Probing the energy levels of organic bulk heterojunctions by varying the donor content

The performance of organic solar cells (OSCs) is dependent on, for instance, molecular design, device architectures, and fabrication conditions, offering a large parameter space to explore and optimize. Compared to other solar cell technologies, e.g., perovskites, which are mainly limited by charge carrier recombination between material layers,¹ OSCs are currently mainly limited by large energy losses in the bulk of the photovoltaic active layer and specifically at donor (D)–acceptor (A) interfaces. While bulk energy levels influence exciton diffusion and charge transport,² the energetics at the D–A interface impact the properties of charge transfer (CT) states³ and determine the efficiency of (1) charge transfer from singlet (S₁) to CT states, (2) exciton dissociation from CT to charge separated (CS) states, and (3) recombination to the ground state (GS). Therefore, studying and optimizing energetic landscapes and transitions, especially of excited S₁ and CT states in D–A blends, is crucial for increasing photocurrents and reducing recombination losses of OSCs. Nonetheless, an accurate determination of molecular and blend energy levels, in particular, around buried D–A interfaces in bulk heterojunctions (BHJs), remains a challenge.⁴ In the context of molecular energy levels, the ionization energies (IE) and electron affinities (EA) of molecules are often used as zeroth order approximations to judge the photovoltaic potential of different D–A pairs⁵ and are generally determined via density functional theory (DFT),^4,6 cyclic voltammetry (CV),⁷ ultraviolet photoelectron spectroscopy (UPS),^8,9 and inverse photoemission spectroscopy (IPES) on single component films.⁶ While these specialized techniques are not always accessible to researchers, the different measurements are rarely comparable⁴ and are not directly device relevant if energy levels are probed for isolated molecules, molecules in solution, or thin films grown on specialized substrates rather than full device stacks. In addition, many of these experimental techniques are surface-sensitive and cannot directly probe the bulk. However, studying the energetics of BHJs in device-relevant configurations is paramount to ensure consistency with device characterization techniques. This is because energy levels are strongly impacted by intermolecular interactions (Fig. 1), e.g., polarization effects that reduce the IE–EA gap in thin films compared to isolated molecules, or electrostatic interactions (e.g., van der Waals forces, dipole or quadrupole moments^8,10–13), that change with D:A mixing ratios, molecular orientation, dielectric constants, and microstructure.^9,14–18 In this context, especially charge–quadrupole interactions have gained significant interest, as they were shown to cause large electrostatic shifts in crystalline^10,11,13,19 and disordered organic films.^8,20,21 More specifically, a molecule with a non-zero quadrupole moment creates an electrostatic field that causes the energy levels of molecules in its vicinity to shift. The resulting change in IE and EA occurs in the same direction and by roughly the same amount. While the direction of the energetic shift depends on the sign of the quadrupole,⁸ the magnitude depends on the strength of the electrostatic field and hence the magnitude of an individual quadrupole, the number of quadrupoles, and the distance between the quadrupole and surrounding molecules.²⁰ With this, the precise effect of charge–quadrupole interactions on the energy levels in BHJs is dependent on the quadrupole moments of the chosen donor/acceptor molecules and D–A mixing and is generally different in the bulk than at the interface [Figs. 1(b) and 1(c)].⁸ As a result, specifically, an exciton dissociation from CT to CS states is strongly influenced by charge–quadrupole effects⁸ and either supported or hindered depending on the precise molecular interactions.

