High open-circuit voltage small-molecule p-DTS(FBTTh2)2:ICBA bulk heterojunction solar cells – morphology, excited-state dynamics, and photovoltaic performance

Institute of Materials Research and En Technology and Research (A*STAR), Singapo Max Planck Research Group for Organic O Polymer Research, Ackermannweg 10, D-55 mpip-mainz.mpg.de Center for Polymers and Organic Solids, De University of California at Santa Barbara, USA. E-mail: quyen@chem.ucsb.edu † Electronic supplementary information ( in Fig. S1–S12. See DOI: 10.1039/c4ta0625 ‡ These authors contributed equally to th Cite this: J. Mater. Chem. A, 2015, 3, 1530


Introduction
The open-circuit voltage (V OC ) is one of the gures of merit determining the power conversion efficiency (PCE) of solar cells. It is also a critical parameter in electrochemical reactions such as photocatalytic splitting of water into hydrogen and oxygen using solar fuels. [1][2][3][4] What exactly determines the V OC is still intensely investigated. In fact several factors inuencing the V OC of a bulk-heterojunction (BHJ) solar cell have already been suggested, for instance the work-function difference of the device electrodes according to the metal-insulator-metal model, [5][6][7] Fermi level pinning between the metal electrode and the fullerene reduction potential, 8 the surface dipoles at the interfaces, 9,10 the formation of ground-state charge-transfer complexes between the acceptor and donor material, 11 and shunts in the photoactive layer causing leakage currents. 12 However, it appears that the V OC mainly depends on the energy offset between the highest occupied molecular orbital (HOMO) of the donor and the lowest unoccupied molecular orbital (LUMO) of the acceptor. 8,13 Based on this rather simple and empiric design rule, donor materials with deep HOMO levels have been designed to increase the V OC of BHJ organic solar cells. The V OC of the heavily researched poly(3-hexylthiophene):phenyl-C61-butyric acid methyl ester blend (P3HT:PC 60 BM) is only 0.6 V (in fact it varies between 0.58 V and 0.67 V depending on the lm processing conditions and interfacial layer structure) 14,15 as P3HT has a rather high HOMO level of À5.2 eV, whereas conjugated polymers with lower HOMO levels such as poly(N-9 00 -heptadecanyl-2,7-carbazole-alt-5,5-(4 0 ,7 0 -di-2-thienyl-2 0 ,1 0 ,3 0 -benzothiadiazole)) (PCDTBT) and poly[2,3-bis-(3-octyloxyphenyl)quinoxaline-5,8-diyl-alt-thiophene-2,5-diyl] (PTQ1) (HOMO levels of À5.5 eV and À5.7 eV, respectively) show signicantly higher V OC s of $0.88 V and 0.89 V, respectively. 16,17 Furthermore, fullerene bis-adduct acceptors have been used to increase the V OC , because of their higher LUMO level compared to the ubiquitously used PC 60 BM. For instance, replacing PC 60 BM with indene-C60 bis-adduct (ICBA) (LUMO levels around À3.74 eV vs. À3.91 eV for PC 60 BM) signicantly increases the V OC of a P3HT-based solar cell from 0.58 V to 0.84 V. [18][19][20][21] However, there are limitations in the tunability of energy levels of materials. Firstly, lowering the HOMO level of the donor without changing its LUMO level increases the optical bandgap of the donor and hence blue-shis the absorption spectrum of the photoactive layer thereby reducing the spectral overlap with the solar radiation. Secondly, if the LUMO level of the donor is lowered concomitantly to maintain the bandgap, then the effective energy offset between the LUMO levels of the donor and acceptor decreases and the probability of exciton dissociation into free charges can be reduced. 