Chemical Composition of Semiconductor Interconnects

Standard energy dispersive X-ray spectroscopy (EDS or EDX) using detector areas of 30 mm2 on standard scanning transmission electron microscopes (STEM) can deliver element mappings with nm resolution within a few minutes. The condition is, that the detector head is small enough (in slim-line design) to get as close to the specimen for (high solid angle) and as high above the specimen (for high take-off angle) as possible. The latter helps to avoid shadowing and absorption effects.

Standard STEM retrofitted with smaller 30 mmactive area EDS with light element window, achieving 0.09 sr collection angle at 22° take-off angle, was used to analyze semiconductor interconnects (Fig. 1). The element distribution was mapped. The EDS data were processed quantitatively using the Cliff-Lorimer method. The calculation of theoretical Cliff-Lorimer factors in the ESPRIT software is based on the following information: 

  • an extensive and constantly updated atomic data base with values for radiation cross sections and fluorescence yields 
  • the geometric relation of detector and specimen 
  • and information on detector quantum efficiency 

Within one series of specimens investigated under the same conditions, the Cliff-Lorimer method, using theoretically calculated factors, can be accurate within a few atomic percent relative to a chosen reference sample out of the specimen series. EDS data clearly reveal the tantalum and titanium of the Ta and TiN interconnect lining, as well as the copper and tungsten filling (Fig. 2). The titanium signal can be separated from the nitrogen signal. Si, Ta and W can be deconvolved and correctly assigned (Fig. 3).

Fig. 1: High angle annular darkfield image of an interconnect structure. Specimen courtesy: Synergie4.
Fig. 2: Data extracted from a 355 pixel x 678 pixel element mapping, acquisition time: 15 min. Left: Net count representation of a few relevant elements. Middle: Quantitative analysis of Ta using 4x4 pixel binning. Right: Ti distribution using 8x8 pixel binning.
Fig. 3: Deconvolution of Si, Ta and W EDS element lines in ESPRIT.