Bulk heterojunction morphology of polymer:fullerene blends revealed by ultrafast spectroscopy

Morphology of organic photovoltaic bulk heterojunctions (BHJs) – a nanoscale texture of the donor and acceptor phases – is one of the key factors influencing efficiency of organic solar cells. Detailed knowledge of the morphology is hampered by the fact that it is notoriously difficult to investigate by microscopic methods. Here we all-optically track the exciton harvesting dynamics in the fullerene acceptor phase from which subdivision of the fullerene domain sizes into the mixed phase (2–15 nm) and large (>50 nm) domains is readily obtained via the Monte-Carlo simulations. These results were independently confirmed by a combination of X-ray scattering, electron and atomic-force microscopies, and time-resolved photoluminescence spectroscopy. In the large domains, the excitons are lost due to the high energy disorder while in the ordered materials the excitons are harvested with high efficiency even from the domains as large as 100 nm due to the absence of low-energy traps. Therefore, optimizing of blend nanomorphology together with increasing the material order are deemed as winning strategies in the exciton harvesting optimization.

. Conceptual representation of the spectroscopic technique used. The PC 71 BM phase is selectively excited by the ultrafast laser pulse (red). The photogenerated exciton (the blue-purple circle) diffuses to the interface where it dissociates into charges via hole transfer (green straight arrow) to the polymer phase (orange spaghetti). The number of accumulated holes at the polymer phase is probed by a delayed IR probe pulse via polaron-induced absorption.
wavelength below the bandgap of the polymer where PC 71 BM has a significantly higher absorption coefficient (680 nm for RRa-P3HT and RRe-P3HT, and 630 nm for MDMO-PPV, see Supplementary Section 1 for details).
PC 71 BM exciton dissociation into charges was monitored by probing the charge-induced (polaron) PIA of the polymers in the mid-IR region 44,45 . For this, the wavelength of the probe IR pulse was set close to the maximum of the low-energy polaron band at ~3 μ m for all three polymers (see Supplementary Section 2 for details). As the exciton is harvested (i.e. reaches the interface), the polaron absorption increases proportionally to the amount of charges (holes) at the polymer. By changing the delay between the excitation and the probe pulses, exciton diffusion preceding exciton dissociation is monitored in the real time 35,36 . Note that after having reached the interface, the excitons do not necessary produce free charges but also the charge-transfer (CT) states which eventually either dissociate into free charges or geminately recombine in a ns timescale 39,[46][47][48][49] . However, from the point of view of the PIA response, the exact route of the exciton dissociation makes no difference as even the (interfacially) bound charges produce a similar PIA signal 50 . Therefore, in both cases the gradual build-up of the PIA signal reflects the diffusion time needed for the PC 71 BM excitons to reach the interface.
Dynamics of the exciton dissociation into charges are shown in Fig. 2 for the blends with different PC 71 BM weight ratios. The transients at low and high PC 71 BM loads were corrected for the weak pristine polymer response due to the finite excitation contrast, and IR response of the PC 71 BM excitons, respectively (see Supplementary Section 3). All transients were also normalized to the PC 71 BM absorption at the excitation wavelength so that the transient amplitudes represent the charge yield per absorbed photon (i.e. exciton harvesting efficiency) to allow for direct comparison of the transient amplitudes at different PC 71 BM loads.
The exciton harvesting dynamics for the blends with the three polymers have a number of similar features that can be summarized as follows: for the blends with low PC 71 BM content, the transients exhibit a large amplitude and a rapid rise time (< 1 ps), whereas for the blends with high PC 71 BM content the amplitudes are decreased (except of the RRa-P3HT blends) while the rise of the response becomes substantially slower, up to 100 ps. The latter dynamics are assigned to the PC 71 BM exciton diffusion followed by the dissociation to charges at the PC 71 BM-polymer interface via hole-transfer process 26,40,42 . We attribute increasing rise time to variations in the PC 71 BM domain size: the larger the PC 71 BM domains, the longer it takes for excitons to reach an interface and more excitons are lost.
The exciton harvesting dynamics are quite analogous in MDMO-PPV-and RRa-P3HT-based blends with low PC 71 BM content (< 40%). The similar timescale of the initial signal build-up combined with the close-to-unity amplitudes point to the nanomorphology of the mixed phase with a phase separation of ~10 nm. At higher PC 71 BM concentrations a dramatic drop of the signal amplitude is observed with the signal reducing to naught at > 70% PC 71 BM concentration. This indicates the formation of large PC 71 BM domains 51 with the size much larger that the exciton diffusion length (i.e. ~10 nm) separated from the mixed phase. The sharp decrease of the signal amplitude points to an increase of the volume fraction of the large domains, which reaches almost 100% in blends with > 70% PC 71 BM content (i.e. the polymer and fullerene phases are fully separated).This is consistent with the known property of the MDMO-PPV-based blends to form large fullerene domains above a certain acceptor weight fraction [52][53][54] .