In this work, we investigate the role of D:A mixing on the energetic landscape of BHJ devices based on zinc phthalocyanine (ZnPc) or its fluorinated derivatives (F₄ZnPc, F₈ZnPc, and F₁₆ZnPc) as donors and C₆₀ as the acceptor [Fig. 2(a) and Fig. S1]. Fluorination of ZnPc molecules simultaneously shifts the IE and EA away from the vacuum level,^22,23 thereby increasing the interfacial CT state energy (and decreasing $ΔECT=ES1−ECT$⁠, the driving force for CT) in blends with C₆₀. Gradual fluorination of ZnPc also increases the molecular quadrupole moments,^8,20 resulting in stepwise changes in the electrostatic environment. In the F_xZnPc:C₆₀ blends, the donor molecules have considerable quadrupole moments [Fig. 2(a)] while the quadrupole moment of C₆₀ is zero. Consequently, the energy level shifts of F_xZnPc molecules in thin film blends are expected to directly increase with donor content and are larger in pure donor phases than at the interface with C₆₀.⁸ Similarly, C₆₀ energy level shifts are smaller (or even zero) in pure phases of C₆₀ compared to the D–A interface. For each donor molecule, we systematically vary the donor content from neat C₆₀ to neat F_xZnPc (x = 0, 4, 8, and 16) and study device performances via current density–voltage (J–V) measurements. We use sensitive external quantum efficiency (sEQE) measurements as a relatively accessible technique to study energy levels in full device stacks.^24,25 This is done by fitting the energy of singlet (⁠$ES1$⁠) and CT states (E_CT)—the most device relevant energy levels—from sEQE spectra using Marcus theory.^26–31 We find that device performances are strongly dependent on D:C₆₀ mixing ratios for all materials, with ZnPc:C₆₀/F₄ZnPc:C₆₀ and F₈ZnPc:C₆₀/F₁₆ZnPc:C₆₀ displaying similar performance trends.³² Compared to our previous work, where we studied charge transfer and charge separation in low donor content blends (D:A [5:95]) of F_xZnPc and C₆₀,³² this work explores how donor content impacts energy levels and device performance. This is especially interesting considering that aggregation-induced polarization effects and charge–quadrupole interactions are absent at low donor content when donor aggregation is suppressed.^33–35 In agreement with recent research,⁸ we find an empirical correlation between the sign of the donor quadrupole moment (Q_D ≠ 0; $QC60=0$⁠) and whether photocurrents increase or decrease with donor content. While D:A mixing series are often only used to optimize device performances, our work demonstrates that simple analyses of sEQE spectra can provide valuable insights on the energy levels of D:A blends and help uncover the underlying principles governing device performance.

We fabricated BHJ devices of F_xZnPc (x = 0, 4, 8, 16) and C₆₀ at 0, 5, 25, 50, 75, 95, and 100 wt. % donor content. All samples were fabricated via vacuum thermal evaporation, co-depositing donor, and acceptor molecules in the desired weight ratio. For comparability with other literature reports, Table S1 in the supplementary material summarizes the conversion of donor wt. % to the commonly used mol. % or vol. %. Apart from the active layer, all samples were fabricated with nominally identical device architectures and through subsequent evaporation of MoOx (3 nm, 0.05–0.1 Å/s), F_xZnPc:C₆₀ (50 nm, combined rate of 0.4 Å/s), BPhen (8 nm, 0.1–0.2 Å/s), and Al (100 nm, 1–2 Å/s). The substrates were held at room temperature during deposition, and all devices were encapsulated with a glass cover slide in inert conditions prior to measurement.

