Accurate Elemental Mapping of Semiconductor Devices Using EDS – Deconvolving Overlapping Peaks

Energy Dispersive X-ray Spectroscopy (EDS) is a critical tool for the electronics industry in research and development and quality control environments, where elemental mapping is used to visualize the distribution(s) of dielectric layers and electrical contacts within individual devices of integrated circuits. However, many of the materials that are used in the construction of electronic devices and their electrical interconnects have, traditionally, been thought to be difficult to map by EDS due to the inability of EDS detectors to resolve important characteristic X-ray peaks, such as Si K, Ta M and W M (energies 1.740, 1.709, and 1.774 keV respectively). The best energy resolution of commercially available EDS detectors varies by model but is typically 123–130 eV. Consequently, EDS spectra of samples containing Si, Ta, and W suffer from severe peak overlap. For instance, the EDS spectrum in Figure 1 shows that the Si K and W M peaks from a WSi2 sample cannot be separated. Traditionally, it was believed that these peak overlaps could only be resolved using Wavelength Dispersive X-ray Spectroscopy (WDS), as this technique has better energy resolution than EDS (see the WDS spectrum in Figure 1). However, WDS usually requires operation at relatively higher beam current and accelerating voltage, which causes the sample to suffer more drift and charging. Many features that need to be analyzed in the cross-section area of semiconductor devices are sub-micron in scale, and the sample is typically uncoated, making drift and charging mitigation critical and WDS impractical for this application. Nevertheless, appropriate mathematical deconvolution techniques [1-3] can be used with EDS to separate the individual elemental peaks, providing sufficient X-ray counts at a single pixel level. Figure 2 shows the Si K and W M distributions in the cross-section of a solid-state drive. For the ROI maps, a region of interest (ROI) is defined for each energy line. The width of the region is often based on the energy resolution of the detector at the given energy. The X-ray counts within each ROI are summed for each pixel and used to assign a brightness level to the pixel. Since the Si K and W M lines are close in energy, their ROI maps display similar distributions. Net maps can overcome the limitations of ROI maps by performing background subtraction and peak deconvolution. The net maps in Figure 2 clearly reveal the distinct distributions of silicon and tungsten. Pixel-level statistics are critical for the success of peak deconvolution, making it essential to collect map data with a fast amp/process time to increase throughput. As detector resolution degrades with faster amp/process times, it is crucial to use an EDS detector with good resolution stability to minimize this degradation at high throughput and obtain accurate deconvolution results. In comparison to EDS, although WDS has better energy resolution to intrinsically resolve such peak overlap, the higher operating beam current and accelerating voltage required, as well as the longer acquisition time due to sequential collection, result in severe drift and charging, making it unlikely to resolve sub-micron scale features. Therefore, an EDS detector with good resolution stability and adequate mathematical deconvolution techniques can provide an accurate representation of elemental distribution and is more suitable for this kind of application. EDS (red outline) and WDS (cyan) spectra overlay of a WSi2 sample. Si K and W M ROI (top) and net (bottom) EDS maps of the cross-section of a solid-state drive collected at 5 kV accelerating voltage.