Increased Exciton Dipole Moment Translates into Charge-transfer Excitons in Thiophene-fluorinated Low-bandgap Polymers for Organic Photovoltaic Applications

: In this contribution, we investigate the role of thiophene fluorination in a low band gap polymer for organic photovoltaic applications. We use a combined theoretical and experimental approach to investigate charge generation and recombination dynamics, and their correlation with blend microstructure and polymer dipole moment. We find that fluorination results in an increased change in the dipole moment upon exciton formation, which is correlated with the appearance of charge-transfer excitons, as evidenced from ultrafast transient absorption studies. Fluorination also results in smaller yet purer domains, evidenced by atomic force microscopy and resonant soft x-ray scattering, and in agreement with photoluminescence quenching measurements. This change in film morphology is correlated with a modest retardation of non-geminate recombination losses. The efficient charge generation and slower recombination are likely to be partly responsible for the enhanced device efficiency in the fluorinated poly-mer/fullerene devices.


INTRODUCTION
Bulk-heterojunction (BHJ) polymer solar cells have been intensively studied in the last fifteen years because of their potential to constitute flexible, light-weight, low-cost devices for energy generation. One of the early strategies adopted to optimize power conversion efficiencies (PCEs) involved the use of low-band gap polymers which, besides absorbing light in the infrared part of the solar spectrum, can increase open circuit voltage of the devices. 1 In this regard, the design of copolymers with different electron densities or donor-acceptor character has been widely used with very positive results. [2][3][4] One way to induce a difference in monomer electron density is through the introduction of strong electron-withdrawing atoms such as fluorine, 5 which has resulted in efficiencies as high as 10.8%. 6 Fluorinated polymers often show improved PCEs compared to their non-fluorinated counterparts, although the reasons for this vary according to the system studied. [7][8][9][10][11][12][13][14][15][16][17] In some cases, increasing fluorination leads to detrimental effects on performance, an observation often explained by greater sensitivity to processing conditions due to reduced solubility, and greater tendency of fluorinated polymers to aggregate. 7,11,18,19 Multiple reasons have been suggested to explain the positive effects of fluorination on PCE. The most common of these is the inductively withdrawing nature of fluorine atoms and subsequent stabilization of the highest occupied and the lowest unoccupied molecular orbitals (HOMO and LUMO respectively). The result is an enhanced open circuit voltage (V OC ), with little or no detrimental effect on the optical band gap, which results in a net gain in efficiency, so long as the energetics still allows efficient charge separation and collection, and the morphology remains similar. However, even replacing a single hydrogen atom on the backbone repeating unit with a fluorine atom rarely leaves morphology unchanged, and can therefore affect the fill factor (FF) and short circuit current (J SC ). Fluorination tends to considerably enhance aggregation, and as a consequence blends of fluorinated polymers with fullerene acceptors typically exhibit larger domain sizes, with purer polymer-rich phases than in their non-fluorinated counterparts. [11][12][13][14] High domain purity can produce a reduction in both geminate and non-geminate recombination, as observed with time-delayed collection field measurements on a fluorinated version of PCPDTBT. 12 This facilitates the collection of charge carriers, however, large domain sizes and increased purity can also hamper exciton dissociation by increasing the effective distance the exciton has to travel before reaching the nearest acceptor molecule. 20 Additionally, molecular modeling studies have indicated that partial fluorination of the acceptor unit of the polymer can induce a larger polarization of the polymer excited state, corresponding to an increase in the change in dipole from the ground to the excited state (∆µ ge ). 13,[21][22][23][24] Yu and coworkers 25 recently suggested a linear correlation between PCE and ∆µ ge as a guideline for material design. The hypothesis behind this relationship is that the increased polarisation of excitons generated in polymers with a large ∆µ ge facilitates charge separation, by decreasing exciton binding energy, which ultimately results in an increased J sc . However, this relationship breaks down at higher ∆µ ge values, where the acceptor unit is too strong, presumably because of an excessive electronwithdrawing nature of the polymer's acceptor units, which lowers the polymer LUMO level thus reducing the energetic driving force for charge separation.
Despite the amount of work carried out on polymer backbone fluorination, studies to date have not, to the best of our knowledge, included a detailed spectroscopic analysis of charge separation and recombination as a function of excited state polarisation upon polymer fluorination in polymer/fullerene blend films. With this motivation, we set out to synthesize and investigate two polymers, fluorinated and nonfluorinated, the preliminary results of which were discussed in a previous publication. 26 Whilst many studies focus on fluorination of the 2,1,3benzothiadiazole (BT) unit, we opted for the introduction of fluorine atoms on flanking thiophenes (T), resulting in a TFDTBT acceptor unit. Copolymerisation of this unit and its non-fluorinated analogue (DTBT) with an alkylated dithieno[3,2-b:2',3'-d]germole donor unit resulted in directly comparable polymers PGeDTBT and PGeTFDTBT, which for the sake of simplicity will be referred to as F0 and F4 respectively and are shown in Figure 1. These polymers, previously reported by our group 26 , exhibited similar molecular weights and polydispersities, thus allowing their properties to be fairly compared.
In our previous publication, we demonstrated that fluorination directly results in an increased V OC , related to the lowering of the HOMO level in F4, as well as an almost two-fold increase in J SC that correlated with a stronger tendency to aggregate as observed by the vibronic features in the F4 UV-Vis spectrum, along with the formation of smoother films when mixed with PC 70 BM, as visible by AFM ( Figure S2).
In this contribution we explore the influence of thiophene fluorination on polymer conformation and backbone planarity to explain the changes observed in molecular packing, as well as the effect of fluorination upon the photophysics of neat polymer and polymer/fullerene blend films. Through a combination of time-dependent density-functional theory (TD-DFT), grazing-incidence wide angle X-ray scattering (GIWAXS) and resonant soft X-ray scattering (R-SoXS), photoluminescence quenching (PLQ) and transient absorption spectroscopy (TAS) we show that the improvement in device performance upon fluorination is likely to be related to two factors: 1) slower charge recombination from the sub-nanosecond timescale, that correlates with stronger polymer aggregation and π-stacking in F4 blends as obtained by GIWAXS and 2) the appearance of polaron-like charge transfer excitons in the F4 polymer, as observed in ultrafast TAS, which we propose to be related to the equally efficient charge generation observed in F4 blends, despite its lower driving energy for charge generation resulting from a lower-lying LUMO level compared to F0. This effect is attributed to the strong withdrawing nature of the fluorinated DTBT moiety within the polymer resulting in a high excited state polarization, as shown by our TD-DFT calculations.

