSCs with significant T 1 formation. [20,23,38]

Transient Absorption of PTB7-Th:PC 60 BM
We next discuss the TA from the NIR region of the PTB7-Th:PC 60 BM blend (Figure 2c). At 0.1-0.15 ps after photoexcitation at 700 nm, we observe two PIA features: one peaked at 1125 nm and a broader band extending towards the edge of our probe range at 1450 nm. The PIA at 1125 nm is attributed to the hole polaron located on PTB7-Th, [59] while the band around 1450 nm is confirmed to be the residual PTB7-Th S 1 PIA though comparison with the TA spectrum of a neat PTB7-Th film ( Figure S3, Supporting Information). As with PM6:PC 60 BM, the rapid quenching of the S 1 PIA indicates ultrafast electron transfer from the electron donor polymer to the fullerene. However, unlike PM6:PC 60 BM, we do not observe the formation of a clear T 1 PIA over timescales of hundreds of picoseconds. Rather, we see an apparent broadening on the low energy edge of the hole polaron PIA. Kinetic traces from this region around 1250-1300 nm reveal a clear fluence dependence (Figure 2d), indicating a bimolecular formation mechanism. Thus, in-line with a previous report, [59] we conclude that the PIA of PTB7-Th T 1 states, formed via the nongeminate BCT process, overlaps with the hole PIA.

Using trEPR Spectroscopy to Detect Triplet States in Organic Solar Cells
Having explored the nongeminate BCT pathway, we now investigate the geminate BCT and direct ISC pathways for T 1 formation using trEPR spectroscopy. trEPR typically has a time resolution on the order of hundreds of nanoseconds and is sensitive to the presence of states with unpaired spins; [60,61] in OSCs, this primarily includes T 1 states, spin-correlated radical pairs (which can be considered analogous to CT states in an OSC blend), and free charge carriers. [43,52,53] With a focus on T 1 states, trEPR not only provides information on the molecular Figure 2. Transient absorption spectroscopy of the organic solar cell blends studied. a) The TA spectra of a PM6:PC 60 BM blend film, excited at 600 nm with a fluence of 2.8 µJ cm −2 . The PM6 S 1 PIA centred at 1175 nm decays within the first picosecond due to electron transfer to PC 60 BM. Over hundreds of picoseconds, a new PIA band around 1650 nm begins to grow in, indicating recombination into PM6 T 1 states. b) The TA kinetics a PM6:PC 60 BM blend film, excited at 600 nm with varying fluence. The fluence dependence of the T 1 PIA growth shows that T 1 formation occurs following the bimolecular recombination of free charge carriers. c) The TA spectra of a PTB7-Th:PC 60 BM blend film, excited at 700 nm with a fluence of 3.6 µJ cm −2 . The PTB7-Th S 1 PIA band around 1400 nm decays within the first picosecond due to electron transfer to PC 60 BM, leaving behind the PTB7-Th hole polaron PIA at 1125 nm. Over hundreds of picoseconds, a new PIA band on the low energy edge of the hole polaron around 1300 nm begins to grow in, indicating recombination into PTB7-Th T 1 states. d) The TA kinetics a PTB7-Th:PC 60 BM blend film, excited at 700 nm with varying fluence. The fluence dependence of the T 1 PIA region growth shows that T 1 formation occurs following the bimolecular recombination of free charge carriers. All TA measurements were performed at 293 K. location of T 1 and the local structure in the blend through the zero-field splitting (ZFS) parameters of the spin Hamiltonian, but also on the T 1 formation mechanism through the spin-polarization of the signal. [60,62] In trEPR spectroscopy, the spin-polarization results from non-Boltzmann population of the triplet sublevels, which manifests as a characteristic polarization pattern of absorptive (a) and emissive (e) microwave-induced EPR transitions between the three triplet sublevels. For example, direct ISC mediated by spin-orbit coupling (SOC) from S 1 results in a spin selective population of the zerofield triplet sublevels (Table S2, Supporting Information), [63] which in turn is converted into a polarized population of the three high-field triplet sublevels T + , T 0 , and T -. By applying microwave irradiation, transitions between these sublevels result in an aaaeee, eeeaaa, eeaeaa, aaeaee, aeaeae or eaeaea polarization pattern. In contrast, T 1 created through the geminate BCT mechanism will possess a characteristic aeeaae or eaaeea signature. This is because ISC in the CT state manifold, which is primarily mediated by the HFI with paramagnetic nuclei (mainly protons), results in an overpopulation of the T 0 sublevel of the 3 CT state and thus also the molecular T 1 state following BCT. [27,53] In contrast, nongeminate recombination does not produce spin polarization as the spin-statistical recombination of uncorrelated free charge carriers to T 1 via 3 CT results in an equal population of the T + , T 0 , and Tsublevels. The equal sublevel population will establish Boltzmann population within the spin-lattice relaxation time, but in the lowfield (≈330 mT) regime that we explore here with X-band EPR, the spin polarization is too low to be detected with trEPR. [62] Therefore, the nongeminate BCT pathway does not induce a sufficiently high spin-polarization in T 1 , and states populated through this mechanism are not observed in trEPR. Thus, trEPR provides an excellent complement to TA spectroscopy, in which only the nongeminate pathway can be reliably detected.