J–V measurements (Fig. S3 in the supplementary material) of the best performing (highest PCE) devices and sEQE measurements on a linear scale [Figs. 3(a)–3(c) and Figs. S6(a) and S6(c)] reveal that while devices based on neat C₆₀ show a high open-circuit voltage (V_oc), active layer absorbance is limited and charge separation is poor, leading to a low short-circuit current density (J_sc). Adding even small amounts of ZnPc drastically improves the photocurrent [Fig. S3(a)] due to improved charge separation at the D:C₆₀ interface.^24,25,36 The same increase in J_sc is observed for F₄ZnPc [Fig. S3(b)]. While the J_sc of the F₄ZnPc blends is comparable to the ZnPc blends, the V_oc is generally 0.1–0.2 V larger due to the fluorination-induced shift of the donor energy levels that reduce the interfacial energetic offset with C₆₀.^8,22,23 Compared to ZnPc and F₄ZnPc, the photocurrents of all F₈ZnPc [Fig. S3(c)] and F₁₆ZnPc [Fig. S3(d)] blends are smaller than for the neat materials, indicating that free charge generation at the D:C₆₀ interface is inefficient. Due to the similarities in device performance trends of the ZnPc/F₄ZnPc and F₈ZnPc/F₁₆ZnPc blends, we focus our subsequent discussion on comparing ZnPc:C₆₀ and F₈ZnPc:C₆₀ as representative examples. Device characterization of all F_xZnPc:C₆₀ blends is included in the supplementary material for completeness. Figure 2 (and Fig. S4) summarizes the trends in V_oc, J_sc, and fill factor (FF) across donor molecules and donor content. The V_oc of the ZnPc series is strongly dependent on D:C₆₀ mixing and first decreases for intermixed ZnPc:C₆₀ (D:A [50:50]), and then increases with higher ZnPc content [Fig. 2(b)]. This is in agreement with previous reports^9,37 and is attributed to changing E_CT (discussed again later) and increasing non-radiative losses for the intermixed blends [Figs. S12(a) and S12(b)]. In comparison, the V_oc of the F₈ZnPc blends [Fig. 2(c)] decreases from 1.05 ± 0.12 V (low donor content) to 0.66 ± 0.02 V (neat F₈ZnPc). This decrease is coupled with a strong increase in non-radiative recombination from $∼0.2$ to $∼0.5$ eV [Figs. S12(e) and S12(f)]. The J_sc [Fig. 2(d)] and FF [Fig. 2(e)] of the ZnPc blends both peaks for intermixed ZnPc:C₆₀. The F₄ZnPc system (Fig. S4, second column) follows the same trends as ZnPc, although the change in V_oc is less pronounced. In comparison, the J_sc of F₈ZnPc [Fig. 2(e)] shows the opposite trend and decreases for intermixed F₈ZnPc:C₆₀, while the FF stays roughly constant (⁠$∼0.3$⁠). The same is observed for F₁₆ZnPc [Figs. S4(d), S4(h), and S4(l)]. Rationalizing these performance trends, especially the increased (decreased) photocurrents for intermixed ZnPc:C₆₀/F₄ZnPc:C₆₀ (F₈ZnPc:C₆₀/F₁₆ZnPc:C₆₀), we consider that the J_sc is strongly dependent on active layer absorbance, charge separation, and charge carrier collection. While this work focuses on interface-related processes of free charge generation and recombination, the microstructure of the blends and implications for charge transport are briefly discussed. Through grazing incidence wide-angle x-ray scattering (GIWAXS) measurements (Fig. S13 in the supplementary material), we find that the microstructure at specific D:C₆₀ mixing ratios is largely comparable for all donor molecules. Comparing the different D:A mixing ratios, neat F_xZnPc films show preferential ordering of the molecules in the π–π stacking direction and peak broadening along the azimuthal angle, indicating a large spread in the orientation of individual crystalline domains.^32,38 The low acceptor content blends show weaker F_xZnPc scattering features but an equal spread in the orientation of the F_xZnPc domains (Fig. S13, second column). In comparison, the low donor content blends show no scattering features, indicating that even small amounts of the donor molecules effectively disrupt C₆₀ packing (Fig. S13, first column). From our GIWAXS measurements and literature reports of intermixed ZnPc:C₆₀,³⁹ the morphology of our blends can be pictured with some degree of donor/acceptor aggregation at high enough molecular concentrations but no clear phase separation. Phase separation is only clearly observed for samples that are heated during deposition,^40–42 compared to the room temperature substrates used in this study. In terms of percolation, we note that the low donor (acceptor) content blends still achieve charge transport to the electrode, even in the absence of bi-continuous percolation pathways, through mechanisms like long-range tunneling.^35,43 In summary, while the trends in J_sc and FF are certainly the result of superimposed effects, the qualitative trends between ZnPc and F₈ZnPc cannot be explained via differences in the microstructure and charge transport. Based on the strong similarities in the donor absorption profiles,³² absorption changes can also not explain the trends in photocurrent. This is confirmed by transfer matrix modeling (TMM) of the J_sc of the ZnPc [Fig. 2(d)] and F₈ZnPc [Fig. 2(e)] blends, which predicts equally high (6–10 mA/cm²) photocurrents for both donors that peak for 25–50 wt. % donor loading. While the measured photocurrents of the ZnPc blends largely agree with the simulations, aside from the lower J_sc for the low donor and low acceptor content blends, F₈ZnPc not only shows severely reduced photocurrents (⁠$∼0.5−1$ mA/cm²) but also displays the opposite trend predicted by TMM. This is likely caused by inefficient charge separation in the F₈ZnPc and F₁₆ZnPc blends. Although an in-depth discussion of exciton separation efficiencies is beyond the scope of this study, we note that charge separation occurs after exciton transfer from S₁ to CT states and relies on efficient exciton dissociation from CT to CS states. Especially, the latter is strongly influenced by electrostatic shifts, e.g., caused by charge–quadrupole interactions.^8,20 In blends with C₆₀, the energetic barrier for exciton dissociation (ΔE_diss = E_CT − E_CS) was previously found to decrease for donor molecules with negative quadrupole moments (e.g., ZnPc and F₄ZnPc), thereby assisting free charge generation, while positive quadrupole moments (e.g., F₈ZnPc and F₁₆ZnPc) increase ΔE_diss.^8,20 Since charge–quadrupole effects are dependent on film composition and scale with molecular blending,^8,20 they could help rationalize the low photocurrents of the intermixed F₈ZnPc:C₆₀ and F₁₆ZnPc:C₆₀ blends, which is again discussed later. We next turn to investigate the energetics of S₁ and CT states in our blends to further rationalize device performances.