Synthesis
Polymers F0 and F4 were prepared according the previously reported procedure. 26

Computational Studies
All Density Functional Theory (DFT) calculations were performed using Gaussian 09 Revision C.01, 27 at either the the B3LYP/6-311G(d) or WB97XD/6-311G(d,p) level of theory. Alkyl side-chains were replaced by a simple sp 3 methyl group to reduce computational time. Energies of the DTBT and TFDTBT units as a function of thiophene-BT dihedral angle were calculated by first optimizing the structure (B3LYP/6-311G(d)) and, using the redundant coordinate editor, running a scan of the dihedral angle in 36 steps of 10° increments. Single point energy calculations were also performed using WB97XD/6-311G(d,p) for comparison purposes. The resulting energies were converted from Hartrees to kilojoules/mol and plotted relative to the respective minima of each structure to give the graph in Figure 2. The minimum energy conformations were then used to calculate the optimised geometries of the donor-acceptor units (WB97XD/6-311G(d,p)). Excitedstates were calculated using Time-Dependent DFT (TD-DFT), and correspond to the first excited state, before relaxation.

Film Preparation
All blend films were prepared using the same optimized conditions as for active layers of the best performing devices 26 , that is, they were spin coated from 12 mg/mL 1:2 polymer to PC 70 BM solutions in o-dichlorobenzene (o-DCB), after heating overnight at 90 °C. Neat films were spun from a 15 mg/mL polymer solution. All films were kept under Nitrogen atmosphere unless otherwise stated. For GIWAXS measurements, the films were prepared in the same way as for the spectroscopy measurements. For R-SoXS, films were prepared on NaPSS-coated glass slides and subsequently floated off onto silicon nitride membranes.

Steady-state UV-Visible and Photoluminescence (PL) spectroscopy
Steady-state UV-Visible spectra were carried out on all the films used for TAS and PL, with a Perkin Elmer Lambda 25 UV-vis spectrometer detecting from 300 to 1100 nm. Steadystate photoluminescence measurements were carried out in neat and blend films of F0 and F4 with a Fluorolog FM-32 spectrofluorometer using either a visible or an infrared detector depending on the fluorescence wavelengths. All the signals were corrected for absorbance at the excitation wavelength. The magnitude of the PL quenching in blend films was used to estimate the distance L that the exciton diffuses before encountering a fullerene molecule, deconvoluting the polymer and fullerene emission to enable separate determination of their respective PLQ. The following equation was used, as reported previously: 28 where, L ex is the exciton diffusion length in the absence of quenchers (in the neat film). For the polymer exciton, this was assumed to be 10 nm, a value typical for narrow bandgap polymers. 29,30 For PC 70 BM, L ex was taken to be 5 nm. 31 We note this analysis neglects the finite size of the exciton, and assumes efficient quenching when a polymer exciton reaches fullerene acceptors. As such, it gives only a relative indication of the length scale of exciton diffusion occurring in the blend films.
Transient absorption spectroscopy. Transient absorption spectroscopy in the sub-microsecond timescales was performed using a commercial optical parametric oscillator (Oppolette) pumped by a Nd:Yag laser to generate excitation pulses with time duration of < 20 ns and excitation densities in the range of 0.4 to 30 µJ/cm 2 at 660 nm for the intensity dependent studies, and 5 µJ/cm 2 for obtaining the transient spectra at 150 ns. Probe light generated from a tungsten lamp was focused onto the sample and then sent to a single grating monochromator. A long-pass filter was placed before the sample to prevent bandgap excitation of the sample with the probe light. Transient absorption decays were recorded with the aid of a LabView program referenced to a twochannel oscilloscope, using silicon and InGaAs photodiodes (Costronics Ltd.) for detection in the visible and near-IR respectively. The time resolution of this set-up was 150 ns.
For ultrafast transient absorption spectroscopy, the experiments were carried out with a commercial setup that comprises a 1 kHz Solstice (Newport Corporation) Ti:sapphire regenerative amplifier with 800 nm, 90 fs pulses. The output of this laser was split into two parts that ultimately generate the pump and the probe pulses. The tunable pump pulse was generated in a TOPAS-Prime (Light conversion) optical parametric amplifier and used to excite the sample with energies between 1 and 3 µJ/cm 2 at 710 nm. The probe light was used to generate either a Near-IR continuum (900-1450 nm) or a visible continuum (450-800 nm) in a sapphire crystal. A HELIOS transient absorption spectrometer (Ultrafast Systems) was used for collecting transient absorption spectra (450-1450 nm) and decays up to 6 ns. The time resolution of this set-up was 200 fs. The samples were kept at all times under a Nitrogen atmosphere.