For the RRe-P3HT-based blends, the exciton dissociation dynamics are different. For low PC 71 BM content (< 40%), the initial build-up of the signal is significantly faster as compared to RRa-P3HT-and MDMO-PPV-based blends. This indicates extremely fine intermixing of polymer and PC 71 BM in the mixed phase, probably even isolated PC 71 BM molecules dispersed in the polymer matrix. At higher PC 71 BM loadings, the initial build-up slows down indicating the coarser intermixing within the mixed phase. Simultaneously, the decrease of the exciton harvesting is observed, similarly to the MDMO-PPV blends, but to a significantly smaller extent. The observed difference in dynamics between the RRa-P3HT and the RRe-P3HT originates from the different morphology: the blends with RRa-P3HT are completely amorphous while in blends with RRe-P3HT semi-crystalline domains of RRe-P3HT are formed prior to the aggregation of PC 71 BM 55 . Hence, the PC 71 BM molecules are pushed outside the RRe-P3HT nanocrystals 56 to aggregate into the domains. Therefore, we assign exciton losses (Fig. 2c) in the blends of RRe-P3HT with 40-60% of PC 71 BM to the formation of the PC 71 BM domains with sizes much larger than the exciton diffusion length.
At 70% of PC 71 BM, the exciton harvesting suddenly increases which indicates an abrupt change in the RRe-P3HT:PC 71 BM nanostructure. Simultaneously, around these blend compositions the absorption shoulder in the red spectral region, which is associated with the absorption by the RRe-P3HT nanocrystals, vanishes (see Supplementary Section 4). Additionally, the GIWAXS data show a significant change of the blend morphology at 70% PC 71 BM contents (see Supplementary Section 5). All these point to disruption of the RRe-P3HT nanocrystals 57,58 for high PC 71 BM load 57 .
The results of PIA measurements were independently verified by the time-resolved PL quenching technique 19,30 (Supplementary Section 7). Due to intrinsic limitations of the PL technique such as spectral overlap of PL from PC 71 BM, polymers and CT states, and limited time resolution (~5 ps), it is nearly impossible to quantitatively characterize the PC 71 BM domain sizes. Nonetheless, the case of MDMO-PPV-based blends allows the direct comparison of PL quenching efficiency with exciton PIA harvesting efficiency, to produce an excellent match (Supplementary Section 7, Supplementary Figure 13). This lends additional support to the proposed PIA method.
Summarizing the discussion above: despite some similarities, the three PC 71 BM:polymer blends exhibit very different exciton harvesting dynamics as a function of the blend composition. Interestingly, the exciton harvesting is sensitive to subtle changes in the morphology as is for instance shown by the changes of the dynamics upon disappearance of nanocrystals in RRe-P3HT. This clearly indicates that a more detailed insight in the characteristic sizes of the mixed phases and the PC 71 BM domains can be obtained through modeling of the experimental data.

Characterization of the nanomorphology from Monte Carlo simulations. Monte Carlo (MC) sim-
ulations for modeling exciton dynamics have the important advantage over analytical description 4,5,30,59 that they allow for inclusion of energetic disorder 60,61 , which cannot be neglected for the solvent-processed materials and blends. We modeled the exciton diffusion as random hopping in disordered PC 71 BM domains of a mixed phase of spherical domains with diameter d m and large domains with larger diameter d c and a volume fraction f (see the Methods section for details).
The MC simulations reproduce the experimental data fairly well (Fig. 2, solid lines). The estimated domain size in the mixed phase varies from 2 nm to 15 nm depending on the blend composition and the particular polymer (Fig. 3). In the low (< 40%) PC 71 BM load blends only the mixed phase is present with typical PC 71 BM domain sizes of 6-8 nm in amorphous RRa-P3HT and MDMO-PPV polymers, and 2-3 nm in RRe-P3HT-based blends.  With the increase of PC 71 BM load, coarsening of the mixed phase is observed in RRa-P3HT and RRe-P3HT based blends. In contrast, in MDMO-PPV blends higher PC 71 BM load does not result in any significant change in domain sizes of the mixed phase but leads to explosive growth of extremely large PC 71 BM domains (up to 1 μ m, Fig. 4), which results in dramatic decrease of the PIA signal (Fig. 2b). Interestingly, the typical domain size of the mixed phase does not significantly increase with increasing of PC 71 BM content even when the large domains begin to form. In contrast, in MDMO-PPV blends with high PC 71 BM content the fine mixed phase coexists with the separated PC 71 BM domains, which volume share depends on PC 71 BM concentration (Fig. 4a).