trEPR of PM6:PC 60 BM
In Figure 3a,b, we show the trEPR spectra and associated simulations of the PM6:PC 60 BM film taken at two representative time points (1 and 5 µs) after excitation at 532 nm. A summary of the best fit simulation parameters for the blends studied is included in Table 1, with more detailed information on all samples in Table S2 in the Supporting Information. At 1 µs, we observe an intense and spectrally narrow aeae feature centered at ≈346 mT, assigned to CT states; by 5 µs, this evolves into a pure a signal, indicative of free charges (polarons). [61] Conversely, the broader signal between 290 and 410 mT is assigned to T 1 states. From the best-fit spectral simulation at 1 µs, an eeeaaa polarization pattern is obtained, confirming that the T 1 states are formed via SOC-ISC; this results from photogenerated S 1 states that do not undergo charge transfer at the donor:acceptor interface. The T 1 ZFS parameters, D = 1300 MHz and E = 140 MHz, are comparable to the those obtained at 1 µs in a neat PM6 film ( Figure S4, Supporting Information) and are significantly larger than those found for a neat PC 60  Therefore, we assign this feature to the T 1 of PM6. By 5 µs, there is a clear evolution of the polarization pattern, and a more complex spectrum is observed. We attempt to simulate this new spectrum using a single SOC-ISC component but find through examining the residual that this species alone is not sufficient to obtain a high-quality fit ( Figure S6, Supporting Information). An excellent fit is only obtained when two T 1 species (D = 1220 MHz and E = 40 MHz) are included in the simulation with distinct aeaeae and eaaeea polarization patterns. The aeaeae species is the same as the PM6 T 1 formed via SOC-ISC at 1 µs, with the apparent spectral inversion attributed to an unequal rate of decay from the three high-field triplet sublevels. [64] However, the new eaaeea contribution represents T 1 formed via the geminate BCT mechanism, confirming that this pathway is also present in the PM6:PC 60 BM blend.

trEPR of PTB7-Th:PC 60 BM
We now turn to the PTB7-Th:PC 60 BM blend film (Figure 3c,d). At both 1 and 5 µs, this blend shows a strong ea feature at ≈346 mT, which is assigned to CT states. [61] In addition, we also observe a signal between 300 and 390 mT at both 1 and 5 µs; as the spectrum is significantly broader than the PC 60 BM T 1 , we attribute this feature to T 1 states on PTB7-Th. We verify this assignment through comparison of the ZFS parameters obtained from simulations of the neat PTB7-Th ( Figure S7, Supporting Information) and PTB7-Th:PC 60 BM blend films (Table S1, Supporting Information). The T 1 spectra cannot be well-described by a best-fit simulation with single aaaeee SOC-ISC species ( Figure S8, Supporting Information), confirming that there is more than one T 1 generation mechanism present. Therefore, we have simulated the T 1 spectra at 1 and 5 µs with two species (see Table S1 in the Supporting Information); one with an aaaeee and the other with an eaaeea polarization pattern, representing T 1 states formed via SOC-ISC and the geminate BCT mechanism, respectively. Thus, we are able to definitively confirm that geminate BCT T 1 formation is also occurring in the PTB7-Th:PC 60 BM.

