RESUMO
To improve the performance of organic light-emitting diodes (OLEDs), it is essential to understand and control the electric potential in the organic semiconductor layers. Electron holography (EH) is a powerful technique for visualizing the potential distribution with a transmission electron microscope. However, it has a serious issue that high-energy electrons may damage the organic layers, meaning that a low-dose EH is required. Here, we used a machine learning technique, three-dimensional (3D) tensor decomposition, to denoise electron interference patterns (holograms) of bilayer OLEDs composed of N,N'-di-[(1-naphthyl)-N,N'-diphenyl]-(1,1'-biphenyl)-4,4'-diamine (α-NPD) and tris-(8-hydroxyquinoline)aluminum (Alq3), acquired under a low-dose rate of 130 e- nm-2 s-1. The effect of denoising on the phase images reconstructed from the holograms was evaluated in terms of both the phase measurement error and the peak signal-to-noise ratio. We achieved a precision equivalent to that of a conventional measurement that had an exposure time 60 times longer. The electric field within the Alq3 layer decreased as the cumulative dose increased, which indicates that the Alq3 layer was degraded by the electron irradiation. On the basis of the degradation of the electric field, we concluded that the tolerance dose without damaging the OLED sample is about 1.7 × 105 e- nm-2, which is about 0.6 times that of the conventional EH. The combination of EH and 3D tensor decomposition denoising is capable of making a time series measurement of an OLED sample without any effect from the electron irradiation.