22 Similarly, liing the LUMO level of the acceptor to increase V OC also decreases the energy offset and can reduce the yield of free charges. However, the minimum energy offset required to overcome the exciton binding energy for efficient charge separation is still controversially debated and is a subject of recent studies and reports vary from 0.1 eV to 0.5 eV depending on the donor properties. 23,24 Furthermore, the reduction of the energy offset leads to the condition, in which the energy level of the triplet state is lower than that of the interfacial charge transfer state, resulting in relaxation of the charge transfer state to the triplet state. 25 Thus, while the V OC of some BHJ systems has reached values higher than 1 V, the ll factor (FF) and PCE oen remained low. 26,27 In fact, a BHJ system with decent efficiency (PCE > 5%) and V OC approaching 1 V is a very rare case. 28 Recently, we reported a BHJ solar cell with a PCE of $8% and a V OC of $0.78 V using a small molecule (SM) donor, namely 7,7 0 -(4,4-bis(2-ethylhexyl)-4H-silolo[3,2-b:4,5-b 0 ]dithiophene-2,6diyl)bis(6-uoro-4-(5 0 -hexyl-[2,2 0 -bithiophene]-5-yl)benzo[c]- [1,2,5]thiadiazole) (p-DTS(FBTTh 2 ) 2 ), and phenyl-C71-butyric acid methyl ester (PC 70 BM) as an acceptor. 29 The energy offset between the LUMO levels of p-DTS(FBTTh 2 ) 2 and PC 70 BM, however, is very large ($1 eV), as illustrated in Fig. 1. Thus, a signicant fraction of the photon energy is lost in the charge transfer process. However, there is huge potential to improve the V OC by reducing the energy offset between the LUMO levels of the donor and acceptor in this BHJ system. Hence, in the present study we investigate the photovoltaic performance of blends of p-DTS(FBTTh 2 ) 2 and ICBA, as the ICBA has a higher LUMO level than the PC 70 BM previously used. We demonstrate a high V OC ($1 V) with a PCE of $5% using the p-DTS(FBTTh 2 ) 2 :ICBA system. Despite a more than 200 mV improvement in V OC , the PCE of the p-DTS(FBTTh 2 ) 2 :ICBA system is lower than that of our previously reported p-DTS(FBTTh 2 ) 2 :PC 70 BM system mainly due to a low FF. Transient absorption spectroscopy covering a wide wavelength region (that is from the visible to the near-infrared) and spanning a wide dynamic range (from 100 fs to several ms) reveals that the formation of weakly bound charge transfer (CT) states in the p-DTS(FBTTh 2 ) 2 :ICBA system, and their subsequent recombination, plus trap-assisted recombination of free charges are the main reasons for the low FF.

Photovoltaic performance of p-DTS(FBTTh 2 ) 2 :ICBA blends
In this section, we present the photovoltaic performance of p-DTS(FBTTh 2 ) 2 :ICBA blends prepared under various processing conditions. The use of solvent additives is an effective and common method to inuence the nanomorphology of the BHJ.
To examine the effect of solvent additives on the morphology and efficiency of p-DTS(FBTTh 2 ) 2 :ICBA blends, we spin-cast lms from solutions containing 0% to 0.6% (v/v%) 1,8-diiodooctane (DIO) and compared the photovoltaic performance of the devices. The photovoltaic parameters of the p-DTS(FBTTh 2 ) 2 :ICBA solar cells processed with various contents of DIO are listed in Table 1 and their J-V characteristics under AM 1.5G irradiation at 100 mW cm À2 are shown in Fig. S1, ESI. † The results demonstrate that the device processed without any additive exhibits a PCE of only 3.01% with a short-circuit current density (J SC ) of 7.42 mA cm À2 , a V OC of 1.017 V and a FF  When the 0.4% DIO BHJ lm was annealed at 120 C, the PCE increased to 5.07% with J SC ¼ 10.45 mA cm À2 , V OC ¼ 0.962 V and FF ¼ 50.4%. In contrast to other BHJ systems, [31][32][33][34] one observes that the gures of merit of the p-DTS(FBTTh 2 ) 2 :ICBA system are not strongly dependent on the annealing temperature. The PCE remains almost unchanged in the annealing temperature range from 100 C to 130 C ( Table 2).