Grazing Incidence Wide-Angle X-ray Scattering (GIWAXS)
Measurements were performed at the SAXS/WAXS beamline at the Australian Synchrotron. 32 11 keV photons were used with 2D scattering patterns recorded on a Pilatus 1M detector. The sample-to-detector distance was calibrated using a silver behenate standard. Scattering patterns were recorded as a function of X-ray angle of incidence, with the angle of incidence varied from 0.05 degrees below the critical angle of the organic film to 0.2 degrees above the critical angle. The images reported were taken at an angle of 0.3 degrees, well above the critical angle and hence sensitive to the bulk of the film. Data acquisition times of 3 seconds were used, with three 1 s exposures taken with offset detector positions to cover gaps in the Pilatus detector. X-ray diffraction data are expressed as function of the scattering vector, q, that has a mag-nitude of (4ߨ ߣ ⁄ ) sin ߠ, where ߠ is half the scattering angle and ߣ is the wavelength of the incident radiation.

Resonant Soft X-ray Scattering (R-SoXS)
Resonant soft X-ray scattering was collected at beamline 11.0.1.2 of the Advanced Light Source at Lawrence Berkeley National Laboratory. 33 Horizontally polarized X-rays at 284.0 eV were aligned normal to the film surface. 284.0 eV was chosen by calculating the scattering materials contrast between the polymers and PC 70 BM. Comparison with non-resonant 270 eV photons determined that the low-q component of scattering was from roughness, and not materials contrast. Scattered photons were collected by a Princeton PI-MTE in-vacuum CCD detector with 27.6 um x 27.6 um pixels. Two scattering patterns of 100 s exposure time were collected at 30 mm and 150 mm sample to detector distances respectively, and combined in software. Two dimensional scattering patterns were reduced to one dimensional profiles by using a customized version of NIKA. 33

RESULTS
Device efficiency data for 1:2 F0/PC 70 BM and F4/PC 70 BM blends have been reported previously 26 , with F4 blends exhibiting an 80% higher short circuit photocurrent than F0 blends, (7.5 vs 13.5 mAcm -2 for F0 and F4 blends respectively) for blends of similar thickness, as well as an improved open circuit voltage (0.66 vs 0.71 V for F0 and F4 blends respectively) resulting in a power conversion efficiency of 5.47% in F4 devices, compared to 3.51% for F0 devices. We note that the F0 and F4 blend films exhibited similar absorption spectra, and absorption coefficient, such that the differences in device performance cannot be assigned to differences in light harvesting, as shown in Figure S1 in the Supporting Information. The increase in V oc was principally attributed to the ca. 0.2 eV reduction in the ionization potential, as measured by both photoelectron spectroscopy in air and cyclic voltammetry. 26 The different FFs for the two devices are suggested to result from changes in the charge collection efficiency for the two cells, as a result of different blend morphologies. We note that the overall efficiency of the devices may be partly limited by the relatively thin blend films (ca. 80 nm) utilized which may limit light absorption. Attempts to fabricate thicker films in the case of the fluorinated polymer were limited by solubility of the polymer, which is partly attributed to its high tendency to aggregate.