Discussion of the trEPR Results
When comparing the blends, we observe the presence of geminate BCT T 1 and CT states at both 1 and 5 µs in PTB7-Th:PC 60 BM. In contrast, we only see BCT T 1 states at 5 µs in PM6:PC 60 BM and CT states at 1 µs; by 5 µs, the CT states have evolved into free charges in this blend. To better understand these results, we first consider that the time resolution of our trEPR measurements (hundreds of nanoseconds) is too slow to directly monitor the rapid separation of most CT states in an efficient polymer:fullerene blend, which largely takes place on sub-picosecond timescales. [65] Thus, what we observe in trEPR are only the longer lived (trapped) CT states responsible for the geminate BCT T 1 formation mechanism; as discussed above, CT states formed via spin-statistical nongeminate recombination are not detected in trEPR. In OSCs, CT states can be discussed in the context of the spin-correlated radical pair model. [36] Spin-correlated radical pairs are characterised by four eigenstates that in the high field approximation are two mixed singlet-triplet levels with M S = 0, | CT  The trEPR spectrum of a PM6:PC 60 BM blend film at 5 µs, excited at 532 nm. The aeae CT state has now evolved into a single a signal, indicative of free charges. The PM6 T 1 species requires the use of two T 1 contributions to successfully describe the observed spectrum. The first is an aeaeae component, which is the same ISC T 1 state as the 1 µs spectrum, except inverted. The second is an eaaeea component, which is characteristic of T 1 states formed via the geminate BCT mechanism. c) The trEPR spectrum of a PTB7-Th:PC 60 BM blend film at 1 µs, excited at 532 nm. The central ea species is assigned to a CT state. The PTB7-Th T 1 species requires the use of two T 1 contributions to successfully describe the observed spectrum. The first is an aaaeee component, indicating T 1 states formed via direct ISC from undissociated S 1 states. The second is an eaaeea component, which is characteristic of T 1 states formed via the geminate BCT mechanism. d) The trEPR spectrum of a PTB7-Th:PC 60 BM blend film at 5 µs, excited at 532 nm. The ea CT state is still present, indicating slower generation of free charges in this blend. The PTB7-Th T 1 species requires the use of two T 1 contributions to successfully describe the observed spectrum. The first is an aaaeee component, indicating T 1 states formed via direct ISC from undissociated S 1 states. The second is an eaaeea component, which is characteristic of T 1 states formed via the geminate BCT mechanism. All trEPR spectra were acquired at 80 K.  [53,66] Owing to the energetic splitting due to the Zeeman interaction, mixing can only occur between | CT 1 0 and | CT 3 0 . [67] The spin-mixing rate in the studied blends is primarily controlled by the HFI and results in a periodic oscillation of the M S = 0 population between | CT 1 0 and | CT 3 0 on timescales of tens of nanoseconds. [68] Thus, as the spin mixing timescales are faster than the time resolution of our trEPR measurement, the CT states observed can be considered as an admixture of | CT 1 0 and | CT 3 0 in both the PM6:PC 60 BM and PTB7-Th:PC 60 BM blends. Having clarified this point, we turn to the discussion of our trEPR spectra. To explain why the BCT triplet only appears at 5 µs in PM6:PC 60 BM, we propose that the eventual appearance of free charges in this blend suggests that the CT states are on average more weakly bound than in PTB7-Th:PC 60 BM. This is supported by the PLDMR of the blends (vide infra), where the narrow and negative sign signals arising from free charges (positive and negative polarons) are much more intense in PM6:PC 60 BM ( Figure S15, Supporting Information), indicating that a higher proportion of the electrons and holes are far enough apart to no longer be magnetically interacting. A more weakly bound CT state in PM6:PC 60 BM with a larger average electron-hole separation could result in slower BCT for two (related) reasons. First, the BCT rate depends strongly on distance, [20] meaning a 3 CT state with greater electron-hole separation would recombine to T 1 more slowly. Second, the charges in a more separated 3 CT would be expected to approach the donor:acceptor interface, where BCT is fastest, less frequently, also slowing the formation of BCT T 1 states. Thus, these two factors can explain the slower appearance of the geminate BCT T 1 states in PM6:PC 60 BM.

