Role of the energy o ﬀ set in the charge photogeneration and voltage loss of nonfullerene acceptor-based organic solar cells †

The trade-o ﬀ between short-circuit current density ( J SC ) and open-circuit voltage ( V OC ) has been one of the largest challenges in improving the power conversion e ﬃ ciencies (PCEs) of organic solar cells (OSCs). Although the energy o ﬀ set between the excited and charge transfer (CT) states should remain minimal to achieve a high V OC , a very small energy o ﬀ set typically leads to degradation of J SC , even when novel nonfullerene acceptors (NFAs), such as Y6, are used. Therefore, understanding the limit to what extent the energy o ﬀ set can be minimized and the physics underlying the trade-o ﬀ relationship is important to optimize the design of new materials and further improve the PCEs. This study provides a threshold energy that can ensure high charge photogeneration quantum e ﬃ ciencies for Y-series NFA-based OSCs and discusses the role of the energy o ﬀ set in device performances. We found that an insu ﬃ cient energy o ﬀ set led to not only slow hole transfer at the donor:acceptor interfaces, but also ine ﬃ cient long-range spatial dissociation of the CT states and degradation of the ﬁ ll factor (FF). This study also discusses the interplay of the energy levels of the two NFAs that constitute ternary blend OSCs. We found that, by introducing a low-e ﬃ ciency NFA into a high-e ﬃ ciency donor:acceptor blend, the voltage loss can be reduced while maintaining a high charge photogeneration quantum e ﬃ ciency. Our ﬁ ndings highlight the importance of overcoming the trade-o ﬀ between FF and V OC for further improving the PCE.


Introduction
2][3][4][5][6][7] Nevertheless, a large voltage loss DV, which is dened as the difference between the optical bandgap E g and open-circuit voltage V OC (DV = E g /q − V OC , where q is the elementary charge), is a signicant disadvantage of OSCs, restricting further improvement in the PCEs of OSCs. 8,9Although DV has continuously decreased in the past decade, state-of-the-art OSCs still exhibit DVs of ∼0.5 V or more, which remains considerably larger than those of their inorganic and perovskite counterparts, where DV of less than 0.4 V can be achieved. 10arge voltage losses in OSCs primarily originate from the following two sources.The rst source is the voltage loss incurred during the charge generation process as OSCs require donor:acceptor (D:A) heterointerfaces to split excitons into holes on the donors and electrons on the acceptors.2][13][14] In contrast, recent studies have shown that efficient charge separation with an energy offset of less than 0.3 eV can be achieved for various nonfullerene acceptor (NFA)-Yasunari Tamai received his PhD degree from Kyoto University in 2013 in the excited-state dynamics in nanostructured polymer systems.He joined the Optoelectronics group at the University of Cambridge as a postdoctoral fellow under the supervision of Prof. Sir Richard Friend, where his research focused on the ultrafast charge separation at organic semiconductor heterojunctions.Since 2016, he has been an Assistant Professor at Kyoto University.From 2018 to 2022, he was also a JST PRESTO researcher.His current research interests include exciton and charge dynamics in organic semiconductors, particularly conjugated polymers.
based devices.We have revealed that a representative highefficiency NFA-based OSC consisting of PBDB-T-2F (also referred to as PM6) and Y6 as an electron donor and acceptor, respectively, exhibits a near-unity and temperature-independent charge separation efficiency despite a small energy offset of ∼0.12 eV. 15 However, to what extent the energy offset can be minimized, while maintaining a high charge photogeneration quantum efficiency is unclear.8][19] In fact, an OSC consisting of PBDB-T-2F paired with Y5 exhibited a poor photovoltaic external quantum efficiency (EQE PV ) of 36.1%, whereas DV of this device was ∼80 mV smaller than that of the PBDB-T-2F:Y6 device owing to the smaller energy offset of the former. 20he other source for the large DVs in OSCs is because of the voltage loss incurred during the charge recombination process.3][24][25][26] Conventional fullerene-based OSCs typically exhibit DV nr s of approximately 0.4 V, 25 which are substantially larger than those of their inorganic and perovskite counterparts.The origin of the large DV nr can be rationalized by the extremely low photoluminescence quantum yields (PLQYs) of the CT states.When a free carrier encounters an opposite charge at the D:A interface, a CT state is regenerated at the interface followed by deactivation to the ground state.As CT states predominantly decay nonradiatively due to their signicantly small oscillator strengths, charge recombination in fullerene-based OSCs leads to a large DV nr s.This is also valid for NFA-based OSCs.For instance, the PBDB-T-2F:Y6 device exhibits a relatively large DV nr of ∼0.25 V. 20 Recent studies have highlighted the importance of reducing the energy offset in minimizing not only the voltage loss incurred during charge generation, but also the DV nr . 17,20,27,28We have demonstrated that the DV nr s of OSCs continuously decreased with a decrease in energy offset, and the abovementioned PBDB-T-2F:Y5 device exhibited a small DV nr of 0.145 V while the PCE of this device was poor due to the low EQE PV . 17,20n the other hand, some recent reports have claimed that self-ionization of excitons without a heterojunction occurs in Y6 pristine lms, 29,30 and hence, donor polymers, such as PBDB-T-2F, only serve as hole transport materials to suppress the bimolecular charge recombination.If this is true and is an intrinsic property of the Y-series NFAs, the low EQE PV of the abovementioned PBDB-T-2F:Y5 device may stem from other processes than charge separation.Therefore, again, it is crucial to unveil whether a D:A interface with a certain amount of energy offset is necessary for efficient charge separation and to what extent the energy offset can be minimized.
