Spectrum Imaging


High-resolution mapping in the SEM by low-voltage EDS spectrum imaging and multivariate statistical analysis (MSA)


SHaRE Facility research by I.M. Anderson (ORNL)













Traditionally, X-ray microanalysis and elemental mapping of bulk specimens in the scanning electron microscope is performed with accelerating voltages of more than 20 kV. Consequently, the resultant spatial resolution, which is determined by the electron-specimen interaction volume, is typically 1 µm. Also, elemental maps are traditionally generated from pre-selected “regions-of-interest” of the X-ray spectrum (i.e. energy windows at characteristic peak positions). Three major developments now allow a breakthrough in performance. With the use of low voltages (<=5 kV), spatial resolution is improved to 200 nm or better; with the use of spectrum imaging (a full spectrum recorded for every pixel in the image) comprehensive data for all elements present (not just those thought to be there) is obtained; and with the use of multivariate statistical analysis (MSA), the Bremsstrahlung continuum in the spectrum is properly accounted for and the statistical significance of features is correctly assessed without any prior assumptions.






Data were acquired from a cross-sectioned computer chip manufactured by a major semiconductor company with the SHaRE User Facility Philips XL30FEG SEM operating at 4 kV. Spectrum imaging was accomplished with a newly installed EMiSPEC Vision system; acquisition of a 200 x 150 pixel spectrum image required ~2 h and generated a data file of ~30 Mbyte. Software developed at ORNL was used to perform the MSA analysis, which identified 9 spectral variations above the noise level governed by Poisson statistics. Images corresponding to spectral components 0-5 are shown in the figure. Component image 0 (a) shows the total X-ray intensity acquired per pixel; it is the image that would be produced by a non-dispersive X-ray detector. The intensities of the remaining component images represent contrast from a mixture of microstructural features, but the brightest regions of these images represent (b) Al lines, (c) SiO2 dielectric, (d) W plugs, (e) TiN coating and (f) CoSix. The Si substrate (bottom) is dark in all images. The other components identified by MSA correspond to a contamination spot, which is visible in (e) near the top of the middle W plug, and to subtle first-derivative-type spectral effects. Excellent contrast between the Si- and W-rich regions of the specimen is achieved in spite of the strong overlap between Si-Ka and W-Ma (~34 eV separation). Thin TiN layers around the W plugs are also clearly revealed. The resolution is a remarkable 160 nm (8 pixels) as determined by the 90% amplitude variation for the W-Si interface or by the FWHM of the CoSix layer. These mapping methods are currently also being applied to an alumina-CuAg(Ti) braze (SHaRE collaboration with Sandia National Laboratories) and Si3N4 composites. SHaRE is unique among the user facilities in offering these capabilities and many future applications are expected.






 Oak Ridge National Laboratory