Conformational Analysis
The conformation of the DTBT unit, fluorinated or not, has been the subject of much debate in the recent years. 13,18,34 Indeed, the thiophene can exist with the sulfur atom either on the same or opposite side of the thiadiazole unit of BT. The substitution pattern on the flanking thiophenes has been shown to affect the preferred relative orientation. 13 We investigated the relative conformation of the two monomers DTBT and TFDTBT using DFT calculations, before performing calculations on the full donor-acceptor unit. Initially, the conformation was probed using the B3LYP functional, and a basisset of 6-31G(d), in the interest of saving computational time and to allow ready comparison to many published works which use this functional. The dihedral angle between one thiophene and the BT unit was fixed (see highlighted bonds in Figure 2), and the rest of the structure was allowed to relax to  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 the minimum energy conformation. These energies were calculated at 10° increments, covering a full rotation of a thiophene ring. The non-fluorinated thiophene appears to prefer having the sulfur atom on the opposite side of the heteroaromatic part of the BT unit (anti conformation). In contrast, the fluorinated thiophene in this conformation exhibits an energy maximum. In the latter case the closest local minima are significantly twisted out of plane by about 40°, and are therefore still rather high in energy due to reduced conjugation. Both systems also have an energy minimum with the thiophene moiety pointing in the same direction as the BT unit (syn conformation), but while the system is completely planar in the F4 case, it is twisted by about 10° in the F0 analogue, and is therefore slightly higher in energy than the anti conformation. Since the B3LYP functional suffers from an overdelocalisation error and often inadequately represents dispersive forces, we checked the results by performing similar calculations using the WB97XD functional (with a 6-311G(d,p) basis-set) which includes a long-range correction and accounts for dispersive forces. 35 In order to minimize computational time, selected dihedral angles, from 0° to 180° were calculated. As shown in Figure 2, both functionals support the switch in preferred conformation upon fluorination, as well as the high energy anti conformation for TFDTBT. The barrier to rotation between syn and anti local minima is however, significantly overestimated by B3LYP calculations, in agreement with other studies on torsional potentials. 36,37 In both cases though, the barrier is significantly lower for the fluorinated versus the non-fluorinated materials. We note that preference for the anti-anti (a-a) for DTBT agrees with the single crystal structure published for a 4-methylated analogue, 38 whereas lengthening the side chain to hexyl results in a crystal with an a-s conformation. 39 The crystal structure of the unsubstituted DTBT shows substantial conformational disorder, but with a majority a-s conformation. 40,41 In all cases, deviations from a fully co-planar backbone are observed (ca. 5-6°). This planarization and apparent switch in conformational preference can potentially be explained through hydrogenbonding type interactions (N-H and F-H), or electrostatics. 37,42 The DFT calculations show that while the hydrogen atoms present on the thiophene ring exhibit a partial positive charge (Mulliken), fluorine atoms in the same positions have a partial negative charge of a greater magnitude.   1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60 ously reported studies for the DTBT unit. 13,18 As Mulliken charges have been shown to inconsistently predict charges for certain heteroatoms, we performed calculations on the s-a conformer using the CHelpG-based electrostatic potential fit for further confirmation. 37 We indeed observed a slight difference in the magnitude of the charges, but no change in signs of the atoms of interest.
The main conclusion of the calculations is that the TFDTBT is likely to show a preference for the planar s-s conformation. Experimentally we observed that the F4 polymer has a higher tendency to aggregate in solution than the F0 polymer, and as discussed later the GIWAXS results also show a smaller πstacking distance of F4 over F0. Both results would tend to support a more co-planar backbone.
Change in dipole moment from ground to excited-state (∆µ ge ) calculations Yu and co-workers first suggested that the magnitude of differences in the ground and excited state dipole (∆µ ge ) for various donor-acceptor systems correlated with solar cell performance. They suggested that a larger ∆µ ge would lower the coulumbic binding energy of the exciton and facilitate charge separation. 8,21 They calculated ground and excited state dipoles for one repeat unit of the donor and acceptor co-monomer using the Austin model (AM1), although others have used DFT (B3LYP). 13 Recently, Ratner highlighted that accounting for all possible minimum energy conformations is crucial to understand and interpret results from DFT calculations, particularly when considering energy levels and dipole moments. 37 We therefore computed the dipoles in both the ground and excited state for the various possible conformations of each polymer using DFT and TD-DFT, both at the B3LYP level in accordance with previous reports, 13 as well as at the WB97XD/6-311G(d,p) level of theory, which as noted earlier accounts better for dispersive forces. ∆µ ge was then calculated by accounting for the x, y and z components of the dipole moment in the ground and excited states, as outlined by Yu and co-workers. 8 Table 1 shows the data for WB97XD and the data from B3YLP is shown in the Supporting Information (Table S1). Although the values of ∆µ ge are substantially different for the two different functionals (with B3YLP being consistently higher), the trend that fluorination results in a noticeable increase in ∆µ ge is consistent for both calculations. We also note that although the conformation of the DTBT and TFDTBT unit plays a role on the value of ∆µ ge , in all cases F4 consistently exhibits a higher ∆µ ge than F0 for the equivalent conformation.
When we consider the respective minimum energy conformations (a-a for F0, and s-s for F4), ∆µ ge increases from 5.53 D to 6.31 D upon fluorination (see the values in Table 1). This is mainly due to a greater change in the x-component of the dipole moment of F4, which lies along the length of the polymer backbone. This effect is predominantly attributed to fluorine atoms increasing the acceptor strength of the TFDTBT unit relative to the DTBT analogue. As a consequence, we can reasonably deduce that excitons produced on the F4 backbone will be more polarized along this axis. In the interest of establishing whether the high ∆µ ge for F4 could indeed be related to increased charge generation, picosecond-to microsecondresolved TAS measurements were performed, the results of which are presented below.

