Measurement of nanometre-scale gate oxide thicknesses by energy-dispersive x-ray spectroscopy in a scanning electron microscope combined with monte carlo simulations

“A procedure based on energy-dispersive X-ray spectroscopy in a scanning electron microscope (SEM-EDXS) is proposed to measure ultra-thin oxide layer thicknesses to atomic scale precision in top-down instead of cross-sectional geometry. The approach is based on modelling the variation of the electron beam penetration depth and hence the depth of X-ray generation in the sample as a function of the acceleration voltage. This has been tested for the simple case of silica on silicon (SiO2/Si) which can serve as a model system to study gate oxides in metal-on-semiconductor field-effect transistors (MOS-FETs). Two possible implementations exist both of which rely on pairs of measurements to be made: in method A, the wafer piece of interest and a reference sample (here: ultra-clean fused quartz glass for calibration of the effective k-factors of X-ray lines from elements O and Si) are analysed at the same acceleration voltage. In method B, two measurements of the apparent O/Si ratio of the same wafer sample need to be made at different acceleration voltages and from their comparison to simulations the SiO2 layer thickness of the sample can be inferred. The precision attainable is ultimately shown to be limited by surface contamination during the experiments, as very thin carbonaceous surface layers can alter the results at very low acceleration voltages, while the sensitivity to ultra-thin surface oxides is much reduced at higher acceleration voltages. The optimal operation voltage is estimated to lie in the range of 3–15 kV. Method A has been experimentally verified to work well for test structures of thin oxides on Si-Ge/Si.”

3. Energy-Dispersive X-ray Spectroscopy Measurements in SEM Using Method A

In order to enable sets of coherent experiments under exactly the same conditions, four specimens were mounted onto a common stub for investigation in a Hitachi (Hitachi High Technologies, Tokyo, Japan) tabletop SEM 3030 plus at U = 15 kV, a TOA of 22° and magnification of 40×. The microscope is equipped for EDXS with a 30 mm2 Bruker (Bruker Nano Analytics, Berlin, Germany) XFlash 430 silicon drift detector that is nominally 450 µm thick, has a 0.029 mm dead layer, an AP3.3 polymer entrance window and provides 126 eV full width at half maximum resolution at 5.9 keV and a sampling of 10 eV/channel.
The surfaces of all specimens were subsequently cleaned in 99.8% pure ethanol and acetone. No conductive coatings were applied to any sample as none of them showed any signs of charging. This can be seen in Figure 7c where the Bremsstrahlung background of all spectra clearly extends out towards the Duane–Hunt limit at 15 keV in much the same way. The specimens in Figure 6b,c are, clockwise and starting at the top right: a SiGe specimen of known structure that had been previously studied by cross-sectional TEM [28] and the (Si,Ge)O2 layer thickness on top of which is known to be now 6.5 nm thick from Figure 8, a 2 mm thick piece of fused quartz glass of type Spectrosil B that serves as an ultra-pure calibration specimen of composition SiO2 (trace elements < 10 ppm), a Si wafer piece that had been polished an hour before inspection using 4000 grit silicon carbide paper so the newly formed native oxide on top should be only 1 h old and rather thin, and a piece of Si wafer left over from a previous project that ended in 2011, providing a native oxide on top that was 10 years old.
Figure 7. EDX spectra of four specimens at 15 kV: epitaxial Si with 10-year-old native oxide, mechanically polished Si(001) wafer, surface of chemical vapour deposited SiGe 22C86 specimen (cf Figure 8) and fused quartz glass specimen Spectrosil B ©. (a) spectral range 0–2 keV on logarithmic scale; (b) same spectra as before on linear scale over 0–2.4 keV, with additional data for C and Pt blanks; (c) same spectra range on magnified linear scale over 0–15 keV, demonstrating Bremsstrahlung extending out to 15 keV limit even for the silica sample (purple), which showed no charging.
Figure 8. Bright field transmission electron microscopy (TEM) image of sample 22C86, showing from top in cross-section: glue line/6.5 ± 0.3 nm SiO2/66.5 nm Si0.83Ge0.17/Si(001).
X-ray spectra were collected at around 10,000 counts per second for 300 s (for Spectrosil B ©: 400 s) while keeping the electron beam scanning regions ≈ 1 mm in diameter in mapping mode to avoid strong contamination often observed otherwise for stationary electron probes or at high magnifications. The resulting spectra are shown in Figure 7. In Figure 7a we concentrate on the energy range up to 2 keV (above that only the SiGe specimen gave a weak Ge K X-ray signal which is irrelevant for our method here). The vertical axis shows the counts on a logarithmic scale so that the rather faint signals of C K, O K and Ge L can be seen before the much larger Si K-line at 1.74 keV. For integration, windows of 475–575 eV are used for the O K-line centred at 525 eV and 1635–1935 eV to include the Si Kα-line at 1740 eV and Kβ at 1836 eV. These windows are marked as bars in light grey in Figure 7a,b. The background is fit linearly over 0.3–1.6 keV around but excluding the O K-line itself and 1.6–2.0 keV around but excluding the Si K-line. The fits are shown for the SiGe and Si specimens as black curves in Figure 7; their apparent curvature in Figure 7a is only due to the logarithmic scale of the plot. Then net counts have been extracted by background interpolation and subtraction. These are reported in Table 1, along with additional X-ray measurements conducted for 99.9% pure elemental standards of graphite and platinum that serve as blank controls as they should be free of any surface oxides when prepared in the cleanest possible way (cleaning as before, plus a subsequent anneal at 250 °C for two hours to desorb any solvent residues).

Leave a Comment