While the device processed without DIO exhibits a low PCE (3.01%), its PCE signicantly increased to 4.95% with J SC ¼ 9.99 mA cm À2 , V OC ¼ 1.0 V and FF ¼ 49.5% aer annealing at 100 C ( Table 2). Hence, the PCE of this device (that is without DIO, but with annealing) becomes similar to that of the device prepared with 0.4% DIO and annealing, suggesting that the impact of annealing supersedes the effect of additive in this system. We also investigated the thermal stability of the devices processed without DIO. The PCE changed from 4.95% to 4.78% by varying the temperature from 100 C to 120 C. More interestingly, the p-DTS(FBTTh 2 ) 2 :ICBA system is also less susceptible to a variation in the donor : acceptor (D : A) ratio. We varied the D : A ratio from 1.5 : 1 to 2.3 : 1 resulting in a virtually constant PCE of 4.95% and 4.94%, respectively, as shown in Table 2. Based on the photovoltaic performance, we conclude that the p-DTS(FBTTh 2 ) 2 :ICBA system is less sensitive to the processing conditions compared to other BHJ systems.
The J-V characteristics under AM 1.5G irradiation at 100 mW cm À2 and the corresponding external quantum efficiency (EQE) spectra of the champion devices fabricated under different processing conditions are shown in Fig. 2. The calculated J SC obtained by integrating the EQE data convoluted with the AM 1.5G solar spectrum deviates only by 1.5% from the experimentally determined value. The maximum EQE of $60% is obtained between 570 nm and 700 nm. A decrease in EQE at 550 nm and below can be assigned to the poor absorption of C60based fullerene in the visible region compared to C70-based fullerenes. The J-V characteristics of the other devices are displayed in Fig. S2-S4, ESI. †

Morphological characterization of BHJ lms
The interpenetration of the acceptor and donor is an important factor in determining the efficiency of a BHJ solar cell. Thus, the nanomorphology of the p-DTS(FBTTh 2 ) 2 :ICBA BHJ obtained under different processing conditions was studied by tapping mode Atomic Force Microscopy (AFM). Fig. 3 presents the AFM topographic and phase images of as-cast lms and lms  Overall, the lms processed with DIO and/or annealing reveal rather ne and evenly distributed domains of a size of tens of nanometers in conjunction with continuous interpenetrating networks without any large-scale aggregation of either the donor or the acceptor with a slight increase in the surface roughness ($1.4 nm, Fig. 3b and c). From the phase images ( Fig. 3e and f), one observes ber-like network formation for lms processed with DIO and/or thermally annealed lms. It is widely accepted that the phase separation must be sufficiently large to create percolation pathways for charge carrier transport, yet small enough to form a large interfacial area for efficient exciton dissociation and charge separation (typically on the order of tens of nanometers). It is likely that the formation of interpenetrating networks of acceptor and donor compounds without large-scale aggregation by the use of DIO and/or annealing fulls the above conditions, and hence leads to an improved J SC , FF, and PCE.
Since AFM reveals information limited to the surface of the lm only, we also probed the morphology of the bulk lm by grazing incidence wide angle X-ray scattering (GIWAXS). Fig. 4 presents 2D GIWAXS patterns of as-cast lms and lms processed with DIO and/or annealing atop of PEDOT:PSS-coated substrates. As shown in Fig. 4, 2D GIWAXS patterns of as-cast lms as well as lms processed with DIO and/or annealing exhibit a (100) peak as well as higher order scattering peaks ((200) and (300)) along the q z (normal to the substrate) axis due to a lamella-like layered structure, and the (010) peak along the q xy (parallel to the substrate) axis due to p-p interchain stacking. This implies that p-DTS(FBTTh 2 ) 2 :ICBA BHJ lms have highly ordered edge-on orientation with side chains sitting perpendicular to the substrate and the p-stacking direction is parallel to the substrate, even for the as-cast conditions. The (100) peak at q z ¼ 0.288Å À1 and (010) peak at q xy ¼ 1.744Å À1 correspond to the alkyl stacking distance of 21.8Å and p-p stacking distance of 3.6Å, respectively (note: the intense peak at q xy $ 2.16Å À1 is caused by scattering from the underlying PEDOT:PSS layer). However, the lms processed with DIO and/ or annealing have a higher scattering intensity, especially the higher order scattering peaks (200) and (300), associated with both lamella-like layers and p-p stacking, respectively than ascast lms, suggesting that the crystallinity of the lm is further improved by using the co-solvent DIO and/or annealing. The crystal correlation length (CCL) which is the estimated crystal size of the lms processed with DIO and/or annealing, calculated from the peak width, is $17 nm, which is in good agreement with the AFM images.