The ternary blend concept has been widely applied to OSCs to enhance the PCEs.2][3][4][5][6] In ternary blend systems, upon changing the blend ratio of the two NFAs, V OC continuously varies between V OC s of the corresponding binary reference systems. 31,32In addition, several studies have shown that when the two NFAs form co-crystals, the ionization energies (IEs) of the co-crystals can be continuously tuned between the IEs of the two NFAs, enabling precise control of the energy offset. 33owever, the role of the energy offsets of corresponding binary reference systems in the charge photogeneration and voltage loss has not been fully elucidated.
Herein, we studied the relationship between the energy offset and EQE PV of NFA-based OSCs to discuss to what extent the energy offset can be minimized.We used PBDB-T-based donor polymers and Y-series NFAs with different highest occupied molecular orbital (HOMO) energies (Fig. 1) to ensure similar photon absorption efficiencies and E g s with minimal difference in their chemical structures.This allowed us to focus on the impacts of energy offsets on the charge photogeneration and voltage losses.We prepared OSCs using 24 D:A combinations and examined the corresponding EQE PV s and voltage losses.The EQE PV sharply decreased when the voltage loss was less than 0.52 V. We found that an insufficient energy offset led to slow charge transfer at the D:A interfaces, as has been expected by Marcus theory.Surprisingly, the insufficient energy offset also resulted in poor long-range spatial dissociation of the CT states.This is contrary to the expectations derived from recent studies, 34 wherein the band bending near the D:A interface driven by the large quadrupole moments of NFAs was supposed to accelerate long-range charge dissociation once the CT states were formed, resulting in barrier-less efficient charge dissociation.Degradation of the ll factor (FF) with a decrease in the energy offset was also observed because of the inefficient charge dissociation.We also discuss the interplay of the energy levels of the two NFAs that constitute ternary blend OSCs.We found that combining two NFAs enables us to reduce the voltage loss overcoming the abovementioned threshold, while maintaining a high charge photogeneration quantum efficiency.

Materials and ionization energies
We used three PBDB-T-based donors paired with eight Y-series acceptors (the chemical structure of the materials employed in this study can be found in Fig. 1), resulting in 24 combinations of D:A blends.We stress that the E g s and absorption properties are similar among all the devices (steady-state absorption spectra are provided in Fig. S1, ESI †), allowing us to focus on the role of energy offset in the charge photogeneration and voltage loss.In contrast, the HOMO energy difference between the donor and acceptor can be substantially tuned with minimal differences in the chemical structure and associated changes in physical properties.
The HOMO energies (or IEs) of organic semiconductors are usually evaluated by photoelectron spectroscopic techniques, such as ultraviolet photoelectron spectroscopy (UPS) and photoelectron yield spectroscopy in air (PYSA), or by electrochemical techniques, such as cyclic voltammetry (CV).Recent studies have pointed out that the HOMO energy differences between the donor and NFA determined by the photoelectron spectroscopies were signicantly larger than those determined by CV. [34][35][36] For instance, the HOMO energy difference between PBDB-T-2F and Y6 determined using CV was only ∼0.1 eV, 35 whereas the IE differences of the same blend determined by UPS and PYSA were >0.5 eV. 34,36The origin of the large discrepancy between these was likely due to differences in sample morphology.Photoelectron spectroscopies probe the IEs of solid states (especially, near the surface).In contrast, materials used for CV measurements are more disordered than those used for photoelectron spectroscopies even when lm samples are applied to CV measurements because the lm is swollen by solvents or sometimes dissolves into solvents during measurements.This leads to large variations in the contributions caused by charge-permanent multipole (mainly quadrupole) interactions to the observed HOMO (IE), 34,37 resulting in a large discrepancy in the measured values.Therefore, photoelectron spectroscopies are considered more reliable for determining the IEs of the "pure" donor and acceptor materials in the solid state; hence, this study determined the IE differences between the donor and acceptor materials in their pristine lm state using PYSA (DIE PYSA ).In contrast, it should be emphasized that the actual D:A interfaces in bulk heterojunction blends are more disordered than pristine lms.Therefore, the actual HOMO differences at the D:A interface likely differ from those determined for pristine lms using photoelectron spectroscopies because the charge-permanent multipole interactions are supposedly weaker in the interfacial D:A mixed region.In fact, time-resolved spectroscopic techniques have shown that the energy offset between E g and E CT of the PBDB-T-2F:Y6 blend is only ∼0.12 eV. 15 This value is considerably smaller than the abovementioned IE difference determined using photoelectron spectroscopies, but similar to the value determined by CV measurements.Therefore, DIE PYSA should be signicantly larger than the actual HOMO energy difference at the D:A interface.Nevertheless, DIE PYSA remains an effective quantitative measure of the energy offset.Fig. 1e shows the IEs of materials employed in this study determined by PYSA (experimental data can be found in Fig. S2, ESI †).These values are consistent with those reported previously with minor variations (see Fig. S3 † for more details). 36Note that the lowest unoccupied molecular orbital (LUMO) energies were estimated using IEs and excited state energies (Fig. S4, ESI †); thereby, these values only serve as a rough estimate for relative comparison.The LUMO energy differences between the donor and acceptor materials were sufficiently large.However, because efficient energy transfer from the donor to the acceptor occurs, the LUMO energy difference was not related to device performances (vide infra).

