Determination of bandgap onset by blind deconvolution of electron energy-loss spectra (EELS): 1D EELS vs 2D EEL spectrum images

Precise determination of the bandgap is important for semiconductor research. It is possible but not straightforward to determine the location of the onset of the Density of States (DOS) from low-loss EELS rather than by optical spectroscopy. A square-root function fit to the low-loss function may work for direct bandgaps but will be affected by the presence of a strong and asymmetric zero-loss peak, phonons, Cerenkov effects and possibly even surface plasmons. Also, the tail of the bulk plasmon will affect the determination of bandgap. Deconvolution methods can be applied to remove these effects. Fourier-log and Richardson-Lucy deconvolution methods are routinely used for one-dimensional (1D) spectra but they tend to enhance noise. An alternative method is 2D deconvolution of a spectrum image. In this method the deconvolution is applied to a spatially or angular resolved EELS, or spectrum image, bringing the extended zero-loss peak to a single point. By this way of deconvolution, even weak information which is hidden by the wide point sperad function of the zero-loss peak can be made visible. In this study, different ways to determine the bandgap of GaAs are compared using blind deconvolution of 1D and 2D EELS. For this, starting from the same 2D spectrum image, the effect of changing the sequence of projection (from 1D to 2D) and deconvolution is thus compared. A Gaussian model of the zero-loss peak is considered as the initial point spread in both cases. A prominent onset of the intensity at low energies is observed and either a square-root function fit or simple smooting and differentiation are used to refine the exact location of the onset. For deconvolution of the 1D vertically projected EELS the value of the band edge apparently decreases with the number of iterations, from ~ 1.8 eV to ~ 0.4 eV (mean value of 0.82 eV, with standard devation of 0.46 eV). For 2D decovolution followed by vertical projection the apparent bandgap stays consistently high, yielding E_g = 1.42 &plusmn 0.03 eV independent of the number of iterations.

Download paper here

Recommended citation: V. C. Angadi, C. Abhayaratne and T. Walther (2017), "Determination of bandgap onset by blind deconvolution of electron energy-loss spectra (EELS): 1D EELS vs 2D EEL spectrum images", In 2017 20th Microscopy of Semiconducting Materials, Oxford, UK.