The rapid development in light-harvesting materials, especially non-fullerene acceptors (NFAs)1,2,3, has enabled exciting progress in organic solar cells (OSCs)4,5,6,7. For the OSCs to be competitive and commercially viable, further improving device performance (power conversion efficiency (PCE) and stability) is urgently needed. Regarding the OSC efficiency, one must improve the open circuit voltage (VOC) along with efficient exciton dissociation and charge transport process, i.e., simultaneously suppress VOC loss (the difference between the bandgap and eVOC, also known as energy loss, Eloss) and achieve ideal nano-scale morphology in OSCs. From a physical point of view, minimizing VOC loss requires suppressing the absorption of charge transfer (CT) state and improving the electroluminescence quantum efficiency (EQEEL) of OSC devices8,9,10, but the rational design of molecules that can achieve these aims remains a challenge. Besides, the Eloss of OSC is not only related to the intrinsic properties of donor (D) and acceptor (A) molecules but also to the nano-morphology of D/A blend, such as the distribution, aggregation and packing of D/A molecules. Therefore, the surface energy and crystallinity of the newly designed molecules should also be considered.

Introducing a proper third component to the benchmark binary blend is a widely used method to further enhance device efficiency, starting from the earlier fullerene era by multiple-donor system to more choices in NFA era7,11,12,13,14,15,16,17,18. Considering the excellent efficiency in state-of-the-art binary NFA OSCs, the third component is naturally better not to disturb the excellent morphology of the host binary blend. In this regard, NFAs with smart halogen substitution positions in end groups have high potential to provide similar molecular orientation and excellent miscibility19, but adjustable different aggregation and photoelectronic properties. In this sense, the development of such NFAs is likely to achieve the efficiency enhancement goal.

Herein, the molecular design started from the optimization of halogen substitution position in terminal groups. Guided by theoretical calculations, we synthesized o-BTP-eC9, an isomer of BTP-eC9 with different chlorine substitutions on the dichlorinated 1,1-dicyanomethylene-3-indanone (IC-2Cl) end groups5. The very similar chemical structures lead to excellent miscibility of these two isomers. Compared to BTP-eC9, o-BTP-eC9 shows shallower LUMO (the lowest unoccupied molecular orbital) level, higher dielectric constant, and weaker crystallinity. When paired with PM6, o-BTP-eC9 based OSC exhibits a comparable PCE of 18.7% with that of BTP-eC9 based OSC (18.9%). And the energy loss in PM6:o-BTP-eC9 device is 41 meV smaller than that in PM6:BTP-eC9 device. Taking advantages of o-BTP-eC9, the good miscibility, the complementary in crystallinity and energetics, we introduced o-BTP-eC9 into the PM6:BTP-eC9 blend to fine tune the nano-scale morphology and the energetics of the host blend. As a result, o-BTP-eC9 induces more favorable phase separation in morphology, more balanced charge transport process and suppressed VOC loss. In PM6:o-BTP-eC9:BTP-eC9 based ternary OSC, a record PCE of 19.9% (19.5% certified by Asymptotic Pmax Scan) was achieved. The overall rational design paves the way for OSC performance breakthrough.


Theoretical calculation and the intrinsic properties of target molecule

The electron-withdrawing ability of IC-2Cl depends on two parts, 2-(4-oxocyclopent-2-en-1-ylidene) malononitrile and dichlorobenzene. As shown in Supplementary Fig. 1, the changes in the position of chlorine atoms have a direct impact on IC-2Cl group’s dipole moment, thereby influencing the LUMO electron distribution of NFA molecule. According to the intensity borrowing mechanism, uplifting the LUMO level of NFA is an effective method to suppress the VOC loss of OSC20,21. Thus, our design starts from optimizing the dipole moment and uplifting the LUMO level of IC-2Cl group. There are three isomers of dichlorobenzene, namely ortho-dichlorobenzene, meta-dichlorobenzene, and para-dichlorobenzene (Supplementary Fig. 1). Combining with the 2-(4-oxocyclopent-2-en-1-ylidene) malononitrile unit, we can get 6 isomers of the IC-2Cl end groups, denoted as α-o-IC-2Cl, β-o-IC-2Cl, θ-o-IC-2Cl, α-m-IC-2Cl, β-m-IC-2Cl, p-IC-2Cl, respectively (Fig. 1). To select the most promising IC-2Cl group, we calculated the dipole moments, LUMO and HOMO levels of various IC-2Cl groups. From the quantum chemistry calculations, θ-o-IC-2Cl group shows the lowest dipole moment of 3.22 Debye and the highest LUMO level of −3.70 eV, thus becoming our primary target to uplift the LUMO level of NFA.