Trade-off between charge generation and voltage loss
Fig. 2a and b show the current density-voltage (J-V) characteristics and EQE PV spectra of four representative solar cell devices employed in this study (the experimental results and photovoltaic device parameters for all devices can be found in Fig. S5, S6 and Tables S4-S6, ESI †).We found a clear trade-off relationship between short-circuit current density (J SC ) and V OC ; V OC increased with a decrease in J SC .EQE PV also decreased with an increase in V OC .These results suggest that the charge separation efficiency decreases with a decrease in the energy offset (i.e., a decrease in DIE PYSA ), as was observed for the conventional fullerene-based OSCs. 13,38Interestingly, the EQE PV spectra followed well with the absorption spectra, indicating that the internal quantum efficiencies (IQEs) of these devices were less sensitive to the excitation wavelength despite large LUMO energy differences between the donor and acceptor.This is also conrmed using Fig. 2c, wherein the maximum EQE PV in the near-IR region (>650 nm), which corresponds to the acceptor absorption region, is plotted against the maximum EQE PV over the entire wavelength region (most devices showed the maximum EQE PV in the visible region).Owing to the large spectral overlap between donor uorescence and acceptor absorption, energy transfer from the donor to the acceptor is likely to outcompete the electron transfer (Fig. S7, ESI †) 39 This leads to predominant acceptor exciton formation regardless of the excitation wavelength, rendering EQE PV insensitive to the excitation wavelength.Therefore, the maximum EQE PV over the entire wavelength region is plotted against DV.Using the maximum EQE PV value allows us to minimize the inuence of imperfect photon absorption.As shown in Fig. 2d, EQE PV drops sharply when DV is less than 0.52 V.The maximum EQE PV is also plotted against the IE difference DIE PYSA (Fig. S8, ESI †), where we found that the threshold value of DIE PYSA that ensured efficient charge photogeneration was ∼0.42 eV.We underline that the DIE PYSA used here, which was determined by the PYSA measurements for the pristine thin lms, is considerably larger than the energy offset between E g and E CT owing to the contribution caused by charge-permanent multipole (mainly quadrupole) interactions, as mentioned above.
A similar trend was also observed for fullerene-based OSCs.However, the threshold value obtained in this study is lower than that observed for fullerene-based OSCs (see Fig. S10, ESI † for more details), 13,38 indicating that Y-series NFAs can achieve an efficient charge photogeneration with a lower threshold voltage loss.
To obtain deeper insights, V OC was divided into two parts, as follows: where V rad OC is the radiative limit of V OC , wherein charge recombination is always accompanied by photon emission (i.e., the maximum achievable V OC when the QY of radiative charge Fig. 2 (a) J-V characteristics and (b) EQE PV spectra of the four representative devices.(c) Maximum EQE PV s in the >650 nm region plotted against those over the entire wavelength region.(d) Maximum EQE PV over the entire wavelength region plotted against DV (= E g /q − V OC ).E g was determined from the EQE PV spectra using the method proposed earlier, 24 wherein the first derivative of the EQE PV spectrum is assumed to be a probability distribution function of the photovoltaic bandgap energy, and the mean value of the distribution is used for E g .The shape and color of the legends in (c) and (d) identify donors and acceptors, respectively.
recombination is unity).Tables S7 and S8 † summarize V rad OC and DV nr for all devices (see the ESI † for the details of the determination procedure).As the CT absorption was buried under the smeared-out absorption edge of acceptors for all devices owing to the small energy offset, V rad OC was largely determined by E g , and hence, showed a narrow distribution of 1.100 ± 0.022 V.In contrast, DV nr exhibited a signicant variation ranging from 0.139 to 0.363 V.In other words, the variation of DV was mostly governed by that of DV nr , as shown in Fig. S12.† As mentioned above, DV nr signicantly depends on the energy offset, indicating that DV nr can be an alternative quantitative measure for the energy offset. 20Therefore, as shown in Fig. 3a, the maximum EQE PV values were also plotted against DV nr .EQE PV again showed a clear threshold at approximately 0.2 V; EQE PV dropped sharply in the region where DV nr was less than 0.2 V.This result is consistent with the fact that few OSCs reported to date exhibited high EQE PV with DV nr less than 0.2 V. 9 Briey, by plotting EQE PV against the three different criteria (DV, DIE PYSA , and DV nr ), we observed a clear threshold above which efficient charge photogeneration could be ensured in any of the three criteria, indicating that a D:A interface with a certain amount of energy offset is required for efficient charge photogeneration, contrary to what some reports have claimed. 29,30Note that the V OC and voltage losses are also affected by the morphology of the active layer.However, differences in morphology cannot explain the existence of the threshold energy observed in this study as morphology is independent of the energy offset.
Ideally, it is the most straightforward to discuss the relationship between EQE PV and the energy offset between E g and E CT ; however, accurately determining the E CT of some blends is challenging. 8For instance, E CT can be determined by timeresolved PL measurements if the CT emission can be distinguished from the corresponding prompt emission. 15However, this requires a near-unity exciton dissociation and reasonable CT emission intensity.Hence, CT emission is buried under the corresponding prompt emission when any of these are not satised, making it impossible to determine E CT from emission measurements.Therefore, we will use DV nr as an alternative quantitative measure for the energy offset in the following sections.Plots using DIE PYSA can be found in the ESI.† The use of DV nr and DIE PYSA has their advantages and disadvantages.The former is determined from the actual device, and hence, the value reects the nature of the D:A interface; however, the value is not directly related to the energy offset.In contrast, the latter can be easily measured, although the value may be overestimated relative to the actual HOMO energy difference at the D:A interface owing to the contribution caused by chargepermanent multipole interactions, which are weakened at the interfacial mixed region.Because DV is a less direct measure of the energy offset than the others, it is not discussed hereinaer.Note that, as an alternative approach, temperature-dependent V OC measurements allow us to estimate E CT at 0 K by linear extrapolation. 19,38,40,41However, E CT at room temperature is typically 0.1-0.2eV higher than that at 0 K. 40,41 Because we discuss the effect of slight energy offset differences, the uncertainty of 0.1-0.2eV is undesirable; hence, the temperaturedependent V OC measurements were not applied in this study.

