Noise reduction procedures applied to XPS imaging of depth distribution of atoms on the nanoscale

We have studied the capability of a new method for 3D XPS imaging and have focused on the influence of noise in the spectra. To this end, we have studied a patterned structure made by thermally oxidising a silicon wafer. We have studied the O 1s, Si 2p and C 1s peaks which have rather low photoionization cross sections. In addition we have not used high spectrometer pass energy. Therefore, the signal-to-noise level for these spectra was very low. We have investigated the extent to which different noise reduction procedures can improve the quantitative images obtained. The original data has been processed using four different methods: (1) smoothing using a quadratic, 7-point Savitzky–Golay filtering followed by averaging the spectrum for each pixel with spectra from nearest neighbours (2) principal component analysis (PCA) (3) PCA followed by smoothing and (4) PCA followed first by smoothing and then by averaging. We have shown that for noisy spectra, PCA significantly improves the images of both the amount of substance (AOS) in the outermost few nanometers and also the XPS-images of the different in-depth distributions of oxygen, carbon and silicon atoms. The images of the depth profiles for the different elements in the sample studied are found to be consistent. This result is important because in imaging, data acquisition time is a limiting factor, which can be reduced by effective noise reduction procedures.