Comparison with PC 70 BM-based blends
The p-DTS(FBTTh 2 ) 2 :ICBA system exhibits a PCE of $5% with a high V OC of $1 V under optimized conditions. Compared to our previously reported p-DTS(FBTTh 2 ) 2 :PC 70 BM system, which showed a PCE of $8% with J SC ¼ 14.2 mA cm À2 , V OC ¼ 0.775 V and FF ¼ 72.8%, the PCE of p-DTS(FBTTh 2 ) 2 :ICBA is still lower even under optimized preparation conditions, despite the more than 0.2 V increase in V OC . The decrease in PCE originates from a low J SC ($10 mA cm À2 ) as well as moderate FF ($50%). It is straightforward to understand that a lower absorption of the p-DTS(FBTTh 2 ) 2 :ICBA photoactive layer compared to that of p-DTS(FBTTh 2 ) 2 :PC 70 BM, due to the lack of the absorption of ICBA in the visible region, causes a decrease in J SC (Fig. S5 and S6, ESI †). However, the origin of the lower FF is not so clear. The dark J-V curve of a p-DTS(FBTTh 2 ) 2 :ICBA device exhibited a high rectication ratio and low leakage current in turn ensuring that there is a high quality interface between the BHJ photoactive layer and the electrodes and that the low FF is not due to the interfacial layers (Fig. S7, ESI †). Thus, to gain insight into the origin of the low FF in the p-DTS(FBTTh 2 ) 2 :ICBA system, we studied the excited states dynamics and related charge carrier recombination mechanisms of these two systems by ultrafast and ns to ms transient absorption spectroscopy.   ground-state bleach present immediately aer excitation. In addition to the ground-state bleach, both lms show a negative feature peaking at 1250 nm at early delay times, which we assigned to exciton-induced absorption of the singlet excitedstate of p-DTS(FBTTh 2 ) 2 . This assignment is further conrmed by TA measurements on a neat p-DTS(FBTTh 2 ) 2 lm (Fig. S8, ESI †). In both lms, a photoinduced absorption at shorter wavelengths, precisely between 900 and 1100 nm, appears on a timescale of tens to hundreds of picoseconds, which we assigned to the charge carrier-induced absorption of p-DTS(FBTTh 2 ) 2 molecules (see Fig. 5).
A noteworthy difference between the two investigated blends is the occurrence of a peak at 720 nm. This is attributed to electro-absorption (EA) as reported and discussed recently by Friend and co-workers. 35 They demonstrated that a large EA signal is caused by long-range charge separation across the donor-acceptor interface and is therefore an indication of the formation of free charge carriers. Clearly, this feature is less pronounced in p-DTS(FBTTh 2 ) 2 :ICBA blends compared to p-DTS(FBTTh 2 ) 2 :PC 70 BM blends. Thus, it appears that blends containing PC 70 BM exhibit more efficient generation of spatially separated (free) charges compared to ICBA blends. We analyzed the TA data by means of multivariate curve resolution (MCR) analysis as recently reviewed by us 36 and also applied to other donor : acceptor systems. 37 The transition of the initial excitonic state (component 1) to the charge-separated state (component 2) can be clearly seen as shown in Fig. 6. Fitting the concentration proles obtained by MCR analysis showed an inverse rate constant between 10 and 40 ps, depending on the excitation intensity used. This is consistent with TA measurements in the nearinfrared (NIR) spectral region on the ps to ns timescale, where the ICBA sample showed a fast and intensity-independent decay in the spectral region around 1050 nm, which is related to charge-induced absorption as shown in the upper panel of Fig. 7. A straightforward explanation for this observation is the geminate recombination of tightly bound CT states at the interface, which did not manage to split into free charge carriers. In contrast, the PC 70 BM sample showed a rather constant TA signal at low excitation intensities. This is further supported by MCR analysis of the NIR TA data, which is presented in Fig. S9, ESI. † Component 1 is assigned to singlet exciton-induced absorption