Comparison of Fullerene and Nonfullerene Acceptor Systems
We note that the presence of triplet excitons generated by geminate BCT in fullerene acceptor blends, as reported here by us and others, [23,43,52,53] is in clear contrast to the results obtained for the NFA blends. [20] This can be rationalized by the widely observed propensity of fullerene molecules to intercalate within the alkyl side chains of the donor polymer, resulting in the formation of mixed polymer/fullerene regions. [69][70][71][72][73] Of particular relevance to our work, observations of fullerene intercalation within the polymer sidechains have also been made in PC 60 BM blends with polymers from the PTB7 series. [74] We note that efficient solar cell operation is obtained when there is excess fullerene that forms local 5-10 nm fullerene inclusions, driving charge separation from mixed fullerene/sidechain regions to the pure fullerene regions. [72,[75][76][77] However, if the fullerene concentration in these mixed regions falls below the percolation threshold for efficient electron transport, charge separation will be impeded and geminate recombination will result. [75][76][77][78][79] Thus, we propose that poorly-connected fullerene domains provide the opportunity for geminate 3 CT formation via HFI-ISC from 1 CT states on nanosecond timescales, [56,80] followed by BCT to T 1 , increasing losses via T 1 states. In contrast, many efficient NFA OSCs have been shown to possess good phase purity, [81][82][83][84] which has previously been shown to facilitate CT state dissociation and reduce BCT T 1 formation. [85] Thus, it appears that engineering good phase purity in the donor:acceptor bulk heterojunction could be helpful for engineering out BCT T 1 generation pathways in OSCs.

Using PLDMR Spectroscopy to Detect Triplet States in Organic Solar Cells
In contrast to TA and trEPR, PLDMR spectroscopy is generally employed as a steady-state technique when investigating organic semiconductors, [86] though transient iterations with a time resolution on the order of tens of nanoseconds are available. [87] Continuous illumination can also yield spin polarization of the triplet sublevels by unequal recombination rates or triplet accumulation, with subsequent triplet-triplet annihilation processes. Furthermore, while trEPR directly detects reflection changes in the applied microwaves following laser excitation, PLDMR uses optical detection. [30,44,45,86,88] Therefore, the experimental sensitivity is greatly enhanced and PLDMR can detect any triplet excitations that are coupled to the photoluminescence of the sample (for example, via the triplettriplet annihilation of two molecular T 1 which reforms, among other possibilities, one bright S 1 state and one dark spin-singlet ground state), [86] not just those that are highly spin polarized. In addition, the enhanced sensitivity of PLDMR allows for "half-field" (HF) signals to be readily resolved; HF transitions represent a first-order forbidden Δm S = ±2 transition between the T + and Tsublevels that is enabled by the strong dipolar interaction between the two localized electron spins of a molecular T 1 state. [30,44] These signals provide an additional tool for determining the molecular location of a T 1 state, since their magnetic field position depends on the strength of the dipolar interaction. [89] Therefore, HF signals are particularly useful when working with systems where the ZFS parameters of the donor and acceptor T 1 states are similar, which is often the case in NFA OSC blends. [20] In Figure 4 (enlarged individual spectra are presented in Figures S9-S13 in the Supporting Information), we display the PLDMR spectra of neat films of PC 60 BM (red), PTB7-Th (light green), PM6 (light blue) and the PTB7-Th:PC 60 BM (dark green) and PM6:PC 60 BM (dark blue) blends. A summary of the simulation parameters used for the films is shown in Table 1, with the full spectral simulations and ZFS parameters for each sample provided in Figure S15 and Table S3 in the Supporting Information. The spectra consist of a full-field (FF) region (260-410 mT), corresponding to Δm S = ±1 transitions, and Δm S = ±2 HF signals (160-172 mT).

PLDMR of PC 60 BM
Beginning with neat PC 60 BM (Figure 4, red), we observe a relatively narrow T 1 feature in the FF spectrum between 320-350 mT, which can be described with the ZFS parameters D = 360 MHz and E = 50 MHz. Two additional features (sharp negative signals) are superimposed on the T 1 signal at 336.25 mT (g = 2.0040) and 336.65 mT (g = 2.0012), seen more clearly in Figure S9c in the Supporting Information. We assign these to the anion PC 60 BM − , as already known from literature, and the cation PC 60 BM + , respectively. [90][91][92] In addition, a HF signal due to PC 60 BM T 1 states is also detected at 168.1 mT, which can be easily distinguished from the HF signals of the polymer in the blends.