Microstructure and morphology of neat and blend materials
The morphology of the blend films was investigated by a combination of atomic force microscopy (AFM), resonant soft X-ray scattering (R-SoXS), grazing incident wide angle X-ray scattering (GIWAXS) and photoluminescence quenching (PLQ). On relatively large length scales (of the order of 100 nm), F4 and F0 appear to have quite significantly different morphologies. Both AFM and R-SoXS measurements suggest smaller domain sizes for the F4 blend compared with the F0 analogue, with R-SoXS revealing typical domain spacings (long period) of around 90 nm and 150 nm for F4 and F0 respectively ( Figures S2 and S3 in the Supporting Information). The reduced domain size for the F4 blend compared to the F0 blend is different to what has been observed for other co-polymer containing the DTBT unit, 12-14 however the increased aggregation of the F4 polymer compared to the F0 polymer may result in smaller domains. 43,44 Comparing the integrated area of the R-SoXS scattering profiles provides information about the relative purity of these domains, 45 with relative purities of 100% for the F4 blend and 68.6% for the F0 blend. PLQ results (vide infra) show that emissions of both polymers are nearly entirely quenched upon blending with PC 70 BM, suggesting very fine, nanoscale mixing of the polymers with the fullerene acceptor within the domains revealed by AFM and R-SoXS. The GIWAXS results ( Figure S4 in the Supporting Information) show only weak crystalline order in neat films, with alkyl stacking peaks just visible. Both polymers have similar lamellar stacking distances (2.3 nm) with a higher alkyl stacking coherence length for F4 (~ 8 nm) compared to F0 (~ 5 nm). π-stacking peaks are also observed, with a significantly lower π-stacking distance of 0.35 nm for F4 compared to 0.40 nm for F0; F4 also has a larger π-stacking coherence length of 4 -6 nm compared to < 2 nm for F0. Thus fluorination of the DTBT unit enhances the degree of order in the π-stacking direction, an effect also observed by other groups. 13,46 F4 also tends to adopt a more edge-on orientation than F0. The microstructural properties of F0 and F4 are more or less preserved in blends with PC 70 BM.
The overall morphological picture that emerges from these studies is that fluorination leads to smaller, yet purer domains, with increased molecular order in the form of π-stacked polymer chains.

Photoluminescence Quenching (PLQ)
In order to gain more insight in the degree of intermixing of PC 70 BM and the polymers, photoluminescence measurements in both the neat and the blend materials were performed, probing the quenching of polymer and fullerene photoluminescence quenching (PLQ) in the blend films. To obtain the polymer PLQ, we excited at 710 nm and detected in the near-IR region. As apparent from Figure 4, emission from both the F0 and F4 polymers in the blends with PC 70 BM is highly quenched (98% for F0 and 95% for F4 compared to their respective neat materials). and their respective blends with PC70BM after exciting at 710 nm, normalized at the respective neat polymer signal maximum. PL quenching was calculated as PLQ = (1 -PL blend / PL neat ). All signals were corrected for absorbance at 710 nm. Inset is the UVvis spectra of the neat and blend films.
Analysis of this PLQ using Equation 1 (and assuming a 10 nm polymer exciton diffusion length) indicates that polymer excitons diffuse only 1 -2 nm before being quenched by an acceptor fullerene for both blends. Such short diffusion distances suggest a high level of PC 70 BM intermixing within the polymer rich domains, and the absence of a significant fraction of large, pure polymer domains. Although coherence lengths as large as 8 nm in the alkyl stacking direction are observed, the overall degree of crystallinity appears to be low, with the PL quenching data reflecting the majority of polymer chains which are disordered. The slightly lower quenching observed in the F4 blend suggests a modest decrease in the intermixing between the polymer and PC 70 BM; consistent with the R-SoXS results discussed above.
For the fullerene PLQ, we excited the blend films at 475 nm and monitored fullerene singlet exciton emission from 650 to 800 nm. As can be observed in Figure 5, the quenching relative to neat PC 70 BM film is noticeably lower as compared to the polymer quenching (69% for F0 blend and 65% for F4 blend) suggesting the presence of relatively large, and relatively pure fullerene domains. Assuming a PC 70 BM exciton diffusion length of 5 nm, these PLQ data suggest that pure fullerene clusters have a size of ~ 6 nm. This low fullerene PLQ ob-tained is indicative of fullerene exciton diffusion being a significant limitation for the efficiency of photocurrent generation for the materials systems studied herein. We note however, that despite the higher polymer PLQ than the fullerene PLQ, the EQE in the blue spectral region is still higher than that of the red region. 26 If we assume that the differences in optical interference are negligible for different excitation wavelengths, these results suggest that additional losses, apart from exciton quenching, limit photocurrent generation from polymer excitons, as we discuss in the following sections. Figure 5. Steady-state Photoluminescence (PL) of a neat PC70BM film and F0 and F4 blends after exciting the fullerene at 475 nm, normalized at the respective neat fullerene signal maximum. All signals were corrected for absorbance at 475 nm.

