Significant advances were recently made in achieving high efficiency organic photovoltaic cells based on non-fullerene acceptors (NFA) [1, 2]. The morphology of bulk heterojunctions (BHJ) is one of the key parameters determining their efficiency. However, optimization still relies primarily on trial and error and personal experience rather than on profound knowledge of the evolving morphology. The similar atomic composition and structures of the donor and acceptor molecules necessitate the use of advanced spatially resolved spectroscopy to identify the different phases at the nanoscale [3]. This is particularly valid for polymer:NFA blends, which show considerable higher similarity in their electronic spectral fingerprint, as compared to polymer:fullerene systems. Deeper knowledge of the morphology and of developing material phases will not only improve the optimization process of the cells [4], they are also crucial for understanding fundamental physical processes at the interfaces within NFA based BHJs. This information may also explain observations, where the absence of a significant driving force for charge separation at the donor/acceptor interface is in the center of an ongoing debate [5].

We demonstrate a novel technique to visualize the surface morphology of organic photovoltaic blends at the nanoscale. We use spatially resolved electron spectroscopy in a prototype aberration corrected ultra-low voltage scanning electron microscope (ULVSEM), the DELTA (Carl Zeiss Microscopy, Germany) [6]. The main advantage of ULVSEM is the extremely low electron landing energy down to 20 eV instead of ≥1 keV in standard SEM and 30 – 300 keV in TEM. Firstly, the interaction volume between electron probe and sample at such low primary energy is drastically reduced – only the first few nanometers below the surface are imaged. Thus, signal mixing at interfaces is minimized (cf. Figure 1a). Secondly, we observe minimal beam damage for organic materials in this ultra-low voltage regime.

In contrast to previous attempts with spectroscopy in SEM [7], we do not compare spectra of the pure components to identify spectrometer settings to maximize contrast between donor and acceptor materials. Instead, our method relies on nano-resolved spectra and is therefore sensitive towards local variations in the electronic structure of the investigated sample. Unsupervised machine learning algorithms are used to reveal spectroscopic similarities within the datasets. This unbiased procedure yields three phases in a PTB7:PC70BM model system investigated at 50 eV primary energy. The phases can be assigned to a polymer rich, a fullerene rich and a mixed phase. Smaller domain structures are observed in comparison to results from analytical transmission electron microscope (ATEM) reference measurements based on the methodology proposed in previous works [8]. The procedure and resulting morphology data are illustrated in Figure 1b,c.

We expect that combined analyses of lateral and cross-sectional surfaces will shed light on the processes at inter-layer as well as donor:acceptor interfaces for an enhanced understanding of charge separation processes.