PLDMR of PTB7-Th Systems
Neat PTB7-Th (Figure 4, light green) shows a T 1 spectrum with large ordering factor, λ, which provides information on the orientational distribution of the molecules in the sample and is reflected by the outer "wings" in the PLDMR spectra. [62,93,94] In contrast to trEPR, ordering factors are only visible in PLDMR for these materials, further discussed in the Supporting Information. The ordering factor λ is given for θ and φ, where θ is the angle between the molecular z-axis and the applied magnetic field, and φ is the in-plane angle. If λ is zero, all molecular orientations occur with the same probability. For PTB7-Th, the ordering for θ is λ θ = 11, corresponding to an extremely narrow orientational distribution of the molecules in the direction of the applied magnetic field ( Figure S16, Supporting Information). Upon blending PTB7-Th with PC 60 BM (Figure 4, dark green), a broad T 1 feature of the polymer is clearly visible in the FF and HF signals of the blend. However, the ZFS D-parameter increases from D = 1190 MHz in pristine PTB7-Th to D = 1310 MHz in the blend, while the ordering factor decreases to λ θ = 7.5. The change in these values suggest that blending PTB7-Th with PC 60 BM affects the polymer chain ordering in the PTB7-Th domains, possibly due to fullerene intercalation within the polymer sidechains. [74]

PLDMR of PM6 Systems
For PM6 (Figure 4, light blue), we observe a broad T 1 spectrum between 280 and 390 mT, corresponding to ZFS parameters D = 1500 MHz and E = 70 MHz. In contrast to trEPR, this spectrum also shows a considerable ordering factor (Table S3, Supporting Information). When blending PM6 with PC 60 BM (Figure 4, dark blue), the broader polymer triplet is again clearly visible in the FF and HF signals. In contrast to PTB7-Th:PC 60 BM, the PM6 T 1 ZFS parameters remain the same before and after blending with PC 60 BM and the ordering factors change only slightly (Table S3, Supporting Information). Thus, the ordering of the polymer chains in PM6 is less disrupted upon mixing with PC 60 BM, when compared to PTB7-Th. Additionally, both blends show a positive (ΔPL/PL > 0) CT state peak with two negative signals at g = 2.0012 and g = 2.0037 in PTB7-Th:PC 60 BM, and g = 2.0012 and g = 2.0034 in PM6:PC 60 BM. The lower g value is identical to the negative polaron PC 60 BM − , as detected in neat PC 60 BM ( Figure S9c, Supporting Information), while the higher g value likely represents positive polarons on the respective polymer. [95]

Detecting Minority Triplet Formation Pathways with PLDMR
In contrast to trEPR, the additional experimental sensitivity of PLDMR enables us to resolve the weaker PC 60 BM T 1 signal in the HF (168.1 mT) regions in both blends (Figures S11a and S13a, Supporting Information) and in the FF (320-350 mT) region of PTB7-Th: PC 60 BM ( Figure S11b, Supporting Information); only the more intense T 1 features of the donor polymers are seen in the corresponding trEPR data (Figure 3). Thus, without PLDMR, minority T 1 generation pathways could easily be missed. When considering the formation mechanism for PC 60 BM T 1 states in the blends, we note that they are generally too high in energy (T 1 ≈ 1.5 eV) to be populated by BCT from the CT states in low band gap polymer:fullerene systems with large S 1 -CT energetic offsets (as is the case in the blends studied here). [34] Furthermore, as the PC 60 BM T 1 will be higher in energy than the T 1 states of PM6 and PTB7-Th, assuming a typical S 1 -T 1 energy gap of ≈0.6-1 eV in most conjugated polymers, [39] any PC 60 BM T 1 states formed near the donor:acceptor interface would be expected to relax into the lower lying polymer T 1 state. Therefore, we conclude that the PC 60 BM T 1 states observed must be located in isolated PC 60 BM domains, which are often found in polymer:fullerene bulk heterojunction blends. [74,76] This observation reinforces the importance of ensuring that domain sizes are on the order of the singlet exciton diffusion length to enable efficient charge generation and suppressed T 1 formation via direct SOC-ISC from undissociated S 1 states. [96]