S₁ and CT state energies provide invaluable information on molecular and blend energetics in full device stacks with complex intermolecular interactions. Figures 3(b) and 3(d) [and Figs. S6(b) and S6(d)] show sEQE spectra of the investigated systems on a logarithmic scale. Only the ZnPc blends show clearly discernible CT state shoulders. From the sEQE spectra, we also confirm that the S₁ of the donor is lower in energy than that of C₆₀, thereby defining the optical gap of the BHJ blends. To determine the donor $ES1$ and E_CT, we follow the approach presented in Refs. 44–46 and fit both the S₁ peaks and CT states in sEQE spectra via Gaussian functions. While S₁ absorption is influenced by homogeneous and inhomogeneous broadening, which leads to an overall Gaussian line shape, charge transfer reactions, such as exciton transfer from S₁ to CT states, are well described by Gaussian functions like Marcus theory.^{26–28,30,31} Fitting both S₁ and CT states is more robust than simple fitting of only CT states, especially for systems with low $ΔECT=ES1−ECT$⁠, where CT states are not easily distinguishable.⁴⁴ We note that the S₁ peak of C₆₀ is hidden in the spectra and cannot easily be fit using the same approach. Figures 4(a) and 4(c) show two example fits for intermixed ZnPc:C₆₀ and F₈ZnPc:C₆₀. For completeness, we also plot the sum of the S₁ and CT state fits, which describe the sEQE tail well in both cases. Figures 4(b) and 4(d) (Figs. S8–S10) summarize the fit results for the ZnPc:C₆₀ and F₈ZnPc:C₆₀ (F₄ZnPc:C₆₀ and F₁₆ZnPc:C₆₀) blends as a function of donor content. For blends based on ZnPc, both $ES1$ and E_CT follow the same U-shaped trend as the V_oc, resulting in a roughly constant ΔE_CT ≈ 0.3 eV across ZnPc:C₆₀ mixing ratios (Table S4). Previous work investigating energy levels of ZnPc:C₆₀ blends found the same trend in E_CT⁹ but observed a constant optical gap (and, therefore, varying ΔE_CT) across mixing ratios by tracking the S₁ peak position in thin film absorbance spectra.⁹ While this approach risks neglecting device specific effects (e.g., modified film growth on quartz compared to full device stacks), it also does not account for the broadness (captured by the reorganization energy $λS1$⁠) of the S₀ → S₁ transition and led the authors to conclude that suppressed crystallization of C₆₀ with the addition of ZnPc is the main driving force for the observed performance trends. Looking only at the position of the ZnPc absorption peaks in our data [Fig. 3(a) and for clarity Fig. S7]—corresponding to $ES1+λS1$ in Fig. 4(b)—would also suggest a roughly constant optical gap across ZnPc:C₆₀ mixing ratios,⁹ that is, significantly larger than $ES1$ determined via reciprocity of emission and absorption of neat ZnPc thin films.⁴⁷ Our approach of fitting $ES1$ directly from sEQE spectra significantly improves on this estimate by accounting for the blend- and device-specific intermolecular interactions, e.g., polarization effects and electrostatic interactions, that change with D–A mixing. With this, our results for $ES1$ of intermixed ZnPc:C₆₀ (1.53 ± 0.01 eV) and neat ZnPc (1.55 ± 0.01 eV) closely match previous reports.^47,48 We attribute the simultaneous shift in $ES1$ and E_CT with donor content to changing polarization environments. As discussed earlier, since ZnPc aggregates with preferential orientation in a π–π stacking geometry,^8,38,49,50 morphologies and, therefore, polarization environments are expected to change for different ZnPc:C₆₀ mixing ratios and aggregation regions.