The sizes of large PC 71 BM domains obtained from the MC simulations and independently from AFM for the MDMO-PPV blends and from TEM/GISAXS for the RRe-P3HT blends (see Supplementary Sections 5, 6 and 8-10) are summarized in Fig. 4b. MC simulations yield only the minimal sizes of the large PC 71 BM domains from the long-time exciton harvesting dynamics, while the volume fraction is straightforwardly obtained from decrease of the signal amplitude. The sizes of the large domains derived from the MC simulations match reasonably well the independently measured values for the RRe-P3HT based blends. In the MDMO-PPV case the deviations are quite substantial because for the proposed technique 0.1 and 1 μ m size domains look identical as no excitons are harvested from either of them. Note, however, that the PC 71 BM domain sizes in the mixed phase ( Fig. 3) lay below either attainable resolution (AFM) or contrast (TEM) of the conventional methods but are readily captured by spectroscopic means.
For the P3HT:PC 71 BM blends with high PC 71 BM loadings (> 40%), the blend composition was also independently verified by GISAXS and GIWAXS (Supplementary Sections 5, 6) 62 . For the mixed phase, the domain sizes of 2 nm and 15 nm with different shares were observed which is in good agreement with the current two-domain model (Fig. 3c). The sizes of large domains were estimated from GISAXS as > 100 nm which matches perfectly both results obtained by TEM and from MC simulations (Fig. 4b).
Influence of energetic disorder. With the MC machinery in hand, we studied the influence of energetic disorder on the exciton losses in large PC 71 BM domains, by performing simulations of exciton harvesting from 1-100 nm domains (Fig. 5a, red line) with and without disorder. If the phase intermixing is fine (< 15 nm, i.e. similar to the mixed phase), almost 100% of excitons are harvested in both cases, i.e. the disorder does not play any crucial role. The reason is two-fold. First, with such small domain sizes, the significant fraction of excitons is generated in the close proximity to the interface and dissociate into charges immediately with 100% efficiency. Second, even those excitons that are generated deeper in the PC 71 BM domains, reach the interface faster (in ~400 ps for 15 nm domain size, Fig. 5b) than the exciton lifetime of 650 ps.
In contrast, the exciton harvesting efficiency rapidly decreases in the large domains (> 50 nm), because the exciton diffusion time needed to reach the interface becomes comparable with the exciton lifetime. This is a direct consequence of ten-fold decrease of the exciton diffusion coefficient within 100 ps (Fig. 5b). The diffusion coefficient at long times (> 100 ps) changes only insignificantly and amounts to D ~ 1.5•10 −4 cm 2 /s, which is consistent with the earlier report 30 . This value can be safely used to estimate the exciton diffusion length since the exciton lifetime (650 ps) is much larger than the time needed for diffusion coefficient equilibration (~100 ps). However,  the exciton dynamics at the early times (< 10 ps) are determined by the highly non-equilibrium D (Fig. 5b) which explains fast exciton harvesting from the mixed phase (< 10 ps for 3 nm domain size).
In materials with negligible energy disorder, where the diffusion coefficient does not depend on time, the excitons are harvested extremely efficiently (> 75%) even from the PC 71 BM domains as large as 100 nm (Fig. 5a, blue line). This fact explains high efficiency of vacuum-deposited TPTPA/C 70 solar cells with > 95% content of C 70 with extremely low disorder of 5 meV 25,35 . Interestingly, in the disordered medium exciton harvesting is limited not only by dynamical decrease of D, but also by the presence of low-energy trap sites. As the result, the exciton is trapped at a low-energy site for a long time, which significantly decreases the diffusion length (Supplementary Section 11). Thus, even though the excitons can be effectively harvested in a BHJ based on disordered materials with fine phase intermixing, decreasing of the energy disorder seems to be more favorable for the blend optimization as in this case larger domain sizes lead to smaller interface area and, therefore, decreased non-geminate charge recombination.