Conclusion
In this work, we have shown that through leveraging the strengths of TA, trEPR, and PLDMR spectroscopies, a complete picture of T 1 generation pathways in OSCs can be obtained. This strategy is fully applicable to both fullerene and NFA blends and will prove valuable in the task of engineering out of recombination via T 1 states in OSCs. Through applying this methodology to two model OSC blends, we have demonstrated that it is possible to unravel the intricacies of spintriplet physics in OSCs by identifying both the molecular localization and generation mechanism for the wide range of T 1 states found in these systems. Consequently, we have shown that fullerene blends frequently exhibit the geminate BCT T 1 formation mechanism, [23,43,52,53] in addition to the nongeminate pathway. This is in clear contrast to the more efficient NFA OSC systems where the geminate pathway appears to be absent, consistent with their improved performance. [20] We propose that the geminate BCT mechanism is associated with isolated fullerene molecules trapped in alkyl side chains of the donor polymers, [69][70][71][72][73] suggesting that engineering good purity in the donor and acceptor phases is key for suppressing this process. Thus, we anticipate that this framework will also be particularly useful for analyzing T 1 loss mechanisms in ternary systems comprised of both fullerene and NFAs, which have demonstrated some of the highest PCEs to date. [40,41]

Experimental Section
OSC Device Fabrication: Indium tin oxide (ITO) patterned glass substrates were cleaned by scrubbing with soapy water, followed by sonication in soapy water, deionized (DI) water, acetone, and isopropanol for 20 minutes each. The substrates were dried using compressed nitrogen and placed in an oven overnight at 100°C. The conventional architecture devices were made by treating the ITO substrates with UV-ozone for 15 minutes and spin-coating a layer of poly(3,4-ethylenedio xythiophene):poly(styrenesulfonate) (PEDOT:PSS, Clevios P VP Al 8043) at 3000 rpm for 40 s onto the ITO substrates in air. The substrates were then annealed in air at 150 °C for 20 minutes. 1 mg/mL of the cathode interlayer PDINO was dissolved in methanol and spin-coated on top of the active layer. The substrates were then pumped down under vacuum (<10 −7 torr), and a 100 nm thick Al electrode was deposited on top by thermal evaporation using the Angstrom Engineering Series EQ Thermal Evaporator. In the case of inverted architecture devices, ZnO was used as the bottom transparent electrode (replacing PEDOT:PSS), where the ZnO solution was prepared in a nitrogen glovebox by mixing tetrahydrofuran and diethylzinc (2:1). The fresh ZnO solution was then spin-coated atop the clean ITO substrates at 4000 rpm for 30 seconds and then placed on a hotplate at 110 °C for 15 minutes. Following active layer spin-coating, the inverted devices were pumped down under vacuum (<10 −7 torr), and 7 nm of MoO x and 100 nm thick Ag electrode were deposited on top of the active layer by thermal evaporation. The electrode overlap area was 0.22 cm 2 for both conventional and inverted devices. The active area of the device was determined using an optical microscope.
OSC Device Testing: Photovoltaic characteristic measurements were carried out inside a N 2 filled glove box. Solar-cell device properties were measured under illumination by a simulated 100 mW cm −2 AM1.5 G light source using a 300 W Xe arc lamp with an AM 1.5 global filter. The irradiance was adjusted to 1 Sun with a standard silicon photovoltaic cell calibrated by the National Renewable Energy Laboratory. No spectral mismatch correction was applied. A Keithley 2635A source measurement unit was used to scan the voltage applied to the solar cell between -2 to 1 V at a speed of 0.43 V/s with a dwell time of 46 ms. Scans were performed in both the forward and reverse directions, with no unusual behavior observed. Between eight and 30 individual solar cell devices were tested for each blend reported. The error associated with the reported PCE values is ±0.2%. External quantum efficiency (EQE) for the OSCs was measured using a 75 W Xe light source, monochromator, optical chopper (138 Hz), and a lock-in amplifier. Power-density calibration of the EQE characteristics was achieved using a calibrated silicon photodiode from Newport.
Photoluminescence Quantum Efficiency Measurements: The PLQE was determined using method previously described by De Mello et al.. [97] Samples were placed in an integrating sphere and photoexcited using a 658 nm continuous-wave laser. The laser and emission signals were measured and quantified using calibrated Andor iDus DU420A BVF Si and Andor CCD-1430 InGaAs detectors.