Charge separation efficiency
We next measured the PL quenching yield F q of the D:A blends aer selective excitation of acceptors (Fig. S13-S15, ESI †).Fig. 3b shows the F q s plotted against DV nr .The F q was close to unity when the energy offset was satisfactory (as a rule of thumb, in the region where DV nr is larger than 0.2 V).Because the F q is a product of the efficiencies of exciton harvesting at the D:A interfaces and charge transfer at the interface, the nearunity F q indicates that all the excitons generated in the acceptor domains can reach the interface and quantitatively dissociate into the CT states.We have previously observed Y6 singlet exciton dynamics in PBDB-T-2F:Y6 blends using transient absorption (TA) spectroscopy. 15Upon selective photoexcitation of Y6, singlet excitons were generated in Y6 domains, which subsequently reached the D:A interface with a time constant of ∼6 ps.Although the discussion on the intrinsic lifetime of Y6 singlet excitons is controversial, 8,42 it is expected to be at least 200 ps, which is sufficient for all the excitons to reach the interface.Strikingly, the hole transfer from Y6 to PBDB-T-2F occurs with a sub-picosecond time scale despite the small energy offset of only 0.12 eV, resulting in the quantitative conversion of excitons to the CT states.In contrast, the quenching yield decreased in the region where DV nr was less than 0.2 V.Because the exciton harvesting efficiency is independent of the energy offset, 8,43 this result indicates that the hole transfer rate decreases with a decrease in the energy offset, as is expected by Marcus theory in the normal region, 14,16 leading to inefficient CT state formation.
To conrm this, we performed TA measurements for inefficient PBDB-T-2F-based blend lms.The PBDB-T-2F:Y5 blend is a representative low-EQE PV system (yellow triangles in Fig. 3) and suitable for comparison with the adequately investigated PBDB-T-2F:Y6 blend; 15,44,45 thereby, the results for the PBDB-T-2F:Y5 blend are shown in the main text, whereas the results for other inefficient blends (Y1, Y2, and Y16) can be found in Fig. S16, ESI.† Since TA data for the inefficient blends are scarce (whereas there are quite a few for the efficient blends), these results will further deepen our understanding.Fig. 4a shows the TA spectra of the PBDB-T-2F:Y5 blend lm aer selective photoexcitation of Y5 at 800 nm.In analogy with the assignments for the PBDB-T-2F:Y6 system, 15,42,44 the large photoinduced absorption (PIA) band observed at around 900 nm immediately aer the photoexcitation can be attributable to Y5 singlet excitons, which gradually decayed thereaer.The broad negative signal observed in the 500-800 nm region is assigned to ground-state bleaching (GSB).In the case of the PBDB-T-2F:Y6 blend, the PBDB-T-2F GSB was observed in the 500-650 nm region already at 0 ps, evidencing the sub-picosecond hole transfer (Fig. S16a, ESI †).In contrast, close inspection of the GSB signal of the PBDB-T-2F:Y5 blend (Fig. 4b) revealed that the spectral shape in the 500-650 nm region at early times was different from that at later times, meaning that the GSB signal in this region at 0 ps should mainly be ascribed to Y5 due to slow hole transfer.In addition, a rise in the PBDB-T-2F GSB signal was not observed in the PBDB-T-2F:Y5 blend, in sharp contrast to the PBDB-T-2F:Y6 blend, where the PBDB-T-2F GSB signal increased with a time constant of ∼6 ps (Fig. S16a †). 15 The rise in the PBDB-T-2F GSB signal was also not observed in other inefficient PBDB-T-2F-based blends (Fig. S16b-d †), indicating slower hole transfer in these inefficient blends than in the PBDB-T-2F:Y6 blend.Fig. 4c shows the time evolution of the TA signals monitored at 550 nm, where the TA signal was initially positive and then turned negative at later times, representing the decay of the Y5 excitons and the generation of the PBDB-T-2F GSB.The growth of the PBDB-T-2F GSB signal continued until at least 100 ps, again indicating slow hole transfer.Note that the later time TA kinetics at this wavelength includes the contribution of the geminate recombination of the CT states (vide infra); thereby, it is possible that the hole transfer occurred even aer the appearance of the apparent peak at ∼100 ps.These results conrmed that the slow hole transfer from Y5 (and also Y1, Y2, and Y16) to PBDB-T-2F was due to insufficient energy offsets.