Fig. 1: The effect of chlorine substitution position on the property of end group.
figure 1

Chemical structures of various IC-2Cl groups and the corresponding quantum chemistry calculations of the dipole moments, LUMO and HOMO distributions.

We then synthesized o-BTP-eC9 following the synthetic routes presented in Supplementary Figs. 2 and 3. And the related chemical structures were carefully characterized with 1H nuclear magnetic resonance (NMR, Supplementary Figs. 4, 6 and 8), 13C NMR (Supplementary Figs. 5, 7 and 9), gas chromatography-mass spectrometer (GCMS, Supplementary Figs. 10 and 11), single crystal analysis (Supplementary Fig. 12 and Supplementary Data 1 and 2), and matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS, Supplementary Fig. 13). Figure 2a–c shows the molecular structures of PM6, BTP-eC9 and o-BTP-eC9, respectively. The only difference between the two acceptor materials is the position of chlorine substitution on the IC-2Cl terminal groups. Unlike the opposite distribution of chlorine atoms on the IC-2Cl groups of BTP-eC9, the chlorine atoms in o-BTP-eC9 molecule are placed on the same side as the carbonyl groups. Due to the orientation effect of 2,3-dichlorobenzoyl chloride, the synthetic procedure for θ-o-IC-2Cl is one step less than that for β-o-IC-2Cl, and the over yield of the new end group (54%) is obviously higher than that of its counterpart (36%). Besides the condensation reaction of θ-o-IC-2Cl with dialdehyde gives a yield of 96%, which is also significantly higher than that of the β-o-IC-2Cl (53%).

Fig. 2: Chemical structures of photoactive materials and the related physical properties.
figure 2

Chemical structures of PM6 (a), BTP-eC9 (b), and o-BTP-eC9 (c). d Normalized UV-vis absorption spectra of PM6, BTP-eC9 and o-BTP-eC9. e Comparison of dielectric constant values of BTP-eC9 and o-BTP-eC9. The electrostatic potential surfaces (ESP) of BTP-eC9 (f) and o-BTP-eC9 (g) calculated from the DFT simulation at the BLYP/6–31G* level (the coordinates of the optimized computational models can be found in Supplementary Data 3). h Energy levels of PM6, BTP-eC9 and o-BTP-eC9. Source data are provided as a Source Data file.

The ultraviolet–visible (UV–vis) absorption spectra of the two NFAs are presented in Fig. 2d and Supplementary Fig. 14, and the related data are summarized in Table 1. The maximum absorption peaks of BTP-eC9 in solution state (λmaxsol) and thin-film state (λmaxfilm) are located at 747 nm and 828 nm, respectively, which are consistent with previous report5. In comparison, the absorption peak of o-BTP-eC9 solution is unchanged (λmaxsol at 747 nm) but the absorption peak of o-BTP-eC9 film is slightly blue-shifted (λmaxfilm at 826 nm), which is related to the weaker electron-withdrawing ability of the modified IC-2Cl groups (as discussed later). It is obvious that the absorption spectra of both BTP-eC9 and o-BTP-eC9 are red-shifted from solution state to film state, this is a commonly observed phenomenon in the organic photoactive materials and should be ascribed to the enhanced π-π interaction in solid state22. The red-shift of o-BTP-eC9 is 79 nm, less than that of BTP-eC9 (81 nm), implying the weaker π-π sacking in o-BTP-eC9 film22. Then the dielectric constant (\({\varepsilon }_{r}\)) values of these two molecules were tested by fabricating capacitors23 (Supplementary Fig. 15a), and the detailed parameters are presented in Fig. 2e and Supplementary Table 1. BTP-eC9 and o-BTP-eC9 show average εr values of 2.4 and 2.9, respectively, implying the exciton binding energy become smaller in o-BTP-eC9. The higher εr of o-BTP-eC9 should be ascribed to the higher molecular dipole moment23, as revealed by quantum chemistry calculations at ground state (Supplementary Fig. 15b, c). From the density functional theory (DFT) simulations (Fig. 2f, g and Supplementary Data 3), the minimum negative value of electrostatic potential surface of o-BTP-eC9 (−1.426 eV) is higher than that of BTP-eC9 (−1.462 eV), implying o-BTP-eC9 should show higher LUMO level than BTP-eC9, which is confirmed by electrochemical cyclic voltammetry (CV) measurements. As shown in Supplementary Fig. 16, Fig. 2h and Table 1, o-BTP-eC9 exhibits an obviously higher LUMO energy level and a nearly unchanged highest occupied molecular orbital (HOMO) energy level compared to BTP-eC9. Correspondingly, the bandgap of o-BTP-eC9 (Egcv = 1.84 eV) is larger than that of BTP-eC9 (Egcv = 1.79 eV), which is in good agreement with the tendency of optical gaps observed from UV-vis absorption spectra (Table 1) and the calculated energy levels (Supplementary Fig. 17).