in agreement with measurements on the pristine donor material (see Fig. S8, ESI †). Component 2 is assigned to charge-induced absorption according to spectra obtained from a chemically oxidized p-DTS(FBTTh 2 ) 2 lm (see Fig. S10, ESI †). From the MCR analysis it appears that only a small fraction of charges recombine on the early timescale up to 4 ns at low excitation densities in the p-DTS(FBTTh 2 ) 2 :PC 70 BM blend. However, we note that the ground-state bleach does not remain constant on the same timescale. This indicates that either a fraction of the singlet excitons decay without being quenched at the interface or during the charge transfer process at the interface or alternatively that a negative charge-induced absorption signal is superimposed on the ground-state bleach and hence reduces the positive ground-state bleach signal. This interpretation is supported by MCR analysis of the visible data as shown in Fig. 6. Here, component 2 is assigned to the spatially separated charges, as the EA signal could clearly be observed. Component 1 is assigned to p-DTS(FBTTh 2 ) 2 excitons with a ground-state bleach and additionally a strong contribution from the stimulated emission ranging from 750-850 nm. In contrast to the rather constant signal observed for p-DTS(FBTTh 2 ) 2 :PC 70 BM, the concentration prole of the second (charge carrier-related) component in p-DTS(FBTTh 2 ) 2 :ICBA decreases by 50-70% on the short timescale.
The TA data also allowed us to draw some conclusions on the nanoscale morphology of the investigated samples. The evolution of the spectra at the 0-0-transition on the short timescale (see lower panels of Fig. S11, ESI †) as well as the evolution of the charge-induced absorption peak obtained by MCR analysis from both the visible and the NIR data ( Fig. 6 and S9 ESI, † lower panels, grey solid lines) implies that exciton diffusion plays a more dominant role in p-DTS(FBTTh 2 ) 2 :PC 70 BM blends compared to p-DTS(FBTTh 2 ) 2 :ICBA, as the respective maxima are reached aer a longer delay time in the PC 70 BM containing blends. A further indication is the amplitude of the charge carrier-induced absorption observed directly aer excitation and obtained by MCR analysis (lower panels of Fig. S9, ESI †).
The initial carrier-induced absorption in p-DTS(FBTTh 2 ) 2 :PC 70 BM is below 25% with respect to the maximum charge carrier concentration at later times, whereas in p-DTS(FBTTh 2 ) 2 :ICBA blends more than 50% of the signal is obtained from the very beginning suggesting a more intimate mixing of donor and acceptor domains and thus a larger fraction of ultrafast charge transfer. This is supported by the AFM images and GIWAXS experiments as presented above. In fact, the domain size in the p-DTS(FBTTh 2 ) 2 :ICBA system is $17 nm, whereas p-DTS(FBTTh 2 ) 2 :PC 70 BM blends show domain sizes of up to 40 nm as reported recently. 38 This can be explained by the stronger tendency of PC 70 BM to cluster and crystallize as the steric hindrance introduced by the side groups is less pronounced than in ICBA. We note also that Gelinas et al. recently observed two charge generation pathways for p-DTS(FBTTh 2 ) 2 :PC 70 BM blendsan ultrafast component contributing about 70% weight with an inverse rate constant of $82 fs and a second component contributing 30% of the amplitude, which was diffusion-limited and had an inverse rate constant of $22 ps. A very similar inverse rate constant of 26.6 ps is also observed in our TA measurements, however, only at the highest excitation uence, while lower excitation uences resulted in values as high as 44.4 ps. This is indicative of some exciton-exciton annihilation. The faster femtosecond component could not be resolved by our setup, as it is below the temporal resolution of $200 fs. However, MCR analysis suggested a smaller fraction of ultrafast charge generation, as the initial signal amplitude related to charges is below 25%.

