Results are reported from a pilot study under the Consultative Committee for Amount of Substance (CCQM) to compare measurements of and resolve any relevant measurement issues in the amount of thermal oxide on (100) and (111) orientation silicon wafer substrates in the thickness range 1.5-8 nm. As a result of the invitation to participate in this activity, 45 sets of measurements have been made in different laboratories using 10 analytical methods: medium -energy ion scattering spectrometry (MEIS), nuclear reaction analysis (NRA), RBS, elastic backscattering spectrometry (EBS), XPS, SIMS, ellipsometry, grazing -incidence x-ray reflectometry (GIXRR), neutron reflectometry and transmission electron microscopy (TEM). The measurements are made on separate sets of 10 carefully prepared samples, all of which have been characterized by a combination of ellipsometry and XPS using carefully established reference conditions and reference parameters.The results have been assessed against the National Physical Laboratory (NPL) data and all show excellent linearity. The data sets correlate with the NPL data with average root-mean-square scatters of 0.15 nm, half being better than 0.1 nm and a few at or better than 0.05 nm. Each set of data allows a relative scaling constant and a zero thickness offset to be determined. Each method has an inherent zero thickness offset between 0 nm and 1 nm and it is these offsets, measured here for the first time, that have caused many problems in the past. There are three basic classes of offset: water and carbonaceous contamination equivalent to ∼1 nm as seen by ellipsometry; adsorbed oxygen mainly from water at an equivalent thickness of 0.5 nm as seen by MEIS, NRA, RBS and possibly GIXRR; and no offset as seen by XPS using the Si 2p peaks. Each technique has a different uncertainty for the scaling constant and consistent results have been achieved. X-ray photoelectron spectroscopy has large uncertainties for the scaling constant but a high precision and critically, if used correctly, has zero offset. Thus, a combination of XPS and the other methods allows the XPS scaling constant to be determined with low uncertainty, traceable via the other methods. The XPS laboratories returning results early were invited to test a new reference procedure. All showed very significant improvements. The reference attenuation lengths thus need scaling by 0.986 ± 0.009 (at an expansion factor of 2), deduced from the data for the other methods. Several other methods have small offsets and, to the extent that these can be shown to be constant or measurable, these methods will also show low uncertainty. Recommendations are provided for parameters for XPS, MEIS, RBS and NRA to improve their accuracy.  Crown

The crystallinity of atomic layer deposition hafnium oxide was found to be thickness dependent, with the thinnest films being amorphous and thick films being at least partially crystalline. Hafnium oxide films fabricated by metalorganic chemical vapor deposition are mostly monoclinic. Formation of hafnium silicate by admixture of 20% Si prevents crystallization. Electronic defects are reflected by an absorption feature 0.2–0.3 eV below the optical bandgap. These defects arise in polycrystalline, but not in amorphous, hafnium-based oxides.

The real and imaginary parts of the refractive index, n(ω) and k(ω), of silicon were measured as a function of photon frequency ω using Fourier transform infrared (FTIR) transmission spectral data. An accurate mechanical measurement of the wafer’s thickness, t, was required, and two FTIR spectra were used: one of high resolution (Δω=0.1to0.5cm−1) yielding a typical channel spectrum (Fabry–Perot fringes) dependent mainly on t and n(ω), and one of low resolution (Δω=4.0cm−1) yielding an absorption spectrum dependent mainly on t and k(ω). A procedure was developed to first get initial estimates for n(ω) for the high-resolution spectrum and then calculate k(ω) from the faster low-resolution spectrum with minimal measurement drift. Then both initial n and final k values were used together as starting point data for a fit to the high-resolution spectrum. A previously derived transmission formula for a convergent incident beam was used for the fit. The accuracy of n(ω) determined using this procedure is mostly dependent upon the measurement error in the sample thickness t and k(ω) is dominated by the accuracy of the absolute transmission values obtained from a sample-in and sample-out methods. Our results are compared with previously published values for n(ω) and k(ω) in the 450–4000-cm−1 spectral region. The reported uncertainty in n(ω) is ±10−4 absolute, a factor of 10 better than published values. The values of n(ω) range from 3.4400 at 4000cm−1 (λ=2.5μm) to 3.4169 at 450cm−1 (λ=22.222μm). The k(ω) values had a standard deviation of ⩽±3% and are in good agreement with previous measurements.

We have measured x-ray absorption spectra (XAS) at the oxygen K edge for hafnium oxide (HfO2) films grown by chemical vapor deposition (CVD) and atomic layer deposition (ALD), as well as for hafnium silicate (HfSiO) films grown by CVD. The XAS results are compared to x-ray diffraction (XRD) and spectroscopic ellipsometry (SE) data from the same films. Features characteristic of crystalline HfO2 are observed in the XAS spectra from all CVD-grown HfO2 films, even for a thickness of 5 nm where XRD is not sensitive. XAS and XRD spectra from the ALD-grown HfO2 films exhibit the signature of crystallinity only for films that are 20 nm or thicker. These characteristic XAS features are absent in all HfSiO films measured, which is consistent with their being amorphous. The appearance of these peaks in XAS and XRD is correlated with sub-band-gap absorption in the SE spectra, which appears to be intrinsic to crystalline HfO2 in the monoclinic phase.

No abstract