To further deepen our understanding, EQE PV was divided by F q .Fig. 3c shows EQE PV /F q plotted against DV nr .Interestingly, EQE PV /F q sharply decreased in the same region (DV nr < 0.2 V).EQE PV can be expressed as: where h abs , h ED , h CT , h CD , and h CC are the QYs of photon absorption, exciton diffusion to the D:A interface, charge transfer to form a CT state, long-range spatial dissociation of the CT state, and charge collection at the respective electrodes, respectively. 8,9Therefore, Fig. 3c represents the dependence of the long-range charge dissociation efficiency h CD on the energy offset because F q = h ED × h CT .Note that this assumes that h abs and h CC are independent of the energy offset.The former is evident.The latter is reasonable at least under short-circuit condition with a weak light illumination, such as in the EQE PV measurements, because bimolecular charge recombination is suppressed under this condition. 46,47This assumption is veried by the good consistency between the measured J SC (under 1 sun condition) and calculated J SC obtained by integrating the EQE PV (measured under short-circuit condition with a weak illumination) (Tables S4-S6, ESI †).Therefore, a decrease in EQE PV /F q in the DV nr < 0.2 V region suggests that the longrange charge dissociation becomes inefficient when the energy offset is too small.To conrm this, we measured the excitation-uence dependence of the TA decays.Fig. 4d shows the recoveries of the PBDB-T-2F GSB under various excitation uences.9][50][51] This is in sharp contrast to what we previously observed for the PBDB-T-2F:Y6 blend, where the geminate recombination was negligible and bimolecular recombination was signicant even at low excitation uences. 15Fig. 4c also shows the time evolution of the TA signals monitored at 680 nm.The negative signal, which is assigned to the Y5 GSB, slowly recovered and nally turned positive aer 300 ps.This behavior was also observed for the other PBDB-T-2F-based blends, whereas the signal amplitude and the time at which the TA signal turned positive depended on the acceptors (Fig. S16, ESI †).We found that the positive signals at 680 nm of the inefficient blends were signicantly small compared with that of the PBDB-T-2F:Y6 blend (Fig. 4c and S16, ESI †).In the PBDB-T-2F:Y6 blend, the TA signal turned positive already at 10 ps aer photoexcitation and increased further.The positive signal at this wavelength is assigned to the transient electroabsorption (EA) signal of PBDB-T-2F. 15,44,45,52hen an exciton dissociates into an electron-hole pair (CT state) at the D:A interface, the electron-hole pair generates a local electric eld in the surroundings.This causes a Stark shi in the steady-state absorption of surrounding molecules, leading to the addition of a transient EA signal to the TA spectra. 53,54Because the transient EA signal increases with an increase in the separation distance of the electron-hole pair, our ndings for the inefficient blends indicate that the charge dissociation in these blends is slow and inefficient.Therefore, the decay kinetics were tted using an exponential function to determine the charge dissociation efficiency.The TA decay kinetics of the PBDB-T-2F:Y5 blend monitored at 630 nm (PBDB-T-2F GSB) were well tted with the exponential function with a CT state lifetime of ∼1.3 ns and the charge dissociation efficiency of only ∼42% (Fig. 4d).Note that the charge dissociation efficiency obtained herein was lower than EQE PV /F q of the same blend (∼46%) because TA measurements were performed under open-circuit condition, as will be discussed later.
We then measured the TA spectra of a PBDB-T:Y5 blend lm to elucidate whether the low charge dissociation efficiency of the PBDB-T-2F:Y5 blend stems from the intrinsic nature of Y5 itself.The PBDB-T:Y5 blend exhibits a DV nr of 0.234 V, which is above the threshold, and hence, comparison between the largeoffset PBDB-T:Y5 and small-offset PBDB-T-2F:Y5 blends allows us to elucidate the origin of the inefficient charge dissociation in the PBDB-T-2F:Y5 blend.Fig. 5a shows the TA spectra of the PBDB-T:Y5 blend lm in the visible region, where the assignments of the TA signals are the same as those of the abovementioned PBDB-T-2F-based blends.We found that the hole transfer occurred in a sub-picosecond time scale in the PBDB-T:Y5 blend owing to the sufficient energy offset.Strikingly, the PBDB-T:Y5 blend exhibited considerably larger transient EA signals compared with the PBDB-T-2F:Y5 blend (Fig. 5b).In addition, the decay kinetics of the PBDB-T GSB depended on the excitation-uence due to the contribution of the bimolecular recombination (Fig. S17, ESI †), which is indicative of efficient free carrier generation.These results are consistent with the relatively high J SC of the PBDB-T:Y5 device.Therefore, we conclude that the low charge dissociation efficiency of the PBDB-T-2F:Y5 blend does not stem from the intrinsic nature of Y5 itself, but rather stems from the absence of sufficient energy offset at the D:A interface.