Table 1 Electrochemical and optical properties of BTP-eC9 and o-BTP-eC9

It is reported that the position of halogen atoms in end groups has a direct impact on molecular stacking22,24, thus we further investigated the orientation and crystallinity of o-BTP-eC9 and BTP-eC9 by using grazing incidence wide-angle X-ray scattering (GIWAXS) measurements. As revealed in Supplementary Fig. 18 and Supplementary Table 2, both BTP-eC9 film and o-BTP-eC9 film show preferred face-on orientation, with strong π-π stacking (010) peaks at q ≈ 1.68 Å−1 in the out-of-plane (OOP) direction and lamellar (100) peaks at q ≈ 0.43 Å−1 in the in-plane (IP) direction. The corresponding crystallite coherence length (CCL) values of o-BTP-eC9 (CCL010 = 15.3 Å and CCL100 = 57.1 Å) are smaller than that of BTP-eC9 (CCL010 = 18.2 Å and CCL100 = 59.5 Å). In general, the changes in chlorine substitution position have little influence on molecular stacking orientation but leads to smaller crystallites, which affords a chance to finely tune the morphology of the active layer by combining o-BTP-eC9 and BTP-eC9 as electron acceptors.

Because the structural difference between BTP-eC9 and o-BTP-eC9 is very small, these two isomers are expected to show excellent miscibility. To verify the hypothesis, we conducted contact angel tests to calculate the surface energy (γ) values of photoactive materials, and to further analyze the miscibility between these materials. As shown in Supplementary Fig. 19, the contact angles of water on PM6, BTP-eC9 and o-BTP-eC9 are 103.5°, 97.4° and 96.1°, respectively, and the contact angles of glycerol on PM6, BTP-eC9 and o-BTP-eC9 are 83.8°, 76.5° and 75.2°, respectively. According to Wu’s model25, the corresponding γ values of PM6, BTP-eC9 and o-BTP-eC9 are calculated as 20.71, 24.94, and 25.57 mNm−1, respectively (Supplementary Table 3). The miscibility between material A and B is commonly evaluated by the equation of \({{{{{{\rm{\chi }}}}}}}_{A:B}=K{(\surd {{{{{{\rm{\gamma }}}}}}}_{A}-\surd {{{{{{\rm{\gamma }}}}}}}_{B})}^{2}\), where \({{{{{{\rm{\chi }}}}}}}_{A-B}\) is Flory–Huggins interaction parameter, the lower \({{{{{{\rm{\chi }}}}}}}_{A:B}\) value, the better miscibility between two materials, and \(K\) is a positive constant26,27. \({{{{{{\rm{\chi }}}}}}}_{{{{{{{\rm{PM}}}}}}}6:{{{{{{\rm{BTP}}}}}}}-{{{{{{\rm{eC}}}}}}}9}\) and \({{{{{{\rm{\chi }}}}}}}_{{{{{{{\rm{PM}}}}}}}{{{{{\rm{6:o}}}}}}-{{{{{{\rm{BTP}}}}}}}-{{{{{{\rm{eC}}}}}}}9}\) are calculated as 0.196 \(K\) and 0.256 \(K\) respectively, while \({{{{{{\rm{\chi }}}}}}}_{{{{{{{\rm{BTP}}}}}}}-{{{{{{\rm{eC}}}}}}}{{{{{\rm{9:o}}}}}}-{{{{{{\rm{BTP}}}}}}}-{{{{{{\rm{eC}}}}}}}9}\) is much lower, only 0.0039 \(K\), indicating the excellent miscibility between BTP-eC9 and o-BTP-eC9. The significant difference between \({{{{{{\rm{\chi }}}}}}}_{{{{{{{\rm{PM}}}}}}}{{{{{\rm{6:o}}}}}}-{{{{{{\rm{BTP}}}}}}}-{{{{{{\rm{eC}}}}}}}9}\) and \({{{{{{\rm{\chi }}}}}}}_{{{{{{{\rm{BTP}}}}}}}-{{{{{{\rm{eC}}}}}}}{{{{{\rm{9:o}}}}}}-{{{{{{\rm{BTP}}}}}}}-{{{{{{\rm{eC}}}}}}}9}\) implies that the PM6:o-BTP-eC9 blend tends to form microscopic morphology with obvious phase separation, but BTP-eC9:o-BTP-eC9 blend is possible to form a well-mixed phase, which is favorable in ternary blend films28,29.