Transient absorption spectroscopy
We used femtosecond to microsecond-resolved transient absorption spectroscopy (TAS) as a probe of exciton and polaron dynamics following polymer excitation in both neat polymer films and blends with PC 70 BM acceptor. In Figure 6 we show the transient absorption spectra (from 200 fs to 6 ns) of a) neat non-fluorinated F0 polymer, b) 1:2 F0/PC 70 BM blend, c) neat fluorinated F4 polymer and d) 1:2 F4/PC 70 BM blend. Films were excited at 710 nm with a beam intensity of 3 µJ/cm 2 , and are corrected for differences in absorbance at the excitation wavelength. These conditions assure firstly, that the excitation is selective for the polymer, and secondly, that non-linear processes are minimized, since this excitation intensity produces signals within the linear response region (analogous data taken at lower, ~ 1 µJ/cm 2 , excitation densities show similar results, see Figure S5 in the Supporting Information).
We will first present the neat spectra for both polymers, and discuss the details of exciton generation and decay, (Figures  7a and 7c) and then we will carry on discussing charge generation and recombination from the analysis of F0 and F4 blends spectra in Figures 6b and 6d.
One can observe that spectra for both F0 and F4 neat films show two main types of signals. First a negative feature from ~ 550 to 790 nm for F0 and from ~ 480 to 775 nm for F4. This negative signal corresponds to the polymers ground state bleaching, (GSB) i.e. the depletion of ground state polymer molecules after the excitation, and to stimulated emission (SE). Secondly, a positive photoinduced absorption band is apparent, extending from ~ 900 to 1400 nm in both polymers. 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60 This photoinduced absorption is assigned to S 1 → S n singlet exciton absorption, consistent with literature assignments of analogous data for other low-bandgap semiconducting polymers. [47][48][49] From the decay of these positive, photoinduced absorption signals, average decay times for the singlet excitons can be extracted; corresponding to τ F0 ≈ 90 ps for F0 and τ F4 ≈ 180 ps for F4. For both polymers, a small, long lived, residual signal is observed at 6 ns. In order to determine whether this long lived signal should be assigned to polaron or triplet states, microsecond-resolved TAS was performed on the neat polymer films in the presence of nitrogen and oxygen atmospheres, see Figures S6a and S6b in the SI. In both cases, we found small but detectable signals under both atmospheres, with similar spectra to the residual spectra observed in our ultrafast data at 6 ns. F4 exhibited a larger amplitude and slower decay kinetics compared to F0 (τ T(F4) ≈ 1.1 ± 0.07 µs and τ T(F0) ≤ 0.7 µs) when measured in nitrogen. F4's microsecond transient absorption signal is strongly quenched when exposed to an oxygen atmosphere, following literature studies, 50-53 observation of strong oxygen quenching indicates this signal should be assigned to triplet excitons. For the F0 film, the shorter lifetime and smaller signal amplitude prevented us from observing such oxygen induced quenching, although it appears most likely that this long lived signal also derives from photogenerated polymer triplet states.