TA Spectroscopy: TA was performed on a setup powered using a commercially available Ti:sapphire amplifier (Spectra Physics Solstice Ace). The amplifier operates at 1 kHz and generates 100 fs pulses centered at 800 nm with an output of 7 W. A TOPAS optical parametric amplifier (OPA) was used to provide the tunable ≈100 fs pump pulses. The probe was provided by a broadband IR noncollinear optical parametric amplifier (NOPA). The probe pulses are collected with an InGaAs dual-line array detector (Hamamatsu G11608-512DA), driven and read out by a custom-built board from Stresing Entwicklungsbüro. The probe beam was split into two identical beams by a 50/50 beamsplitter. This allowed for the use of a second reference beam which also passes through the sample but does not interact with the pump. The role of the reference was to correct for any shot-to-shot fluctuations in the probe that would otherwise greatly increase the structured noise in our experiments. Through this arrangement, very small signals with a T T ∆ = 1 × 10 −5 could be measured.
trEPR Spectroscopy: EPR samples were fabricated by spin-coating solutions under identical conditions to the optimized devices onto Mylar substrates, which were subsequently cut into strips with a width of 3 mm. To ensure the flexible Mylar substrates did not bend during the spin coating process, they were mounted onto rigid glass substrates using adhesive tape. The strips were placed in quartz EPR tubes which were sealed in a nitrogen glovebox with a bi-component resin (Devcon 5-Minute Epoxy), ensuring that all EPR measurements were performed without oxygen exposure.
All trEPR spectra were recorded on a Bruker Elexsys E580 X-band spectrometer, equipped with a nitrogen gas-flow cryostat for sample temperature control. The sample temperature was maintained with an Oxford Instruments CF935O cryostat and controlled with an Oxford Instruments ITC503. Laser pulses for trEPR were collimated into the cryostat and resonator windows from a multi-mode optical fiber, ThorLabs FT600UMT. Sample excitation at 532 nm with an energy of 2 mJ per pulse and a duration of 7 ns was provided by the residual 2 nd harmonic output of a Newport/Spectra Physics Lab 170 Quanta Ray Nd:YAG pulsed laser, operating at 20 Hz. The trEPR signal was recorded through a Bruker SpecJet II transient recorder with timing synchronization by a Stanford Research Systems DG645 delay generator. The instrument response time was about 200 ns. The spectra were acquired with 2 mW microwave power and averaging 400 transient signals at each field position.
From the datasets obtained, the transient EPR spectra at different time delays after the laser pulse have been extracted and averaged over a time window of 0.5 µs. Spectral simulations have been performed using the core functions pepper and esfit of the open-source MATLAB toolbox EasySpin. [94] The parameters included in our best-fit simulations are the ZFS parameters (D and E), the triplet population sublevels (p 1 , p 2 , p 3 ) and the line broadening (assumed as only Lorentzian to not over-parametrize the fitting). For the calculation of spin polarization, the populations of the spin-triplet sublevels at zero field were calculated (T x , T y , T z ) in the fitting program and used by EasySpin to simulate the trEPR spectrum at resonant fields. For all the simulations, the g tensor was assumed isotropic with g iso = 2.002. To carry out our leastsquare fittings, a user-defined simulation function has been developed which allows the fitting of parameters, such as the spin populations of the triplet sublevels. All the fits were carried out using a Nelder/Mead downhill simplex optimization algorithm.
PLDMR Spectroscopy: PLDMR samples were prepared in the same way as the trEPR samples, whereby the EPR tubes were sealed under inert helium atmosphere to measure at cryogenic (10 K) temperatures. PLDMR measurements were carried out in a modified X-band spectrometer (Bruker E300) equipped with a continuous-flow helium cryostat (Oxford ESR 900) and a microwave cavity (Bruker ER4104OR, 9.43 GHz) with optical access. Optical irradiation was performed with a 473 nm continuous wave laser (Cobolt) from one side opening of the cavity. PL was detected with a silicon photodiode (Hamamatsu Si photodiode S2281) on the opposite opening of the cavity, using a 561 nm longpass filter to reject the excitation wavelength. The PL signal was amplified by a current/voltage amplifier (Femto DHPCA-100) and recorded by lock-in detector (Ametek SR 7230), referenced by TTLmodulating the microwaves with 517 Hz. Microwaves were generated with a microwave signal generator (Anritsu MG3694C), amplified to 3W (Microsemi), and guided into the cavity.

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