In the context of quantifying voltage losses, an accurate determination of both $ES1$ and E_CT is paramount for estimating ΔE_CT and radiative recombination (Table S4 and Fig. S12).^44,47 Comparing all materials systems, we find that the trends in $ES1$ and E_CT are well reproduced by $qVocrad$⁠, the radiative limit of the V_oc, which was independently calculated via integration of the sEQE and does not rely on spectral fitting. Interestingly, radiative losses, defined as the energetic difference between E_CT and $qVocrad$⁠, are largely constant across D:C₆₀ mixing ratios. The highest radiative losses (around 0.25 eV) are observed for the ZnPc blends, which is in qualitative agreement with the comparably broad absorption onset observed in the sEQE spectra [Figs. 3(a) and 3(b)]. In agreement with the results presented earlier, our sEQE analysis also reveals roughly constant ΔE_CT across D:C₆₀ mixing ratios for the fluorinated donors. Finally, the general decrease of ΔE_CT with increasing donor fluorination (Table S4) matches our expectations of fluorination-induced shifts of donor energy levels toward the vacuum level.^8,22,23

We now reconcile the detailed analysis of the energy levels in the donor content series of ZnPc and F₈ZnPc with the molecular properties and trends in device performance. As discussed above, for both donor systems, the S₁–CT offset is roughly constant across all mixing ratios (Table S4 and Fig. S12). The energetic shift in S₁ (and hence the CT state) differs qualitatively between the two donor molecules. We attribute the simultaneous shift of the S₁ and CT states with donor content to changing polarization environments of the blends. The trends in V_oc follow the changes in the optical gap and the correspondingly lower CT state energy, which also affects non-radiative losses according to the energy gap law. For example, the S₁ and CT state energy of ZnPc is $∼150$ meV lower at 50% donor content compared to the 5% donor blend. This narrowing of the optical gap already largely explains the 400 mV smaller V_oc in the 50% vs 5% donor blends. Benduhn et al. found the non-radiative voltage losses to inversely scale with CT state energy at $∼180$ mV/eV, corresponding to an expected 30 mV larger non-radiative loss at 50% vs 5% donor content. The experimental difference in non-radiative voltage losses between the two blends is around 130 mV, which shows that the origins of non-radiative recombination are complex and cannot be explained by the energy gap law alone. Apart from polarization effects that lead to gap narrowing, quadrupole fields affect the energy levels with direct implications for device performance.^8,51 Since C₆₀ does not have a quadrupole moment due to its symmetry, only the donors generate an electrostatic quadrupole field, with a higher number of donor molecules leading to stronger fields. Electrostatic quadrupole fields decay on shorter length scales than dipoles, and energetic shifts are most significant within a few neighboring molecules.²⁰ Considering that the quadrupole moments of F_xZnPc act most strongly in the π–π stacking direction,⁸ the quadrupole field that an interfacial ZnPc molecule and an interfacial C₆₀ molecule experience are expected to be largely the same. Consequently, donor and acceptor energy levels are shifted to the same degree, and no change in the CT state energy with donor content due to the quadrupole field is expected (the narrowing via polarization was discussed in Sec. III). This is fundamentally different for non-fullerene acceptor (NFA) systems, where both molecular species carry a non-zero quadrupole moment, and donor and acceptor molecules at the interface impose different electrostatic shifts on their neighbors. As suggested in the literature, the quadrupole moment may have a more pronounced effect on the efficiency of exciton dissociation^8,51 and thus the V_oc. We expect a gradual change in the IE/EA levels when transitioning from a ZnPc domain to an interfacial ZnPc molecule⁵¹ since the quadrupole field becomes weaker closer to the interface [see Figs. 1(b) and 1(c)]. For ZnPc, Q_D < 0 and the gradual shift results in downward energy level bending. This means that holes experience a driving force on ZnPc to move away from the interface, thereby assisting exciton dissociation. Since the degree of energy level bending depends on the strength of the quadrupole field and hence the amount of donors present in the blend, we expect an increase in the driving force for exciton dissociation with increasing ZnPc content. This logic is analogous to a mechanism proposed for NFA blends^51,52 and is consistent with the observed trends in J_sc and FF when increasing the ZnPc content from the low donor to intermixed ZnPc:C₆₀. For F₈ZnPc, Q_D > 0, resulting in an upward bending of the energy levels and hence a barrier to exciton dissociation, which, following the same logic, increases with donor content. Again, this explanation is consistent with the observed J_sc trends for the first half of the F₈ZnPc content series. While the consideration of quadrupole effects describes the transition from low donor to intermixed D:C₆₀ blends well for all systems, it does not fully rationalize the decreasing (increasing) J_sc when the donor content is increased beyond 50 wt. %. We hypothesize that since charge–quadrupole effects primarily influence nearest neighbors, the effect of quadrupole moments on interfacial energy levels might saturate once sufficiently large donor aggregation regions are formed. Furthermore, we highlight again that the J_sc is a highly convoluted quantity that not only depends on the efficiency of exciton dissociation but also on absorption (discussed earlier) and may be affected by changes in the charge collection efficiency. Given the complexity of the presented dataset, elaborate modeling linking molecular and device properties and specifically investigating the influence of quadrupole moments and percolation pathways would be beneficial. In addition, coupling our studies with other specialized techniques, e.g., mobility measurements, or time-resolved spectroscopic techniques like transient absorption spectroscopy (TAS) that can quantify efficiencies of electronic transitions, would aid in disentangling the individual contributions to the photocurrent. Nevertheless, the qualitatively opposing trends in J_sc between ZnPc and F₈ZnPc, and the molecules’ opposing sign in quadrupole moment, provide a justified hypothesis.