Device performance of binary and ternary OSCs

The impact of our molecular design on photovoltaic performance is investigated by devices with conventional structure of indium tin oxide (ITO)/poly (3,4-ethylenedioxythiophene): poly (styrene sulfonate) (PEDOT: PSS)/active layer/poly[(9,9-bis(3’-((N,N-dimethyl)-N-ethylammonium)-propyl)−2,7-fluorene)-alt-2,7-(9,9-dioctylfluorene)] dibromide (PFN-Br)/Ag. In this study, the classical polymer PM6 was selected as the donor material because of its complementary absorption (Fig. 2d) and suitable energy level (Fig. 2h)5. We employed additive 1,4-diodobenzene (DIB) to improve device performance, which we first demonstrated the eutectic phase behavior beneficial for NFA arrangement in 202130. The J-V curves of champion devices based on the two isomers are plotted in Fig. 3a, and the corresponding photovoltaic parameters are summarized in Table 2. o-BTP-eC9 based device shows a significantly higher VOC of 0.901 V than BTP-eC9 based device (VOC = 0.843 V) by 58 mV, which should be ascribed to the larger bandgap and less VOC loss of the PM6:o-BTP-eC9 device (will be discussed later). On the other hand, due to the narrower absorption range of o-BTP-eC9, the current density (JSC) of o-BTP-eC9 based device (JSC = 26.33 mA/cm2) is lower than that of the BTP-eC9 based device (JSC = 28.27 mA/cm2). The external quantum efficiency (EQE) measurements (Fig. 3b) again verified the tendency from J-V tests, with integrated JSC of 25.54 mA/cm2 and 27.39 mA/cm2 for the PM6:o-BTP-eC9 and PM6:BTP-eC9 based devices, respectively. Both devices show excellent fill factor (FF) values of around 79%, and as a result, o-BTP-eC9 contributes a comparable champion PCE of 18.7% (Table 2), with obviously reduced energy loss (will be discussed later).

Fig. 3: Device performance.
figure 3

JV curves (a) and EQE spectra (b) of the binary and ternary OSCs. c Certified results from Enli Tech. Optoelectronic Calibration Lab. d Comparison of PCEs versus year in efficient OSCs. Source data are provided as a Source Data file.

Table 2 Detailed photovoltaic parameters of binary and ternary devices

Considering the higher LUMO level and dielectric constant of o-BTP-eC9, as well as the good miscibility between o-BTP-eC9 and BTP-eC9, we introduced o-BTP-eC9 as the third component in the PM6:BTP-eC9 system to fine tune the micro morphology and energetics of the host blend. The photovoltaic parameters of ternary devices with different ratios of acceptors are summarized in Supplementary Table 4. As we can see, in the range of 10 wt% to 30 wt% for o-BTP-eC9, the VOC of device continuously increases, while JSC and FF undergo a process of first increasing and then decreasing. When 15 wt% BTP-eC9 was replaced by o-BTP-eC9, the ternary device showed a champion PCE of 19.88% (Table 2 and Fig. 3a), with a VOC of 0.860 V, a JSC of 28.75 mA/cm2 (integrated JSC of 27.83 mA/cm2 from EQE measurement) and an excellent FF of 80.41%. The device was further encapsulated and sent to an ISO/IEC 17025:2017 accredited calibration lab—Enli Tech Optoelectronic Calibration Lab. A certified efficiency of 19.48% (Fig. 3c and Supplementary Fig. 20) was rated by asymptotic maximum power (Pmax) scan recommended by the National Renewable Energy Laboratory31,32, which is more reliable than the typical fast current-voltage (I-V) scan. To the best of our knowledge, 19.9% efficiency (19.5% certified by Asymptotic Pmax Scan) is the record value for single-junction OSCs (Fig. 3d). In addition, the ternary OSC shows excellent operational stability under 1-sun illumination stress test at MPP (maximum power point). As presented in Supplementary Fig. 21, the two binary devices suffer from more serious burn-in loss, so the ternary OSC exhibits the higher T80 (the time in which device efficiency drop to 80% of initial value) of 724 h than the PM6:BTP-eC9 (T80 = 428 h) and PM6:o-BTP-eC9 (T80 = 578 h) OSCs.