Chemistry of Materials
Returning to the description of the ultrafast transient spectra, it is also apparent that the photoinduced absorption of the neat F4 film shows a pronounced, rapid blue shift along with an amplitude increase, such that a band can be observed at 1030 nm from early times. This blue-shifting process exhibits a time constant of ~ 2.3 ps (see Figure S7 and Table S2). We rule out that the blue shift corresponds to a rapid intersystem crossing, since this would result in an essentially complete quenching of the steady-state PL. We will discuss the importance of this blue shift further below. An analogous, but much weaker blue shift is also observed for F0. Figure 6. Transient absorption spectra after pump excitation at 710 nm, with a beam intensity of 3 µJ/cm 2 for a) F0 neat film, b) 1:2 F0/PC70BM blend, c) F4 neat film, d) 1:2 F4/PC70BM blend. All signals have been corrected for polymer or blend absorbance, depending on the data belonging to neat polymer or blend transient absorption, respectively. Data with dots was measured in our nanosecond to microsecond setup, 150 ns after exciting at 660 nm and were corrected to match beam intensity. Now we turn to the description of the spectral dynamics of the F0/PC 70 BM and F4/PC 70 BM blend films, and discuss charge generation and recombination after photoexcitation. In Figure 6b and 6d, broadly similar absorption features as the ones for the neat films can be observed: a negative signal in the visible assigned to GSB and a positive photoinduced absorption signal in the NIR It is apparent that the GSB signal decay is much slower than that observed for the neat films. At early times, for both blends, the spectrum of the NIR photoinduced absorption is similar to that observed for the neat films, indicative of the initial formation of polymer singlet excitons. This photoinduced exciton absorption however, is rapidly quenched, with the transient spectra rapidly evolving to a new, blue-shifted absorption signal exhibiting a maximum at ~1150 nm for F0 and ~1100 nm for F4, still present at 6 ns. We notice that the GSB negative signal has also a much larger amplitude at 6 ns compared to the corresponding neat film signal.These observations confirm the presence of long-lived species. We notice the lack of any spectral features corresponding to PC 70 BM excitons for either blend films, consistent with our excitation wavelength being selective for polymer excitation. In order to assign the nature of the long-lived blueshifted signals, we performed sub-microsecond TAS on the blend films. As can be observed in Figures 6b and 6d the spectra obtained at 150 ns is consistent with the 6 ns spectra. Moreover, the µs-TAS transients (Figures S6c, S6d and S10 in the Supporting Information) indicate that both blends exhibit oxygen-independent, power-law decays that can be assigned to non-geminate recombination of polymer polarons, as previously described in a number of TAS studies in polymer/fullerene systems. [54][55][56][57][58] For F4, this assignment is particularly clear, with blend film showing no evidence of the oxygen dependent triplet decay kinetics observed for the neat film. Therefore, the quenching of the early, 1300 nm polymer exciton absorption and blue-shifting corresponds to the formation of long-lived polymer polarons from the initial polymer excitons.
We now address in more detail the decay dynamics in the blends. We observe in Figure 6c and 6d a rapid quenching of the exciton absorption for both blend films. Figure 7 shows the corresponding decays for F0 and F4 blends at 1330 nm, a wavelength in which the exciton absorption contribution to the signal is dominant. It is apparent that both F0 and F4 exciton absorption signals decay with a time constant τ = (1.8 ± 0.1) ps. This is almost two orders of magnitude faster than the corresponding decay dynamics of the neat polymer films. Figure 7. Normalized transient absorption traces of 1:2 F0 and F4 blends with PC70BM excited at 710 nm and 3 µJ/cm 2 and probed at 1330 nm. This decay was assigned to polymer exciton quench-ing, and therefore (see text) to charge generation. The red line is a monoexponential fit to the data, with a mean lifetime of 1.8 ± 0.1 ps.
This fast quenching of the polymer singlet exciton in the blend films is in agreement with the steady-state PLQ results. For F0 blend films a similar time constant (1.7 ± 0.1 ps, see Figure S8) was observed for the rise of polaron absorption at 788 nm, a wavelength in which there is little interfering absorption from excitons in the neat film (see Figure 6a). We thus conclude that exciton separation to form F0 + /PC 70 BMpolaron pairs proceeds in this blend film with a time constant τ = (1.75 ± 0.1) ps. This analysis cannot be carried out for F4 due to the more complex spectra evolution of the neat F4 film spectra, however, we consider 1.8 ps to represent the time constant of the exciton dissociation to form F4 + /PC 70 BMpolaron pairs.
Following the rapid (~1.8 ps) evolution of the photoinduced absorption spectrum from polymer excitons to polarons, the transient polaron absorption in F0 and F4 blend films exhibit a relatively slow, decay that initiates at ~ 50 -100 ps and extends to tens of microseconds, as we show in Figure 8. It is apparent from this figure that the polaron decay dynamics, assigned to charge recombination, are approximately 4 fold slower for the F4 blend than the F0 blend, as estimated from the half-lifetime of the polaron decay. These data are monitored at 1035 nm, a wavelength where the red shift of the polaron absorption, discussed below, has minimal impact. The power law nature of these decays and their extension to microsecond timescales, shown in Figure S10, allows them to be assigned to non-geminate recombination. For F0 blend, similar decay dynamics were observed for the recovery of the GSB, (see Figure S9) consistent with the assignment of this decay to non-geminate charge recombination to the ground state. blends with PC70BM excited at 710 nm and 3 µJ/cm 2 probed at 1035 nm. This wavelength was assigned to the polymer polaron (see text) so that this decay can be associated from ~ 50 ps, with early charge recombination. The decay is shown at 1035 nm so that the red shift does not contribute to the signal.
Note that for F4, a blue-shift of the photoinduced absorption maximum is observed in both the blend and the neat films. As discussed above with respect to the blend films, this blue shift is assigned to polaron formation. The observation of an analogous spectral evolution for the neat F4 film is therefore a strong indicative of a relaxation process of the initially formed F4 excitons into excitons associated with an increased charge transfer or polaron character on the ~ 2 ps timescale. The classification of these excitons as intramoleculecular on intermolecular is unclear, however unless charge transport is highly anisotropic, this has no consequences upon charge separation or collection.
Finally we observe in Figure 6b and 6d that the polymer polaron photoinduced bands exhibit a small red shift (from ~ 40 ps) of 0.02 eV for F0 blend (from ~1170 to 1190 nm) and 0.03 eV for F4 blend (from 1070 to 1090 nm). A similar red shift of polymer photoinduced absorption has been reported previously for other donor polymers, 22,59 and has been assigned to the energetic relaxation and trapping of the photogenerated polarons.