Our nding that the long-range spatial dissociation of CT states is inefficient when the energy offset is too small is very interesting.A key driver for the efficient charge dissociation in the representative efficient PBDB-T-2F:Y6 blend is attributed to the formation of a cascaded energy landscape near the D:A interface, 15 which enables charges moving away from the interface without experiencing activation barriers because the attracting Coulomb barrier is compensated by the cascaded energy landscape.A possible explanation for the origin of the formation of the cascaded energy landscape relies on the large quadrupole moments of NFAs. 34,37Owing to a concentration gradient of NFAs near the D:A interface, charge-quadrupole interaction continuously increases with an increase in the distance from the interface, leading to the formation of a cascaded energy landscape (also referred to as band bending).Previous studies claimed that the drawback of the band bending was attributed to hole transfer becoming inefficient when the energy offset is too small. 34In contrast, within the model proposed in ref. 34, once the CT states are formed through hole transfer, they are expected to efficiently dissociate into free carriers because the band bending is benecial for charge dissociation.However, this study reveals that the decrease in the charge dissociation efficiency was more significant than the decrease in the hole transfer efficiency when the energy offset is too small (Fig. 3).This is because decreasing the hole transfer rate does not necessarily result in low hole transfer efficiency as the hole transfer efficiency is determined by the competition with the intrinsic exciton decay rate.The hole transfer rate in the PBDB-T-2F:Y6 blend is more than 10 12 s −1 , despite the small energy offset of 0.12 eV, as mentioned above.Therefore, even if we assume that the hole transfer rate becomes an order of magnitude lower than that in the PBDB-T-2F:Y6 blend, the hole transfer rate remains on the order of 10 11 s −1 , which is still larger than the intrinsic decay rate of acceptor excitons (<5 × 10 9 s −1 ), leading to maintaining a relatively high hole transfer efficiency.In contrast, if the hole transfer becomes two orders of magnitude slower than that in the PBDB-T-2F:Y6 blend, the hole transfer efficiency should be signicantly decreased.We estimated to what extent the hole transfer rate is decreased in the inefficient blends compared with the PBDB-T-2F:Y6 blend and found that the hole transfer rate supposedly remains only an order of magnitude lower (Fig. S18, ESI †). 8,55e found that the CT state lifetime in the representative inefficient PBDB-T-2F:Y5 blend (∼1.3 ns) was considerably shorter than that in the representative efficient PBDB-T-2F:Y6 blend (∼2.6 ns, determined by the time-resolved PL spectroscopy 15 ).This result implies that the CT states generated in the inefficient blends are more localized and tightly bound at the D:A interface. 51Therefore, we propose that, as schematically shown in Fig. 5c and d, there exists an activation barrier for charge dissociation, and hence, the QY of the long-range spatial dissociation depends on the initial separation distance of the CT states, which is affected by the energy offset.When the energy offset is sufficient, the initial separation distance between electron and hole aer isoenergetic charge transfer is satisfactory for charges to overcome the activation barrier, resulting in barrier-less charge dissociation.In contrast, when the energy offset is insufficient, charges need to overcome the activation barrier, resulting in poor charge dissociation efficiency.This model is similar to those developed for fullerenebased OSCs, 11,12,53,56 except that the origin of the cascaded energy landscape in this model is attributed to the large quadrupole moment of NFAs.Although no clear experimental evidence that supports our hypothesis is yet available at the moment, our ndings encourage us to further investigate the charge separation mechanism in future studies.

Impact of voltage loss on ll factor
We move our attention to the impact of voltage loss on FF.In general, a large V OC is expected to provide a high FF. 9,57,58evertheless, as shown in Fig. 6a, the FFs of our devices decreased in the high V OC region.To deepen our understanding, FF was plotted against DV nr (Fig. 6b, see also Fig. S19, ESI †), where FF dropped in the DV nr < 0.2 V region, similar to the case of EQE PV .This phenomenon can be rationalized by the following reasons.As is discussed in the previous section, the efficiency of the long-range charge dissociation is insufficient when the energy offset is too small.The charge dissociation efficiency h CD of the PBDB-T-2F:Y5 blend obtained by TA measurements was lower than EQE PV /F q (= h abs × h CD × h CC ), as mentioned above.This indicates that the charge dissociation efficiency exhibits bias dependence because TA measurements were performed under open-circuit condition, whereas EQE PV was measured under short-circuit condition, where charge dissociation is accelerated with the aid of a large internal electric eld. 8,11,59,60Therefore, the charge dissociation efficiency obtained by TA measurements being less than EQE PV /F q indicates that charge dissociation in inefficient devices becomes less efficient with an increase in the applied voltage in the forward direction, resulting in a poor FF. 50,61 In addition, these devices may be more prone to suffer from bimolecular recombination loss.When a charge carrier encounters an opposite charge at the interface, a CT state is regenerated.If the CT state deactivates to the ground or lower-lying local triplet states, charge decay ultimately occurs.Instead, if the CT state redissociates into free carriers, bimolecular recombination is suppressed. 46,47With decreasing energy offset, the rate of radiative decay of the CT state and the rate of back charge transfer to the excited state are expected to increase, resulting in a less efficient redissociation of the CT state.Therefore, bimolecular recombination loss should be more severe for low-energy offset devices.
Our ndings suggest that the trade-off between not only J SC and V OC , but also FF and V OC should be managed.FFs of stateof-the-art OSCs have increased to ∼0.8, which signicantly contributed to the improvement in the PCEs in the last two years. 9However, to the best of our knowledge, no OSCs with DV nr far less than 0.2 V exhibited a FF of 0.8.This is probably due to the trade-off between FF and V OC .Therefore, future research should focus more on overcoming the trade-off between FF and V OC .

Ternary blend system
Finally, we demonstrated ternary blend OSCs consisting of two NFAs paired with a common donor to deepen our understanding on the interplay of the energy levels of the two NFAs.The two NFAs were selected based on the following criteria.The majority NFA was chosen such that a high EQE PV and DV nr being as close as possible to the threshold were achieved when paired with the common donor in the binary reference system.In contrast, the minority NFA was selected to achieve a low DV nr of less than 0.18 V when paired with the common donor in the binary reference system.EQE PV can be low in this case (all acceptors satisfying the criterion for DV nr exhibit low EQE PV s, as mentioned above).2][3][4][5][6] In other words, two NFAs in typical ternary OSCs are selected such that both binary references exhibit DV nr s of approximately or larger than 0.2 V and high EQE PV s.In contrast, the minority NFA in this study was unique as it was chosen with only a small DV nr , ignoring EQE PV .In this way, the role of the two NFAs was emphasized, improving the clarity of future material design concepts.