Charge recombination in p-DTS(FBTTh 2 ) 2 :ICBA and p-DTS(FBTTh 2 ) 2 :PC 70 BM
On the ns to ms timescale the TA spectra exhibit similar features as on the ps to ns timescale, specically the EA signal is also observed (see Fig. 8). The EA signal is still more pronounced in the case of p-DTS(FBTTh 2 ) 2 :PC 70 BM blends, again indicating that a higher fraction of free charge carriers contribute to the signal in the more efficient device.
Furthermore, the TA signal of p-DTS(FBTTh 2 ) 2 :ICBA in the spectral range between 1055 and 1080 nm is signicantly longer-lived (up to 10 ms) and its decay is only weakly intensity dependent over the entire dynamic range (Fig. 9, upper panel). In contrast, the TA signal of p-DTS(FBTTh 2 ) 2 :PC 70 BM is vanished aer 1 ms and shows a strongly intensity dependent decay (Fig. 9, lower panel). It has been reported previously that ICBA blends tend to form trapped CT states at the interface that are loosely bound, but in which charges still experience their mutual coulombic attraction. 39 Decay times between 60 and 100 ns were reported for this type of geminate recombination. Additionally, the reduced tendency to aggregate, which is inherent to ICBA compared to PCBM, can lead to the formation of small isolated fullerene domains. CT states formed within these domains also decay via a quasi-geminate recombination mechanism. 40 Aggregated fullerene domains have also been assigned to be "energetic sinks" that drive charge separation in donor-acceptor blends. ICBA largely lacks this type of thermodynamic driving force thus reducing efficient charge generation. 41 Taking the extended lifetime into account, losses via recombination of trapped CT states appear to be in p-DTS(FBTTh 2 ) 2 :ICBA blends.
Nonetheless, both blends produce a relatively high amount of extractable charge carriers as conrmed by a PCE exceeding 5%. This is supported by the dynamics of the photo-induced absorption (PIA) between 720 and 1000 nm (see lower panels of Fig. S12, ESI †). Here, both blends show intensity dependent dynamics. The discrepancy between intensity dependent and intensity independent dynamics observed for different spectral regions might result from convoluted free as well as bound charge-induced absorption spectra. Interestingly, both blends show only a weak intensity dependence of the dynamics of the ground-state bleaching (see the upper panel of Fig. S12, ESI †).
The ns to ms data of the p-DTS(FBTTh 2 ) 2 :PC 70 BM blend was further analyzed using a previously reported two-pool model that combines an intensity independent geminate decay of coulombically bound CT states (rst pool) and an intensity dependent non-geminate decay of spatially separated charges (second pool). The t results are displayed in Fig. 10, where the open symbols represent the experimentally measured data and the solid lines correspond to the ts of the data according to the two pool model. However, this model could not be applied to the p-DTS(FBTTh 2 ) 2 :ICBA. While some intensity dependence  was observed in the range of 900-950 nm, a signal offset was also observed, indicating that the formation of trapped CT states inuences the kinetics (see the lower panel of Fig. S12, ESI †) in this spectral range. Therefore, an analysis using only two pools would not be meaningful due to the likely superposition of spectral features from different species with unknown concentrations and cross-sections.