In this work, we study the role of D:A mixing ratios on the energy levels and device performance of BHJs based on ZnPc or its fluorinated derivatives (F₄ZnPc, F₈ZnPc, and F₁₆ZnPc) as donors and C₆₀ as the acceptor. Through J–V measurements, we find that open-circuit voltages and photocurrents are strongly dependent on D:C₆₀ mixing ratios, with the ZnPc/F₄ZnPc and the F₈ZnPc/F₁₆ZnPc blends displaying similar performance trends. The device performances are rationalized via changes in molecular and blend energy levels probed by careful analysis of sEQE spectra on a linear and logarithmic scale. Our work demonstrates that fitting both singlet and CT state energies from sEQE spectra provides device-relevant information on energy level shifts and driving forces. For instance, we find that ΔE_CT, the energetic difference between the donor S₁ and CT states, is largely invariant to changing D:C₆₀ mixing ratios in our studied material systems. With this, we rationalize trends in the V_oc by changing polarization effects that likely cause energy level shifts on a molecular level. Our discussion of quadrupole moments in fullerene-based blends highlights qualitatively different expected behavior in donor content series depending on the sign of the quadrupole, which is in agreement with our experimental data. Combining these insights with future studies of charge carrier dynamics could unlock a complete picture of the energetics in these blends. Our methodology of preparing D:A mixing series and carefully analyzing sEQE spectra to investigate energy levels in real BHJ devices is equally applicable to NFAs. In this context, the role of molecular quadrupole moments should be studied further to support the rational design of new molecules and donor–acceptor pairs with improved performance.

