Automated background subtraction for electron energy-loss spectroscopy and application to spectral imaging of semiconductor layers
Published in 2015 19th Microscopy of Semiconducting Materials, Cambridge, UK, 2015
Recommended citation: V. C. Angadi, T. Walther and C. Abhayaratne (2015), "Automated background subtraction for electron energy-loss spectroscopy and application to spectral imaging of semiconductor layers", In 2015 19th Microscopy of Semiconducting Materials, Cambridge, UK.
A novel approach for automated background subtraction of core-loss edgeis proposed. In conventional background subtraction methods, prior knowledge of the pre-edge region is inevitable. We exploit the fact that the core-loss edgeis superimposed on an almost exponentially decaying background. Principle component analysis (PCA) is used to detect clusters of positive slope angle between the spectral points. A moving average filter is adopted to minimise thefalse positive detection of core-loss edges due to shot noise.
The clusters of points with positive slope angle clearly indicate the onset of core-loss edges. We choose the threshold for the PCA as 00. The cluster of points is detected by sliding a window across the energy loss axis. If two thirds of the size of the window contain positive slopeangle then the window contains a cluster. If a cluster isn’t detected then the window slides to next position and the previous step is repeated throughout the spectrum.
Background subtraction by a curve fit through all pre-edge regions is used. The background in the spectrum from zero-loss peak to first core-loss edge is modelled using simple interpolation algorithms like spline. This is to reduce large statistical errors after background subtraction. Small pre-edges regions are selected for all the detected core-loss edges. An inverse power-law is used to fit these pre-edge regions separately, beginning at the higher energy loss edge and progressing to smaller energies. The background fitting of these points is modelled individually. The modelled background is an ensemble of spline fit and an inverse power-law fit. If the background is subtracted from the spectrum we get difference spectra for each individual core-loss edge.
Recommended citation: V. C. Angadi, T. Walther and C. Abhayaratne (2015), "Automated background subtraction for electron energy-loss spectroscopy and application to spectral imaging of semiconductor layers", In 2015 19th Microscopy of Semiconducting Materialss, Cambridge, UK.