The results obtained for p-DTS(FBTTh 2 ) 2 :PC 70 BM blends show that a large fraction of free charge carriers are present in the sample aer excitation, precisely $85%, that can, in principle, contribute to the device photocurrent. We note that the fraction of free charge carrier formation of 85% is in line with the high IQE of the blend of around 85% indicating that those charges can be extracted as photocurrent. Furthermore, the inverse geminate recombination rate is determined to be $2 ns, which is very similar to previously reported lifetimes of geminate pairs in polymer-fullerene blends as they were observed for example in samples prepared with P3HT, 42 PCDTBT, 43 PCPDTBT, 44 and PPE-PPV. 45 The ts also yield an effective bimolecular recombination (BMR) coefficient of 1.25 Â 10 À11 cm 3 s À1 obtained by recalculating the extracted non-geminate recombination coefficient to a bimolecular recombination coefficient at a charge carrier density of 5 Â 10 15 cm À3 . This value is rather high compared to other polymer:fullerene blends, such as P3HT:PC 60 BM or PCDTBT:PC 60 BM, 43 which have BMR coefficients about two to one order of magnitude lower, respectively. Interestingly, charge carrier mobility measurements using the space charge limited current (SCLC) model revealed a rather high and balanced electron and hole mobility of 1.08 Â 10 À3 cm 2 V À1 s À1 and 1.20 Â 10 À3 cm 2 V À1 s À1 , respectively, in the p-DTS(FBTTh 2 ) 2 :PC 70 BM blends. In fact, mobilities reported for PCDTBT and P3HT are in the range of 10 À5 cm 2 V À1 s À1 , 46,47 and thus two orders of magnitude lower than the hole mobility of p-DTS(FBTTh 2 ) 2 . This could very well explain the larger effective BMR coefficient in p-DTS(FBTTh 2 ) 2 :PC 70 BM blends as, in principle, charge carriers with a higher mobility more oen encounter the donoracceptor interface where they can recombine. However, the comparably high mobility of charges also explains why the solar cells still exhibit decent efficiencies, despite fast non-geminate recombination. In fact, the high mobility allows fast carrier extraction from the photoactive layer in operating solar cells, thereby shiing the competition between charge carrier extraction and non-geminate recombination towards carrier extraction.

Conclusions
We demonstrate high V OC ($1 V) BHJ solar cells with a PCE of $5% using the p-DTS(FBTTh 2 ) 2 :ICBA system. While compared to the previously reported p-DTS(FBTTh 2 ) 2 :PC 70 BM system a more than 0.2 V improvement in V OC was achieved, the overall PCE remains lower than that of the p-DTS(FBTTh 2 ) 2 :PC 70 BM system. Transient spectroscopic experiments on p-DTS(FBTTh 2 ) 2 blended with either PC 70 BM or ICBA revealed that using PC 70 BM as an acceptor leads to more efficient generation of free charges. In addition, so modeling of the TA data by MCR analysis suggested that charge separation is not only ultrafast but also in large part diffusion-limited. We note that the MCR analysis also provides limited information about the sample morphology and domain sizes, which are in line with AFM and XRD studies on these blends. Furthermore, we nd that the charge carriers in p-DTS(FBTTh 2 ) 2 :PC 70 BM decay via an intensity-dependent recombination, while the dynamics in p-DTS(FBTTh 2 ) 2 :ICBA blends are only weakly intensitydependent implying the recombination of long-lived CT states and trapped charges in the latter. A closer analysis of the ns to ms recombination dynamics of photogenerated charges in the p-DTS(FBTTh 2 ) 2 :PC 70 BM based devices by a two-pool model suggests that about 85% of the charges are free and can potentially contribute to the photocurrent, which is in line with the internal quantum efficiency of these blends.
All in all, the measurements provide insight into the question why blends consisting of p-DTS(FBTTh 2 ) 2 and PC 70 BM outperform those using ICBA as an acceptor with PCEs of 8.01% and 5.07%, respectively. Moreover, the present study supports recent ndings that aggregated (crystalline) fullerene domains are benecial for charge separation at the donor-acceptor interface. 39,41 In contrast, blends containing ICBA form smaller domains giving rise to long-lived CT states and carrier trapping. Whilst previous work suggests that one crystalline component is sufficient to oppress the formation of bound states, we found that although 2D-GIWAXS experiments suggest a pronounced crystallinity of ICBA, it is not sufficient to generate as many free charges as in PCBM blends. Further experiments are required to better understand the initial charge separation such as ultrafast mobility measurements using different fullerene derivatives to evaluate the role of the initial carrier mobility of different acceptors in the charge separation process.