Device physics

Compared to the host PM6:BTP-eC9 OSC, the ternary device displays improvements in all the three photovoltaic parameters (VOC, JSC and FF). To get more insights into the role of o-BTP-eC9, we first analyzed the VOC loss of these binary and ternary devices. The VOC loss in OSCs consists of three components, radiative recombination above the bandgap (∆E1), radiative recombination below the bandgap (∆E2) and non-radiative recombination loss (∆E3 or Eloss, nr)33. As ∆E1 is unavoidable and only determined by temperature and the bandgap of light-absorbing material34, the key to VOC improvement is to strive for a decrease in ∆E2 and ∆E3. The origin of ∆E2 is the redshifted absorption caused by the existence of charge transfer (CT) state35,36. As for ∆E3, according to reciprocity theory, it can be calculated from the electroluminescence quantum efficiency (EQEEL) of the device, the equation is expressed as \(\Delta {E}_{3}=-\frac{{{{kT}}}}{q}{{{{\mathrm{ln}}}}}{{{{{{{\rm{EQE}}}}}}}}_{{{{{{{\rm{EL}}}}}}}}\), where k is the Boltzmann constant, T is the Kelvin temperature and q is the elementary charge37. To be specific, higher EQEEL value corresponds to lower ∆E3 and an ideal photovoltaic device without non-radiative recombination loss should also be a perfect light emitting diode (LED) with 100% EQEEL.

The calculation process is presented in Supplementary Information, and the detailed VOC loss parameters are summarized in Table 3 and Fig. 4a. The bandgap (Eg) of the device was determined with a probability distribution method38,39, as shown in Supplementary Fig. 22, the determined Egs of PM6:BTP-eC9 device, PM6:o-BTP-eC9 device, and the ternary device are 1.396, 1.413, and 1.397 eV, respectively. By subtracting the VOC, the corresponding Eloss values are 0.541, 0.500, and 0.524 eV, respectively. Specifically, these three types of devices showed similar ∆E1 of 0.26 eV. But the ∆E2 of PM6:o-BTP-eC9 device (∆E2 = 0.048 eV) is obviously smaller than that of PM6:BTP-eC9 device (∆E2 = 0.069 eV). We believe the reasons for the reduced ∆E2 in PM6:o-BTP-eC9 device are two-fold. One is the higher ECT (the energy of CT state) caused by the changes in chlorination position, as demonstrated in Fig. 4b, c. Another one is the higher dielectric constant of o-BTP-eC9, which facilitates exciton dissociation and reduce recombination loss in OSC23. The ∆E2 of the ternary device lies between that of two binary devices, implying the incorporation of the guest acceptor o-BTP-eC9 has a positive impact on improving the ECT of the ternary device (Fig. 4d). As for the calculation of ∆E3, there are two widely used methods. One is calculated from the EQEEL, as mentioned in the previous paragraph:\(\Delta {E}_{3}=-\frac{{kT}}{q}{{{{\mathrm{ln}}}}}{{{{{{{\rm{EQE}}}}}}}}_{{{{{{{\rm{EL}}}}}}}}\). Here, the benchmark PM6:BTP-eC9 device showed the weakest EQEEL of 4.0 × 10−4 (Fig. 4e), corresponding to the highest ∆E3 of 0.202 eV. In comparison, the EQEEL of PM6:o-BTP-eC9 device is 75% higher, with a value of 7.0 × 10−4 (∆E3 = 0.188 eV). The optimized ternary device shows EQEEL of 6.6 × 10−4 (∆E3 = 0.190 eV), 65% higher than that of benchmark PM6:BTP-eC9 device. It is worth noting that the ∆E3 values of 0.188 and 0.190 eV are among the lowest non-radiative recombination loss in OSCs with over 18% PCE (Fig. 4f). Another method used to calculate ∆E3 is from J-V characteristic, following the equation of \(\varDelta {E}_{3}^{{{{{{{\rm{cal}}}}}}}}={{E}_{g}-q{V}_{{OC}}-\varDelta E}_{1}-\varDelta {E}_{2}\). Like the tendency observed in the first method, PM6:BTP-eC9 device suffers from more serious ∆E3 (0.211 eV) than that of PM6:o-BTP-eC9 device (∆E3 = 0.189 eV) and ternary device (∆E3 = 0.199 eV). The lowest ∆E3 in the PM6:o-BTP-eC9 system should be ascribed to the lower LUMO energy level offset, which contributes to the hybrid of singlet excited state (S1) and CT state, and thus improves device EQEEL20,21. Overall, the molecular design can obviously reduce the radiative and non-radiative recombination in OSC, therefore the PM6:o-BTP-eC9 device exhibited a high efficiency of 18.7% although with a lower JSC than the benchmark PM6:BTP-eC9 device. Due to the incorporation of o-BTP-eC9, the ternary device also showed lower ∆E2 and ∆E3 than the host PM6:BTP-eC9 binary device.

Table 3 Detailed Eloss parameters of binary and ternary devices
Fig. 4: Device physics.
figure 4

a The comparison of detailed energy loss for binary and ternary OSCs. ECT determination for the PM6:BTP-eC9 based (b), PM6:o-BTP-eC9 based (c) and ternary (d) OSCs, here EL represents the electroluminescence. e EQEEL spectra for binary and ternary OSCs. f Comparison of PCEs versus non-radiative recombination values in reported OSCs with over 18% efficiency. Source data are provided as a Source Data file.