DISCUSSION
The observations herein presented point out that upon fluorination, the photophysics of both the neat and the blend polymers are modified. An important result is the similarity in the blue shift of the polymer exciton photoinduced absorbance in the F4 neat and blends films. Since this blue shift was assigned to the polaron pair absorption in the blend film, in the neat film it is associated with the formation of excitons with a partial polaron or charge transfer character.
The formation of the charge transfer excitons in the F4 neat film is consistent with our findings of an increased exciton dipole moment upon the insertion of the fluorine atoms via the TD-DFT calculations, which is in turn agrees with the idea of a larger exciton delocalization, and thus probably a lower exciton binding energy, resulting in a higher polaron character. The presence of excitations with a high charge transfer character in neat polymers has been reported before in PTB-type of polymers in solution 22 as well as in PCDTBT and PCPDTBT oligomers in solution. 34 We observe that backbone fluorination also has an effect on the exciton dynamics of the neat polymer films. While the decay dynamics of F0 excitons occurs in ~ 90 ps, the decay of the charge transfer excitons in F4 is approximately two times slower (~ 180 ps). This could indicate that upon fluorination, species with larger charge character are not only more efficiently formed but are also stabilized within the polymer structure, which is also consistent with the larger ∆µ ge calculated in the fluorinated polymer. We note however, that the lifetime of these charge transfer excitons in the F4 neat film (180 ps) is substantially shorter than that of the separated polarons generated in the F4/PC 70 BM blend (~ 10 ns), consistent with their assignment to charge transfer excitons rather than spatially separated polarons.
We turn now to discuss charge generation. Although we do not rule out the contribution of instrument response-limited exciton quenching (charges appearing in less than 100 fs) to the early signal of the blends, the main exciton quenching and polaron formation occurs in ~ 1.7 -1.8 ps for these polymers. This timescale for exciton quenching is consistent with high, but sub-unity, photoluminescence quenching observed for these blend films.
This picosecond exciton quenching differs from the ultrafast (< 100 fs) timescale exciton reported for some other donoracceptor polymers. 8,59,60 The observation of similar charge generation kinetics in both polymer blends is rather surprising given that the LUMO level of F4 is 0.23 eV lower than that of F0. The ability of F4 to generate charges efficiently despite its lower energy offset may be related to the higher charge transfer character of F4 polymer excitons. We note that several studies have related a higher degree charge transfer character of excitons to improved charge separation properties in both organic and dye sensitized solar cells. 8,61,62 We also note that the reduced intermixing within the polymer domains in F4/PC 70 BM blend may also aid charge separation in this blend.
When comparing the polaron decay dynamics, we observe that non-geminate recombination of F4 blend polarons is at least 4-fold slower as compared to the decay of F0 blend polarons. Slower non-geminate recombination in F4 blend results in a total higher charge carrier density in the microsecond timescale: with the lowest excitation intensity (0.4 µJ/cm 2 ) F4 blends present in average, a ~ 60% higher charge density between 200 ns and 1.2 µs, a timescale relevant for charge collection in devices. However, a detailed analysis of charge collection efficiency for these devices is beyond the scope of this study. This finding is similar to the observations in a series of PTB-based polymers, where the signal of the best performing polymer, PTB7, assigned to the charge separated state has the slowest recombination time 8 compared to other nonfluorinated and differently structured fluorine-substituted polymers.
We note that the slower non-geminate recombination dynamics observed for F4/PC 70 BM blends do not appear to result from slower charge carrier mobilities. FET hole mobilities obtained for these polymers are 3 x 10 -3 and 6 x 10 -4 cm 2 Vs -1 for F4 and F0 respectively. 26 This suggests that an additional factor, such as an improved microstructure in the fluorinated F4 polymer blend, favors a spatial separation of holes and electrons and thus slows down non-geminate recombination. Indeed, for the F4 blends herein studied, it is likely that the slower non-geminate recombination is a consequence of a microstructure improvement upon fluorination. F4 blend shows a lower content of fullerene molecules intermixed within the polymer domains, as observed in the polymer PL and inferred from the R-SoXS results. This result is consistent with our observation of an increased polymer aggregation in the F4 blends compared to F0 blends via GIWAXS measurements; the origins of the improved π-stacking in the polymer can be explained by our conformer energetic analysis via DFT calculations. Indeed, the calculations indicate that the planar s-s conformation is largely preferred by polymer F4, whereas F0 does not exhibit such a preference for either conformation, and as such, it might adopt conformations which would result in a slight twist in the backbone, impeding aggregation and resulting in a higher π-stacking distance. Additionally, the increased PC 70 BM emission in F4 blends suggests that the more planar, and thus crystallized polymer backbone also results in the expelling of fullerene molecules which also contributes to spatially separate the free electron and holes and thus result in the slower non-geminate recombination observed. This slower recombination is likely to be responsible, at least in part, of the enhanced device efficiency for the F4 polymer compared to F0.