Discussion
The exciton harvesting dynamics from the PC 71 BM phase have been successfully obtained by a PIA technique and modeled by the Monte-Carlo simulations to yield valuable information on the BHJ morphology. The BHJ blends studied herein contain mixed-phase PC 71 BM domains of the size of several nanometers (up to 15 nm) as well as the large PC 71 BM domains with sizes exceeding 50 nm. These findings are fully consistent with the paradigm of a hierarchical BHJ morphology 56,63,64 and were independently confirmed by GISAXS/GIWAXS, AFM, TEM/SEM and time-resolved PL measurements. Note that due to a number of fundamental limitations, the PL technique is not capable to deliver similar quantitative information.
Significant differences of BHJ morphology in terms of formation of the mixed phase and the large (> 50 nm) PC 71 BM domains have been observed for the blends with donor polymers of RRa-P3HT, MDMO-PPV, and RRe-P3HT. The phase separation of the mixed phase varies from 2 to 15 nm in PC 71 BM:RRa/RRe-P3HT blends and is ~7 nm in PC 71 BM:MDMO-PPV blends. RRa-P3HT based blends demonstrate fine intermixing without large PC 71 BM domains within the whole range of PC 71 BM loads investigated. In contrast, the MDMO-PPV and RRe-P3HT based blends exhibit the formation of large PC 71 BM domains. Observed disruption of the RRe-P3HT nanocrystals at the PC 71 BM load from 60% to 70%, verified by GIWAXS, underlines the high sensitivity of the technique used. We have also demonstrated that the exciton losses in the large PC 71 BM domains are related to a high energetic disorder of ~70 meV and in particular to the low-energy trap sites. Decreasing the energetic disorder (e.g. by applying vacuum deposition techniques) 35 dramatically improves the harvesting efficiency from the fullerene domains. This suggests that increasing the material order is a winning strategy in the exciton harvesting optimization.
The main simplification in the MC simulations is an assumption of two types of the spherically shaped domains. Although realistic morphology is much more complex 30,52 , this simple model captures the essential aspects of the PC 71 BM morphology, with two different kinds of domains being among them. Modern computational methods of predicting more realistic BHJ patterns 9,10 could readily incorporate the MC approach used herein. Next, the domain size of the mixed phase can be slightly underestimated due to the exciton delocalization among 4-5 PC 71 BM molecules 34,35 . Additionally, the possibility of the long-range hole transfer from next to the outer layer of PC 71 BM domains 65,66 cannot be ruled out. The observed increase of hole transfer time with increasing of PC 71 BM content and therefore domain size (Supplementary Section 12) is in line with this supposition. Nevertheless, all latter effects do not have serious influence on the results and could be readily accounted for in the MC simulations.
Another concern is the possible dependence of the kinetic parameters (i.e. exciton lifetime, hopping time and the disorder) on the PC 71 BM domain size. We thoroughly tested stability of domain sizes retrieval for both mixed phase and large domains with respect to the kinetic parameters of the model (Supplementary Section 13) and found that vast variations of them do not result in substantial changes of the PC 71 BM domain sizes. This is attributed to extremely fast extraction of the excitons from the mixed phase. Therefore, the obtained domain sizes for the mixed phase are reliable even if the kinetic parameters are different from the bulk PC 71 BM. On the other hand, the large PC 71 BM domains behave as the bulk material so that the parameters derived from the PL measurements can be safely used.
The charge generation after excitation of PC 71 BM is especially important for modern solar cells involving narrow bandgap polymers, where high PC 71 BM loadings are used and the fullerene becomes the main absorber in the green-blue region of the spectrum. In this work, we used the polymers with relatively wide bandgap to selectively excite PC 71 BM and therefore to simplify the analysis to demonstrate the proof of concept. In the case of modern low-bandgap donors, selective excitation of PC 71 BM is hardly achievable even in the blue where PC 71 BM absorption increases, and both donor and PC 71 BM PIA responses have to be considered. As the excitons from donor phase dissociate at a 100-fs timescale 40,42,46,50,67,68 , i.e. significantly faster than any diffusion-delayed exciton dissociation, the donor PIA response can be considered as step-like function (for an example of such retrieval, see Supplementary Section 14). In addition, transient anisotropy might be used as an extra contrast parameter to distinguish between the donor and acceptor PIA responses 26,47 . Overall, the proposed method constitutes a first step towards PIA spectroscopy as a tool that provides a valuable feedback on optimization of BHJ morphology and can be expanded to modern donor materials such as more efficient polymers 1,42,69 and small organic molecules 39,70,71 .