Except for VOC loss, we also systematically investigated the charge transport, recombination, and collection processes of these three OSC systems. As shown in Supplementary Fig. 23 and Supplementary Table 5, we used space charge limited current (SCLC) method to estimate the charge mobilities of binary and ternary OSCs. The hole mobilities (μh) of all these three devices are similar, around 4 × 10−4 cm2/Vs. While the electron mobility (μe) of PM6:BTP-eC9 device (μe = 5.1 × 10−4 cm2/Vs) is faster than that of PM6:o-BTP-eC9 device (μe = 3.5 × 10−4 cm2/Vs), which should be ascribed to the higher crystallinity of BTP-eC9. Due to the introduction of o-BTP-eC9, the ternary device shows slower electron mobility than the host PM6:BTP-eC9 device, but the ratio of μh and μe of ternary device is closer to 1 (μh = 3.8 × 10−4 cm2/Vs, μe = 4.0 × 10−4 cm2/Vs). It is well accepted that the balanced charge transport is beneficial to reducing space current and thereby improving photocurrent and FF.

We then performed transient photocurrent (TPC) and transient photovoltage (TPV) measurements to study the charge dynamics of binary and ternary OSCs. By fitting the current and voltage curves with mono-exponential decay model, we can get the corresponding decay time τ40. TPC testes were operated under short-circuit condition, the larger τ value means the less efficient charge extraction in device. As presented in Supplementary Fig. 24a, the ternary OSC shows the shortest τ of 0.66 μs, illustrating the guest o-BTP-eC9 effectively accelerates charge extraction process. TPV tests were conducted under open-circuit condition, the smaller τ value means the more serious recombination process in device. Supplementary Fig. 24b shows the TPV results, and as we can see, the ternary device shows the longest τ of 1.97 μs, indicating the introduction of o-BTP-eC9 is an effective method to restrict charge recombination. As shown in Supplementary Fig. 25, we further studied the trap-assisted recombination, also known as Shockley-Read-Hall recombination, by measuring the dependence of VOC on light intensity (Plight). The relation of VOCnkT/q ln(Plight) can be used to evaluate the situation of trap-assisted recombination, where n, k, T, and q are the fitted slope of VOC versus Plight on a logarithmic scale, the Boltzmann constant, the temperature of Kelvin, and the elementary charge, respectively41,42. The fitted slope of ternary device is 1.02, lower than the values of two binary devices, which indicates that the trap-assisted recombination in the ternary device is suppressed by o-BTP-eC9.

To estimate the charge collection process, we measured the photocurrent density (Jph) versus the effective voltage (Veff) of binary and ternary devices. Jph can be defined as JL − JD, where JL and JD are the current densities under illumination and in the dark, respectively43,44. Veff is equal to V0 − Vbias, where V0 is the voltage when Jph = 0, and Vbias is the applied voltage bias. As shown in Supplementary Fig. 26, Jph becomes saturated when Veff increases to 2 V. We then calculated the charge collection probability (Pcoll) according to the equation Pcoll = Jph/Jsat. At short-circuit condition, the ternary OSC showed the highest Pcoll value of 99.2%, agreeing well with the TPC results. Combination of the more balanced charge transport, suppressed recombination and more efficient charge collection contributes to the enhanced JSC and excellent FF in ternary devices.

Nano-morphology and molecular packing in binary and ternary blend films

To understand the relationship between device performance and phase-separated morphology, tapping-mode atomic force microscopy (AFM) measurements were conducted. Figure 5a–c are the height images of the PM6:BTP-eC9, PM6:o-BTP-eC9, and the ternary blend films, respectively. All three blend films show delicate nanomorphology, but due to the high crystallinity of BTP-eC9, the PM6:BTP-eC9 film shows the most obvious molecular aggregation, with the highest root-mean-square roughness (Rq) of 0.99 nm. After introducing the less crystalline o-BTP-eC9, the excessive molecular aggregation in the host PM6:BTP-eC9 blend is suppressed, with the smaller Rq of 0.70 nm in the ternary film. The phase-separated nanostructures are more obvious in Fig. 5d–f, all films exhibit visible fibrillar networks, which is beneficial to exciton dissociation and charge transport. The favorable phase-separated morphology should result from the suitable miscibility between the donor and acceptor materials and should be a reason for the excellent FF values in all three devices. But like the phenomenon observed in height images, the PM6:BTP-eC9 film shows obviously larger phase clusters than other two films due to the higher molecular aggregation. To get a quantitative analysis, we plot line profiles of phase images in Supplementary Fig. 27, the PM6:BTP-eC9 film shows the largest average phase cluster size of 18.1 nm. After the incorporation of o-BTP-eC9, the phase clusters of the ternary film become more uniform and the average size (15.7 nm) become smaller, which is more favorable for exciton diffusion, exciton dissociation and charge carrier transport. The AFM images imply that the crystallinity difference and good miscibility between BTP-eC9 and o-BTP-eC9 fine-tuned the aggregation of NFA, thereby improving the morphology in the ternary film.