Device fabrication and characterization
The ITO-coated glass was rst cleaned by detergent, acetone and isopropyl alcohol with ultra-sonication sequentially, followed by the UV/ozone treatment for 10 min. The hole transport material of PEDOT:PSS (Clevious PH) was spin-coated at 5000 rpm for 40 s to obtain a lm thickness of $30 nm. The p-DTS(FBTTh 2 ) 2 :ICBA blend solution was prepared from a total of 35 mg mL À1 in chlorobenzene with various DIO amounts. The solution is dissolved at 60 C overnight. Before spin-casting, the solution was heated at 90 C for 15 min. The BHJ lm was obtained from spin-casting the solution at 2000 rpm for 45 s. The lms were baked to 80 C for 10 min to evaporate residual solvent and then annealed between 100 C and 130 C for 5 min. The thickness of the BHJ lm is $100 nm as determined using a prolometer. Then, the Ca layer was thermally evaporated with a thickness of 20 nm, and the Al cathode was continuously deposited to $80 nm under the vacuum condition of 4 Â 10 À6 Torr. The fabricated solar cells were encapsulated with epoxy and cover glass. The J-V characteristics of the solar cells were measured using a Keithley 2400 source meter unit. The light source was calibrated by using silicon reference cells with an AM 1.5G solar simulator with an intensity of 100 mW cm À2 . During the testing, an aperture with an area of 4.5 mm 2 was used to accurately measure the performance of solar cells. All solar cells were tested in ambient air. The EQE was measured using a QE measurement system (PV measurements, Inc.) aer monochromatic power calibration to conrm the J SC value.

Thin lm characterization
The nano-morphology of the pristine p-DTS(FBTTh 2 ) 2 lm and p-DTS(FBTTh 2 ) 2 :PC 70 BM BHJ lm were investigated by AFM (AFM Asylum MFP3D) to characterize the surface morphology. 2D grazing-incidence wide-angle X-ray scattering (GIWAXS) measurements were performed at the Stanford Synchrotron Radiation Lightsource (SSRL) on Beamline 11-3, with a MAR345 image plate area detector, at 12.7 keV incident photon energy, and at incident angles of 0.10-0.12 . Thin lm illumination occurred in a helium atmosphere to minimize X-ray beam damage. The lms were spin cast on PEDOT:PSS coated glass for AFM and GIWAXS measurements.

Transient absorption spectroscopy
Transient absorption (TA) measurements were performed with a home-built pump-probe setup. To measure in the time range of 1-4 ns with a resolution of $100 fs, the output of a commercial titanium:sapphire amplier (Coherent LIBRA-HE, 3.5 mJ, 1 kHz, 100 fs) was split into two beams that pumped two independent commercial optical parametric ampliers (Coherent OPerA Solo). One optical parametric amplier (OPA) was used to generate the tunable excitation pulses in the visible range, while the second OPA was used to generate the seed beam for white-light generation. For measurements in the spectral range between 550 and 1100 nm a 1300 nm seed of a few mJ was focused into a c-cut 3 mm thick sapphire window for white-light generation. Mostly reective elements were used to guide the probe beam to the sample to minimize chirp. The excitation pulse was chopped at 500 Hz, while the white-light pulses were dispersed onto a linear silicon photodiode array, which was read out at 1 kHz by home-built electronics. Adjacent diode readings corresponding to the transmission of the sample aer an excitation pulse and without an excitation pulse were used to calculate DT/T.
For measurements in the time range between 1 ns and 1 ms with a resolution of 600 ps, the excitation pulse was provided by an actively Q-switched Nd:YVO 4 laser (AOT Ltd. MOPA) at 532 nm. In this case, the delay between the pump and probe was controlled by an electronic delay generator (Stanford Research Systems DG535). TA measurements were performed at room temperature under a dynamic vacuum of <10 À5 mbar.
For TA measurements in the NIR spectral range covering 1100-2000 nm, a 2100 nm seed was used to generate white-light in an yttrium vanadate window. Furthermore, a dichroic mirror was used to separate the residual seed beam (idler of the OPA at 2100 nm) from the broadband NIR supercontinuum. The NIR pulses were dispersed onto a Peltier-cooled 512 pixel long linear extended InGaAs array (Entwicklungsbüro Stresing) and read out as described above.
MCR-ALS analysis is a so-modeling technique that decomposes a three-dimensional data matrix into spectra and the corresponding concentration proles. The data here can be described by two components unraveled by singular-value decomposition (SVD).