Over the last years the efficiency and stability of organic solar cells (OSCs) increased considerably due to the development of new acceptor materials, particularly non-fullerene acceptors (NFAs) [1].

The photovoltaic properties of an OSC depend crucially on the morphology of the bulk heterojunction. Three morphological parameters appear to have the most impact: (i) the crystallinity of the different materials and relative crystal orientation, (ii) the domain sizes and composition of material phases, and (iii) the composition and extent of the donor-acceptor interface. Importantly, several previous reports showed that energetic properties leading to large driving forces for charge separation are not necessarily observed for NFA blends, which reach high efficiencies [2, 3]. It is therefore of great importance to visualize the nanoscale morphology in detail, so that the mechanism for optimized charge separation and hence, charge generation capacity in OSCs may be deduced.

Via correlation of high-resolution information from conventional transmission electron microscopy (TEM) with phase distributions from analytical TEM, we demonstrate that all three morphological parameters can be visualized simultaneously. Figure 1a shows what appears to be one crystallite inside a 30 nm thin PBDB-T:ITIC photoactive layer recorded by conventional TEM. Crystallites have been visualized before in various fullerene-based blends [4]. However, due to beam damage it has so far not been possible to determine material phases in seemingly single crystallites e.g. by 3D high-resolution electron tomography. Here we discuss, how low-dose imaging and analytical TEM can be combined to visualize crystallite composition in 2D. Our results reveal that direct crystal-crystal contacts of donor and acceptor materials exist in NFA based blends.

Low-energy-loss analytical TEM has been shown to allow segmentation of material distributions for PCBM blends [5, 6]. As seen in Figure 1b, only minor spectral differences, which might be used for discriminating PBDB-T and ITIC signals, are found in the optical excitation spectra. However, in the plasmon region, there are spectral differences yielding images with inverse material contrast (cf. Figure 1c). These images were acquired at the same crystalline sample position at the energy regions marked in the diagram in Figure 1b. Applying machine learning to whole series of inelastic images we determine material distribution maps. From such maps we can directly measure the domain sizes of pure material phases and obtain information about the donor-acceptor interface, here marked in yellow. 

The overlay of the conventional TEM image and the material distribution map in Figure 1d shows a direct interface of connected PBDB-T and ITIC crystallites instead of a uniform crystallite of one material, with only marginal interfacial intermixing. This finding is confirmed by a precise measurement of the lamellar spacings from selected parts of the conventional TEM image (cf. Figure 1e).

As shown, correlating high-resolution and analytical TEM imaging opens new possibilities to visualize so far unexpected structural details of OSC blends. At present we focus on NFA-based morphologies, relating optimized function to the occurrence of the newly described donor-acceptor crystal contacts.