Fig. 5: Surface topography of binary and ternary blend films.
figure 5

AFM height images of the PM6:BTP-eC9 (a), PM6:o-BTP-eC9 (b) and the ternary (c) blend films. AFM phase images of the PM6:BTP-eC9 (d), PM6:o-BTP-eC9 (e) and ternary (f) blend films.

We further investigated the influence of o-BTP-eC9 on molecular stacking in BHJ by GIWAXS measurements, and the detailed information of lamellar diffraction peaks and π-π stacking (010) diffractions is summarized in Supplementary Table 6. Figure 6a–c presents the two-dimensional (2D) GIWAXS patterns of the PM6:BTP-eC9, PM6:o-BTP-eC9 and ternary blend films, respectively. Along the in-plane direction (Fig. 6d), all blend films show two lamellar diffraction peaks, ascribing to the lamellar stacking of PM6 (q at around 0.3 Å−1) and NFAs (q at around 0.4 Å−1), respectively. Along the out-of-plane (OOP) direction (Fig. 6e), there are distinct π-π diffractions (q at around 1.7 Å−1) for all films, implying these three blends all show preferred π-π stacking. But compared to the PM6:BTP-eC9 blend, the PM6:o-BTP-eC9 blend present obviously lower diffraction intensity due to the crystallinity difference of o-BTP-eC9. Like the tendency observed in the GIWAXS patterns of neat films, the two binary blend films show similar π-π stacking distance of 3.7 Å, but the CCL010 (22.5 Å) of PM6:o-BTP-eC9 blend is smaller than that (26.4 Å) of PM6:BTP-eC9 blend. Because of the crystallinity difference between these two isomers, the CCL010 value of the ternary film lies between that of two binary films. The role of o-BTP-eC9 in moderating the crystallization of BTP-eC9 was reconfirmed by grazing incidence small-angle X-ray scattering (GISAXS) measurements. Figure 6g–i are the 2D GISAXS patterns of binary and ternary films, and Fig. 6h is the corresponding1D profiles along the IP direction. An 1D GISAXS profile consists of a shoulder in the middle-q region, an upturn in the small-q region and an almost straight line in the high-q region. The shoulder originates from the crystalline regions of the blend, and the blunter shoulder corresponds to the larger size of crystalline region45,46. To get a quantitative comparison, we fitted these profiles by a fractal network model. The fitted average sizes of crystalline region in the PM6:BTP-eC9, PM6:o-BTP-eC9 and ternary blend films are 21, 12, and 18 nm, respectively, again verifying the molecular crystallization of the host blend is effectively optimized by incorporating the guest component. The AFM, GIWAXS and GISAXS measurements jointly demonstrate that the complementary crystallinity and good miscibility between BTP-eC9 and o-BTP-eC9 contributes to a more ideal phase separation morphology.

Fig. 6: GIWAXS and GISAXS measurements.
figure 6

2D GIWAXS diffraction patterns of the PM6:BTP-eC9 (a), PM6:o-BTP-eC9 (b) and ternary (c) blend films. The related 1D GIWAXS line cut profiles along IP (d) and OOP (e) directions. f 1D GISAXS profiles along qr direction. 2D GISAXS diffraction patterns of the PM6:BTP-eC9 (g), PM6:o-BTP-eC9 (h) and ternary (i) blend films. Source data are provided as a Source Data file.


In conclusion, theoretical DFT calculation guided the design and synthesis of a promising NFA by changing the chlorination position in IC-2Cl groups. The o-BTP-eC9 molecule shows shallower LUMO level, higher dielectric constant, and weaker crystallinity than BTP-eC9. The upshifted CT state in the PM6:o-BTP-eC9 OSC leads to suppressed radiative and non-radiative recombination. More excitingly, benefiting from the compatibility of the two NFAs in crystallinity and energy level, o-BTP-eC9 not only optimizes the energetics of ternary OSC but also fine tunes the phase separation morphology of the host PM6:BTP-eC9 blend. As a result, in the PM6:BTP-eC9:o-BTP-eC9 based ternary OSC, a record 19.9% PCE of single junction OSC with reduced energy loss was achieved. This work paves an avenue for designing new NFA guest molecules toward high performance OSCs.