Blends of MDMO-PPV and both P3HTs were prepared with PC 71 BM weight content ranging from 10% to 90%. The preparation procedure was the following: the polymer and PC 71 BM were dissolved separately with concentrations of 3 g/L for MDMO-PPV and 10 g/L for RRa/RRe-P3HT in 1,2-Dichlorobenzene (ODCB) and stirred overnight on the hot plate with temperature of 60°. The solution of PC 71 BM was filtered using polytetrafluoroethylene (PTFE) filter with pore size of 0.2 μ m. The solutions of polymer and fullerene were mixed with the appropriate volumes to obtain 10-90% PC 71 BM content in the solution. The final solutions were drop cast by equal volumes of 0.2 ml on the glass microscope cover slides with the thickness of 150 μ m, and were allowed to dry for several hours making the solvent-assisted annealing 73,74 . During all measurements, the samples were kept under the nitrogen atmosphere to prevent their degradation; none was observed. Linear absorption was measured using standard Perkin Elmer Lambda 900 spectrometer. Film thicknesses were measured using Dektak profilometer. PIA measurements. Time-resolved photoinducedabsorption (PIA) spectroscopy was performed in a home-built Ti:Sapphire-based setup. The output of a 1 kHz amplifier was split to pump a noncollinear optical parametric amplifier (NOPA) 75 and a 3-stages IR OPA 76 . NOPA was producing visible 30 fs, 40 μ J pulses with the wavelength tunable in the range of 500-700 nm. The excitation wavelength was selected for the best absorption ratio between fullerene and polymer: for PC 71 BM:RRa/RRe-P3HT and PC 71 BM:MDMO-PPV as 680 nm and 630 nm, respectively. For the MDMO-PPV blends, the wavelength was set slightly off the maximal contrast point (630 nm vs. 650-660 nm, Supplementary Figure 1) in the region of higher PC 71 BM absorption to increase the signal-to-noise ratio. The IR OPA generated ~80 fs probe pulses at 3.3 μ m wavelength suitable for probing the low-energy polaron absorption (see Supplementary Section 2). The visible pump was focused into a factor of 2 wider spot than the IR probe to minimize the spatial inhomogeneity of the pump.
The PIA response was calculated as the relative transmission change ΔT/T, where T and Δ T stand for transmission and transmission change with and without the excitation pulse, respectively. Pump flux was carefully attenuated with gradient neutral density filter for the PIA response to be in the linear regime (75 μ J/cm 2 for the P3HT blends and 120 μ J/cm 2 for the MDMO-PPV blends). Note that due to the extremely low absorption of the samples at the excitation wavelengths, the absorbed photon density was below 10 −3 photons/nm 3 (i.e. ~1 photon per 10 nm of length) which minimizes bi-exciton annihilation to nihil (see Supplementary Section 15).
The polarization of the probe beam was rotated by 45° by the half-wave plate with respect to the pump. The beamsplitter was placed after the sample to detect parallel and perpendicular components of IR probe polarization with respect to the pump polarization with two wire-grid polarizers (1:100 extinction) and indium antimonide (InSb) photodiodes. PIA signals with parallel and perpendicular polarizations were used to recalculate isotropic component using the following relation 77 : where the indices || and ⊥ denote parallel and perpendicular components, and t is a time delay. The third InSb detector was used as a reference for IR pulses to enhance the signal-to-noise ratio of the PIA signal.
Scientific RepoRts | 6:36236 | DOI: 10.1038/srep36236 The precise pump-probe time-overlap position (zero delay) was carefully checked before and after each scan (every 30 minutes) by measuring the reference sample, a blend of poly[2-methoxy-5-(2-ethylhexyloxy)-1,4-phenylenevinylene] (MEH-PPV) mixed with 2,4,7-trinitrofluorenone (TNF) by weight ratio of 1:0.3. This blend forms a ground-state charge transfer complex with the apparatus-limited PIA response [78][79][80] . The materials were dissolved in chlorobenzene 2 g/L separately and mixed together. The final solution was drop-cast from chlorobenzene solution of2 g/L on the same substrate as samples and allowed to dry. The root-mean-square (rms) drift of the reference zero delay was ~5 fs during 15-hour measuring sessions.