The PBDB-T-2Cl:L8-BO device exhibited a high EQE PV (maximum EQE PV of 84.9% at 580 nm) with a moderate DV nr of 0.209 V, which is approximately the threshold value that ensures a high EQE PV (Fig. 3).In contrast, the PBDB-T-2Cl:Y1 device exhibited a low DV nr of 0.174 V, whereas the EQE PV was low (maximum EQE PV was 36.1% at 580 nm).Therefore, we prepared ternary blend OSCs consisting of PBDB-T-2Cl, L8-BO, and Y1 as a common donor, majority NFA, and minority NFA, respectively, to understand the role of the two NFAs.Fig. 7a shows the EQE PV spectra of the PBDB-T-2Cl:L8-BO:Y1 ternary blend devices with different blend ratios (J-V characteristics and device parameters can be found in Fig. S20, S21 and Table S9, ESI †).The blend ratios were varied over 1 : 1.2-x : x, as indicated in the gure.Strikingly, EQE PV remained almost unchanged even when x was varied from 0 to 0.3.EQE PV was decreased when x was 0.4, although it was still considerably higher than that of the PBDB-T-2Cl:Y1 binary system (x = 1.2).The fact that the addition of a small amount of Y1 will not degrade the charge separation efficiency, despite the poor charge separation efficiency of the PBDB-T-2Cl:Y1 binary reference (Fig. 3 and S16, ESI †), suggests that charge separation preferentially occurs at the PBDB-T-2Cl:L8-BO interface, as schematically shown in Fig. 7d.When a Y1 exciton reaches the D:A interface, it will likely return to the bulk because the hole transfer is slow.On the other hand, the Y1 exciton can move to the L8-BO domain without a large energetic barrier because of similar excited state energies of L8-BO and Y1 (1.45 and 1.43 eV for L8-BO and Y1, respectively, Fig. S4, ESI †).Once the excitons are transferred to the L8-BO domains, they can quickly dissociate into the CT states.Therefore, charge separation preferentially occurs at the PBDB-T-2Cl:L8-BO interface, resulting in a high EQE PV despite the addition of a small amount of Y1.
In contrast, V OC continuously increased with an increase in the Y1 blend ratio.The detailed voltage loss analysis revealed that the DV nr of the ternary blend device (x = 0.2) was 0.199 V, which was lower than that of the PBDB-T-2Cl:L8-BO binary reference (0.209 V) (Table S10, ESI †).To conrm this, we measured the external quantum efficiency of electroluminescence (EQE EL ) of an OSC device at forward biases because DV nr is directly related to EQE EL as follows: 40,41 As shown in Fig. 7c, EQE EL of the ternary blend device was apparently higher than that of the PBDB-T-2Cl:L8-BO binary reference.Rather, EQE EL of the ternary blend device was close to that of the PBDB-T-2Cl:Y1 binary reference, especially at low applied voltages.The DV nr s determined by the two methods were in accordance with each other within a small error (Table S10, ESI †), and DV nr of the ternary blend device was smaller than that of the PBDB-T-2Cl:L8-BO binary reference in both methods.These results suggest that the PBDB-T-2Cl:Y1 interface preferentially served as a recombination center in the ternary blend, as schematically shown in Fig. 7e.Because the redissociation of the CT states is less efficient at the PBDB-T-2Cl:Y1 interface than that at the PBDB-T-2Cl:L8-BO interface, it is expected that charge recombination is more likely to occur at the former.Because the EQE EL of the PBDB-T-2Cl:Y1 blend is higher than that of the PBDB-T-2Cl:L8-BO blend, the preferential recombination at the PBDB-T-2Cl:Y1 interface leads to an increase in the EQE EL of the ternary blend compared with that of the PBDB-T-2Cl:L8-BO blend.Therefore, the ternary blend OSCs consisting of a common donor paired with two NFAs, one with a high EQE PV and the other with a low DV nr , is expected to be an effective approach to overcome the trade-off between J SC and V OC .A remaining challenge with our ternary blend concept is that the trade-off between FF and V OC has not been overcome.As shown in Fig. 7b, although an increase in the blend ratio of Y1 resulted in an increase in V OC , it simultaneously decreased the FF.Therefore, both binary and ternary systems face the same challenge of overcoming the trade-off between FF and V OC .

Conclusions
We  from the NFA to donor decreases with a decrease in the energy offset.In contrast, what is surprising in our ndings is that the quantum efficiency of the long-range spatial dissociation of the CT states also decreased with a decrease in the energy offset.This behavior cannot be rationalized only by the recently proposed model wherein charge dissociation occurs via the downhill relaxation of charges through the cascaded energy landscape.We propose that there still exists an activation barrier for charge dissociation even when a NFA with a large quadrupole moment is used and the quantum efficiency of the long-range spatial dissociation depends on the initial separation distance of the CT states.If the initial separation distance between the electron and hole aer isoenergetic charge transfer is satisfactory, barrier-less charge dissociation can be achieved; otherwise, charges need to overcome the activation barrier, resulting in poor charge dissociation efficiency.Another interesting nding of this study is the observation of a clear trade-off between FF and V OC caused by the deterioration of the charge dissociation efficiency when the energy offset is too small.This may be another reason why the DV nr s of state-of-the-art OSCs remained at approximately 0.2 V.