Organic solar cells were fabricated via vacuum thermal deposition using an evaporation chamber (Creaphys, EVAP300, base pressure 10⁻⁷ mbar). Prior to deposition, the substrates were sonicated for 10 min in a solution of 2.5% Hellmanex in DI water, followed by DI water, acetone, and isopropanol. The substrates were treated with O₂ plasma for 10 min before being loaded into the evaporation chamber. The devices were fabricated on ITO-patterned glass (Eagle XG glass, 20 Ω/sq, rms roughness < 7 Å) and through subsequent evaporation of MoO_x (3 nm, 0.05–0.1 Å/s), F_xZnPc:C₆₀ (varying ratios, 50 nm, 0.4 Å/s total rate), BPhen (8 nm, 0.1–0.2 Å/s), and Al (∼100 nm, 1–2 Å/s). The ITO-coated glass was purchased from Thin Film Devices TFD, Inc., USA. Optoelectronic grade C₆₀ was purchased from Creaphys GmbH, Germany. ZnPc, F₄ZnPc, F₈ZnPc, F₁₆ZnPc, BPhen, and MoO_x were purchased from Luminescence Technology Corp in a sublimed form where possible. The substrates were held at room temperature during the layer-by-layer deposition of the device stacks. Post deposition, all samples were transferred into a N₂-filled glovebox without air exposure and encapsulated by a glass cover slide for further characterization. All devices reported in this work have an active area of 0.08 cm² as defined by the geometric overlap between the ITO and the Al electrodes.

Current density–voltage characteristics were measured under simulated AM1.5g light using a sun simulator (Abet Technologies, Sun 2000, Class AAB) and a source meter (Keithley, 2400 Source Measure Unit). The light source was calibrated using an NREL-certified KG5 filtered silicon reference diode and set to 100 mW/cm² intensity with respect to the reference diode. Scans were performed in both forward and reverse directions, typically covering a range of −1 to 1.5 V and measuring in 0.02 V steps.

Sensitive external quantum efficiency (sEQE) measurements were performed using a custom-built setup. White light from a tungsten-halogen light source (Princeton Instruments, TS-428, 250 W) was diffracted by wavelength using a monochromator (Princeton Instruments, Spectra-Pro HRS300, Triple Grating Imaging Spectrograph). Using spectral filters (Thorlabs, edge pass, and long pass filters), stray light and higher-order diffractions were removed. The light was modulated using a chopper wheel (Stanford Research Systems, SR450, Optical Chopper) before being focused onto the device under testing. The resulting photocurrent was pre-amplified (Zürich Instruments, HF2TA Current Amplifier) before being read out by a Lock-In amplifier (Zürich Instruments, HF2LI Lock-In Amplifier). The sEQE spectra were calibrated using Silicon (Thorlabs, FDS100-CAL) and Indium Gallium Arsenide (Thorlabs, FGA21-CAL) photodiodes.

Sensitive EQE spectra were fit using Marcus theory,^27,31,53 with $EQEE=fCT/E4λCTkBT×exp−ECT+λCT−E2/4λCTkBT$⁠, where f_CT is an absorption pre-factor that is related to the oscillator strength, λ_CT describes the reorganization energy, and E_CT marks the charge transfer (CT) state energy. The singlet (S₁) peak was fit first using an analogous equation and subtracted from the sEQE spectrum. The remaining tail was fit as the CT state, following the methodology outlined in Refs. 44–46.

Voltage losses were calculated following the methodology outlined in Ref. 47. The singlet (S₁) and charge transfer (CT) state energies were used as reference energies. The radiative upper limit of the open-circuit voltage (⁠$Vocrad$⁠) was calculated assuming reciprocity between emission and absorption^47,54 and via $Vocrad=kBT/q⋅ln∫Emin∞EQE⋅ϕAM1.5gdE/∫Emin∞EQE⋅ϕbbdE$⁠. Here, k_B is the Boltzmann constant, T is the temperature, q is the elementary charge, ϕ_AM1.5g is the AM1.5g light spectrum, ϕ_bb is the black-body spectrum, and E_min is minimum integration energy that was chosen such that $Vocrad$ saturates at the D:C₆₀ interface.^36,55 Radiative losses were calculated as $qΔVocrad=ECT−qVocrad$⁠, and non-radiative losses were determined via $qΔVocnr=qVocrad−Voc$⁠.