Monte-Carlo simulations. In the MC simulations, the mixed phase was modelled as spherical PC 71 BM domains surrounded by the polymer (Fig. 1). The coexistence of the mixed phase and large PC 71 BM domains 81 was taken into account by including two types of domains with different sizes. Exciton diffusion was simulated as random hopping between discreet PC 71 BM molecules in the cubic grid cells. The boundaries of the domains were determined as spheres of variable diameters d m (for the mixed phase)and d c (for the large PC 71 Energetic disorder of the potential energy landscape of PC 71 BM was taken into account by a Gaussian disorder model 82 . Energies within the Gaussian distribution, with standard deviation σ, were randomly assigned to the PC 71 BM molecules. Initially, an exciton with finite effective lifetime T 1 is randomly placed in one of the two domains with probability f which reflects the volume ratio of the mixed phase to the large domains. At each step, the exciton hops into a random direction by one grid point with hopping time τand hopping probability p ij which depends on the energies of the starting E i and target E j grid points: where kT is the Boltzmann factor. Finally, exciton dissociation into charges at the surface of PC 71 BM domain occurs with a finite HT time τ ht 26 , after which the resulting hole begins to contribute to the PIA signal. The weak contribution of PC 71 BM excitons was taken into account by assigning it the relative cross-section α with respect to the hole response (see Supplementary Section 3). The total PIA signal was convoluted with a Gaussian apparatus function of 70-100 fs width.
The exciton lifetime T 1 , hopping time τ and energy disorder parameter σ were obtained independently from the photoluminescence data of PC 71 BM films with TPTPA quenchers as T 1 = 650 ps, τ = 0.3 ps, and σ = 70 meV (see Supplementary Sections 16, 17). The fact that a single set of kinetic parameters is needed to describe as broad range of quencher concentrations as 0.0125-50%, signifies similar exciton diffusion in PC 71 BM domains of different sizes. Therefore, the remaining fit parameters for each sample are the PC 71 BM domain sizes d m and d c , the hole transfer time τ ht and the volume fraction of the large domains f. Each of the four fit parameters is responsible for the particular feature of the PIA transients which makes them independent in the fitting procedure. The early-time dynamics (< 0.5 ps) are mainly driven by the HT process characterized by the hole transfer time τ ht (see Supplementary Section 11). The intermediate time window (1-10 ps) accounts for exciton dissociation from the mixed phase and is determined by the size of small domains d m . Size of the large domains d c is responsible for the later dynamics (> 10 ps), while the volume fraction f determines the PIA transient amplitude. To collect the necessary statistics, each simulation was run 1000 times for each sample (3000 times for MDMO-PPV:PC 71 BM blend with 60% PC 71 BM content). The dependence of the three-dimensional exciton diffusion coefficient on time was obtained from the exciton displacement as: 2 where L(t) is the exciton displacement and < > denotes averaging over the whole exciton ensemble.

TEM measurements.
For the TEM measurements, free-standing thin films were prepared in a clean room using the procedures described in refs [83 and 84] (see Supplementary Section 10 for details). The 400 mesh copper grids were used to pick up the freestanding films from water and put into Philips CM120 electron microscope operating at 120 keV. Just before performing the TEM measurement, all PC 71 BM:RRe-P3HT films were stained with iodine vapors for several minutes to improve the contrast between PC 71 BM and P3HT 15 .
GISAXS and GIWAXS measurements. GISAXS and GIWAXS measurements were performed on the custom made MINA X-ray instrument built on a Cu rotation anode (λ = 1.5413 Å). GISAXS measurements were performed using a sample-to-detector distance of 3 m. 2D patterns have been collected using a Vantec2000 detector (2048 × 2048 pixels array with pixel size 68 × 68 microns). Samples were aligned using Huber motors at an incident angle of α i = 0.3°. The in-plane GISAXS intensity cuts (along the 2θ f or the q y direction) were calculated at the Yoneda peak height (α f = 0.18°). The direct beam center position on the detector and the sample-to-detector distance were calibrated using the diffraction rings from a standard Silver Behenate powder. As the incident angle was kept fixed during the measurements, transmission functions reduce to constant values and the in-plane intensity measured directly the scattering factor of the objects inside the film. GIWAXS measurements were performed using a sample-to-detector distance of 130 cm. 2D patterns have been collected using a Vantec500 detector (1024 × 1024 pixels array with pixel size 136 × 136 microns). The direct Scientific RepoRts | 6:36236 | DOI: 10.1038/srep36236 beam center position on the detector and the sample-to-detector distance were calibrated using the diffraction rings from a standard Silver Behenate powder. Integrated intensities have been obtained by radially averaging of the intensity with respect to the beam center using the GIXGUI software 85 .