All materials are provided by commercial suppliers: PEDOT:PSS (Clevios P VP AI. 4083 (Heraeus)), PM6 (Solarmer Energy Inc.), BTP-eC9 (Solarmer Energy Inc.), BTP-eC9-CHO (Jiangsu ji’a Biotechnology Co., Ltd.), 2,3-Dichlorobenzoic Acid (Shanghai Titan Scientific Co., Ltd.), DIB (Tokyo Chemical Industry Co., Ltd.), PFN-Br (Solarmer Energy Inc.), Chloroform (Sigma-Aldrich, Ltd.), and isopropyl alcohol (Sigma-Aldrich, Ltd.), Ag pellet (ZhongNuo Advanced Material (Beijing) Technology Co., Ltd.). And all reagents and solvents are used directly without further purification.

Device fabrication and testing

The OSCs were fabricated with a conventional structure of ITO/PEDOT: PSS/active layer/PFN-Br/Ag. At first, the ITO-coated glass substrates were cleaned sequentially with detergent, de-ionized water, acetone, and isopropyl alcohol (IPA) for 15 min under sonication. Then the substrates were dried in nitrogen flow and treated with UV ozone for 30 min. After that, about 50 ul PEDOT:PSS was dripped on ITO substrates and spin-coated at 6000 rpm for 20 s, followed by thermal annealing on a hot plate at 120 °C for 10 min to remove the water in PEDOT:PSS film. Then, the substrates were transferred into a glovebox filled with nitrogen (O2 < 10 ppm; H2O < 10 ppm). The precursor solutions were prepared by mixing 7.3 mg donor, 8.7 mg acceptor(s), and 12.5 mg DIB in 1 ml chloroform. For the optimized ternary device, the weight ratio of BTP-eC9: o-BTP-eC9 is 1.05:0.15. And the precursor solution would be heated to 60 °C for an hour to fully dissolve the solutes then cooled to room temperature before use. The thickness of the active layer was controlled at around 120 nm, then the active layer experienced a process of thermal annealing at 100 °C for 5 min. The next stage is to coat electron transport material, about 5 nm PFN-Br (0.5 mg/ml in methanol) was coated on the top of the active layer. Finally, these semi-finished cells were transferred into a thermal evaporation chamber with a base pressure of ~2 × 10−4 Pa, where 100 nm Ag was deposited through a shadow mask with the active area of 11 mm2. The J–V curves of OSCs were tested by a Keithley 2400 source meter and an AAA grade solar simulator (SS-F7-3A, Enli Tech. Co., Ltd., Taiwan) along with AM 1.5 G spectra whose intensity was corrected by a standard silicon solar cell at 1000 W/m2. The J-V curves are measured in the forward direction from −0.2 to 1.2 V, and all devices were covered with anti-reflection films (Mitsubishi Chemical Corporation) when conducted J–V test. The external quantum efficiency (EQE) was measured by a certified incident photon to electron conversion (IPCE) equipment (QE-R) from Enli Technology Co., Ltd.


AFM images were acquired using a Bruker Dimension EDGE in tapping mode. The GIWAXS data was obtained from the PLS II 3C SAXS-I and 9A U-SAXS beamline of the Pohang Accelerator Laboratory in Korea. The active layer samples were deposited on the Si/PEDOT:PSS substrates following device conditions. The X-rays coming from the in-vacuum undulator (IVU) were monochromated (wavelength λ = 1.10994 Å) using a double crystal monochromated and focused both horizontally and vertically (100 (H) × 20 (V) μm2 in FWHM @ the sample position) using K-B type mirrors. The grazing incidence wide-angle X-ray scattering sample stage was equipped with a 7-axis motorized stage for the fine alignment of the sample, and the incidence angles of the X-beam was set to be 0.11°–0.13°. The GIWAXS patterns were recorded with a 2D CCD detector (Rayonix SX165) and an X-ray irradiation time within 100 s, dependent on the saturation level of the detector. Diffraction angles were calibrated using a sucrose standard (monoclinic, P21, a = 10.8631 Å, b = 8.7044 Å, c = 7.7624 Å, and β = 102.938°) and sample-to-detector distance was ~231 mm. The GISAXS measurements were performed with a Xeuss 2.0 SAXS/WAXS laboratory beamline using a Cu X-ray source (8.05 keV, 1.54 Å) and a Pilatus3R 300 K detector.

Highly sensitive EQE and EQEEL measurements

Highly sensitive EQE was measured using an integrated system (PECT-600, Enlitech), where the photocurrent was amplified and modulated by a lock-in instrument. EQEEL measurements were performed by applying external voltage/current sources through the devices (ELCT-3010, Enlitech).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.