Although we have focused on PBDB-T:Y-series blends to minimize the difference in the chemical structure and associated changes in physical properties, the observed trends can be considered as general for other NFA systems. 34As the threshold energy that can ensure high charge photogeneration quantum efficiencies may depend on the chemical structure, 13,34,38 more efforts should be dedicated to unveiling the complete details of the charge separation mechanism.Extending a similar study to other donor:acceptor systems can reveal what determines the threshold energy and how the threshold energy can be reduced.In this way, clear material design guidelines can be obtained for further improving the PCE.
The role of the two NFAs in ternary blend OSCs was also elucidated.In this study, the majority NFA was selected such that a high EQE PV and DV nr being as close as possible to the threshold were achieved, whereas the minority NFA was chosen such that a low DV nr of less than 0.18 V was achieved, ignoring EQE PV .In this way, the role of the two NFAs was emphasized.We found that by introducing the minority NFA, V OC could be increased while maintaining a high EQE PV .This is likely because charge separation preferentially occurs at the donor:majority NFA interface, whereas charge recombination is more likely to occur at the donor:minority NFA interface.One drawback of this ternary blend concept is the decrease in FF upon the introduction of the minority NFA because of the same reason as that for the decrease in FF in binary blends with a too small energy offset.Therefore, future research should focus more on overcoming the trade-off between FF and V OC .If this challenge can be overcome, the PCE of our ternary blend device can be further improved.

Fig. 1
Fig. 1 Chemical structures of (a) donor and (b-d) acceptor materials employed in this study.PBDB-T-2F and PBDB-T-2Cl are also referred to as PM6 and PM7, respectively.(e) Ionization energies (IEs) of pristine donor and acceptor films determined by photoelectron yield spectroscopy in air (PYSA).

Fig. 4
Fig. 4 (a) TA spectra of a PBDB-T-2F:Y5 blend film.The excitation wavelength was 800 nm with a fluence of 6.3 mJ cm −2 .(b) Normalized TA spectra of (a) at 630 nm (GSB peak of PBDB-T-2F).(c) Time evolution of the TA signals monitored at 550 and 680 nm.(d) Excitation-fluence dependence of the TA decays monitored at 630 nm.The red dashed lines represent the best fitting curves with the sum of an exponential function and a constant fraction, DmOD = a exp(−t/s) + b, where a, s, and b are the fitting parameters and s is shared for all the four decay curves.The charge dissociation efficiency was determined as b/(a + b).

Fig. 5
Fig. 5 (a) TA spectra of a PBDB-T:Y5 blend film.The excitation wavelength was 800 nm with a fluence of 6.2 mJ cm −2 .(b) Time evolutions of the TA signals of the PBDB-T:Y5 and PBDB-T-2F:Y5 blend films monitored at 630 nm (GSB peak of donor polymers) and 680 nm (transient EA).The excitation wavelength was 800 nm with fluences of 6.2 mJ cm −2 (PBDB-T:Y5) and 6.3 mJ cm −2 (PBDB-T-2F:Y5).Schematic showing the charge separation process in (c) efficient D:A blends and (d) inefficient D:A blends.Ex, CT, FC, and Gr refer to the excited state, CT state, free carrier, and ground state, respectively.(c) When the energy offset is sufficient, the initial separation distance between the electron and hole after isoenergetic charge transfer is satisfactory to overcome the activation barrier, resulting in barrier-less charge dissociation.(d) In contrast, when the energy offset is insufficient, charges need to overcome the activation barrier, resulting in poor charge dissociation efficiency.

Fig. 7
Fig. 7 (a) EQE PV spectra of PBDB-T-2Cl:L8-BO:Y1 ternary blend devices with different blend ratios.The blend ratios were varied over 1 : 1.2-x : x, as indicated in the figure.(b) Blend ratio dependence of V OC (blue circles, left axis) and FF (yellow triangles, right axis).(c) EQE EL of the PBDB-T-2Cl : L8-BO : Y1 (1 : 1 : 0.2) ternary blend device as well as its binary references.(d) Schematic showing the charge separation process.Hole transfer is slow at the donor:minority NFA (D:A2) interface, and the CT states are not likely to form.On the other hand, excitons can move between the majority and minority NFA domains without a large energetic barrier.Therefore, charge separation preferentially occurs at the donor:majority NFA (D:A1) interface following exciton (energy) transfer from the minority to majority NFAs.ED, ET, CT, and CD refer to exciton diffusion, energy (exciton) transfer, charge transfer, and charge dissociation, respectively.(e) Schematic showing the charge recombination process.Because the redissociation of the CT state is less efficient at the D:A2 interface compared to that at the D:A1 interface, charge recombination is expected to be more likely to occur at the former.Because the EQE EL of the D:A2 blend is higher than that of the D:A1 blend, the preferential recombination at the former leads to an increase in the EQE EL of the ternary blend.BR and CS refer to bimolecular charge recombination to regenerate the CT states, and charge shift between A1 and A2, respectively.
have investigated the threshold to what extent the energy offset can be minimized for Y-series NFA-based OSCs.We found that the Y-series OSCs exhibited a clear threshold DV nr of 0.2 V, below which EQE PV sharply decreased.This is the reason why the DV nr s of state-of-the-art OSCs remains approximately 0.2 V and very few studies have reported that DV nr can be exceeded beyond the 0.2 V threshold while maintaining a high EQE PV .It is easily expected from Marcus theory that the hole transfer rate