Grazing incidence wide-angle x-ray scattering (GIWAXS) measurements for all except two samples were carried out at the Surface and Interface Diffraction beamline (I07) at the Diamond Light Source (DLS) using a beam energy of 20 keV (0.62 Å) and a Pilatus 2M area detector. The sample-to-detector distance was calibrated using silver behenate (AgBeh) and determined as 664.8 mm. GIWAXS measurements of the 5 and 95 wt. % F₄ZnPc:C₆₀ blends were carried out at the European Synchrotron Radiation Facility (ESRF) at BM28 (XMAS/THE UK CRG). X-ray energy of 20 keV (0.62 Å) was used, and 2D images were recorded using a Pilatus 2M detector. Air scattering was reduced by using a flight tube with Kapton windows, and the sample-to-detector distance was determined to be 421.6 mm using AgBeh. Images were converted to 2D reciprocal space using the DAWN software package with an applied polarization and solid angle correction.⁵⁶ Reciprocal space maps are shown with a logarithmic color scale to facilitate the readers’ understanding of scattering features. The positions of the primary out-of-plane peaks were determined from conical slices between 2° and 5° of the out-of-plane axial orientation, which fitted as Lorentzian functions with a linearly varying background. The angular off-set is due to the missing wedge, which arises from the conversion to reciprocal space. The position of the in-plane π–π stacking peaks was determined from a rectangular region around the Yoneda band (roughly q_z = 0.05 Å⁻¹) and fitted in the same way. The corresponding distance in real space was found by d = 2π/q.

Additional data and results, including extended performance characterization, analysis of voltage losses, and microstructure measurements are given in the supplementary material.

A.J. acknowledges funding from the Wolfson-Marriott Graduate Scholarship from Wolfson College, Oxford, the EPRSC Doctoral Training Accounts, and the Department of Physics at the University of Oxford. P.K., T.D., and M.R. acknowledge funding from the Global Challenges Research Fund (GCRF) through Science and Technology Facilities Council (STFC) under Grant No. ST/R002754/1: Synchrotron Techniques for African Research and Technology (START). P.K. also thanks EPSRC for funding a Postdoctoral Fellowship under Grant No. EP/V035770/1. A.E.L. thanks the EPSRC for funding through the Doctoral Training Partnership under Grant No. EP/N509711/1 as well as the STFC, ISIS Neutron, and Muon facility and studentship under Grant No. 1948713.

GIWAXS measurements were performed at the I07 beamline at the Diamond Light Source (experimental session NT26630-1) and at XMaS: The UK Materials Science Facility (BM28) at the European Synchrotron Radiation Facility (ESRF; experimental session 28-01-1274). The data can be accessed at https://data.esrf.fr/ and via https://doi.esrf.fr/10.15151/ESRF-DC-772678050. We thank Didier Wermeille (University of Liverpool), Oier Bikondoa (University of Warwick), and Jonathan Rawle (Diamond Light Source) for their invaluable help and support during the respective beamtimes.

This research was funded in part by UKRI ST/R002754/1 and EP/N509711/1. For the purpose of Open Access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript (AAM) version arising from this submission.

The authors declare no competing interest.

A.J., P.K., and M.R. designed and coordinated the project. A.J. carried out all sample fabrication and device measurements. A.E.L. and T.D. performed the GIWAXS measurements and analysis. All authors contributed to the analysis and manuscript writing.

Anna Jungbluth: Conceptualization (equal); Data curation (lead); Formal analysis (equal); Investigation (lead); Methodology (lead); Project administration (equal); Software (lead); Validation (equal); Visualization (lead); Writing – original draft (lead); Writing – review & editing (equal). Pascal Kaienburg: Conceptualization (equal); Data curation (supporting); Formal analysis (equal); Investigation (supporting); Methodology (supporting); Project administration (equal); Supervision (supporting); Validation (equal); Visualization (supporting); Writing – review & editing (equal). Andreas E. Lauritzen: Data curation (supporting); Formal analysis (supporting); Investigation (supporting); Writing – review & editing (supporting). Thomas Derrien: Data curation (supporting); Formal analysis (supporting); Investigation (supporting); Writing – review & editing (supporting). Moritz Riede: Conceptualization (equal); Formal analysis (supporting); Funding acquisition (lead); Investigation (supporting); Methodology (supporting); Project administration (equal); Supervision (lead); Validation (equal); Visualization (supporting); Writing – review & editing (equal).

The data that support the findings of this study are available from the corresponding author upon reasonable request.