Fining experiment

The aim of this experiment is to see what changes in the spectrum, if the thickness of a wafer is reduced and how the concentration changes, if the protocols are applied to these spectra.

5.2 Thinner samples

We start with a set of wafers, for which the oxygen protocol is applicable. Then these wafers are thinned by means of a potash solution. Assuming, that the bulk oxygen concentration does not change, possible changes are due to the methodology.

Our starting material were three Czochralski-grown n-type (phosphorous) wafers with re-sistivity between 3.8 Ω cm and 6.0 Ω cm and initial thicknesses from 511µm to 516µm. As reference material twon-type (phosphorous) floating-zone wafers with resistivity 3.8 Ω cm and 4.4 Ω cm and thicknesses 507µm and 505µm are used. Here it can already be noticed, that the 0.5 % tolerance for the reference specimen thickness is hard to achieve. In absolute values this means a coincidence of thickness within 2.5µm, which is not even fulfilled by wafers from the same box. This might be a drawback for our experiment. Another handicap is, that the wafers on one hand are only polished on one side and on the other hand, the potash solution acts as a polisher - changing the surface from step to step. This has influence on the spectra and could therefore also change the results.

The thinning was carried out by means of a potash solution. First, 500 ml deionized water are heated in a glass beaker. When 55^◦C were reached, 200 g KOH in pellets are slowly and under stirring added. Then 70 ml of 2-propanol were added to KOH in small portions. Caution is advised, because the potash solution as well as the added propanol may explode if too much is added at one time. After that, the temperature was maintained at 80^◦C. Then the wafers can be immersed. As this treatment showed many inhomogeneities on raw wafers, we decided to carry out a RCA-2 cleaning preceding every fining step. This cleaning consists of a 10 minute immersion in a mixture of 500 ml H₂O, 100 ml HCl and 100 ml H₂O₂, heated up to 80^◦C.

In the first thinning step, the samples were immersed for five minutes in the potash solution.

The result was rather inhomogeneous, so we decided to introduce a RCA-2 cleaning step before each following thinning step. For step two, a new potash solution was prepared, which was used until fining step five. The immersion in step two and three lasted ten minutes, in step four 16 minutes with 1⁰40 s extra time for C-A210t and C-A310t to come to the same thickness as the other wafers. The immersion in step five lasted 20 minutes. After the immersion, the wafers were rinsed three times in deionized water and dried by pressurized nitrogen. Then two spectra were registered with resolutions 4 cm⁻¹ and 8 cm⁻¹ and 32 scans. Although this is half of the number of scans required by the protocol, the 100 % line in the range of 900 to 1300 cm⁻¹ is met even within 0.3 % and the noise is below 0.1 % for resolution 4 cm⁻¹ as well as for 8 cm⁻¹. Hence, this number of scans is sufficient to apply the protocol.

The thicknesses after each step are shown in Figure 5.2b. After the first two steps, 44µm have been etched off, in the third 31µm, in the forth 40µm and in the last 33µm. These values are the mean of the values of all samples. It can be noted, that the etch speed decreases with the step number, which could be expected, as reactants vanish. Another observation is, that the polished side seems not to be affected by the etching. The polished surface remains polished, but the name, graved into the polished side by a diamant point, is deepened. Maybe this surface is so perfect, that the base does not get any point of attack.

In the spectra, Figure 5.2b, the influence of the surface change is clearly visible. As the absorption coefficient of crystalline silicon is very low above 1000 cm⁻¹and our sample is mod-erately doped, the absorption in this region is mainly determined by reflection and scattering at the surface. We attribute the decrease in absorbance for every fining and polishing step

Cz-A210t Cz-A301t Cz-A310t F-A115t F-A205l

4000 wavenumber in cm⁻¹

0.0 Figure 5.2: Thicknesses and spectra obtained after the different etching steps

to the decrease of number and size of scattering centers on the surface. Fortunately, these changes seem to lead only to a shift of the absolute values, but not to a deformation of the peak shape, at least from step two on.

From these spectra, the oxygen concentration can be calculated according to protocol SEMI-MF1188-1107. Since we have two reference specimen, two oxygen concentrations can be cal-culated for each test specimen. In Figure 5.3a the average of these concentrations is plotted.

The errorbars indicate the height of these two concentrations. It can be seen, that the effect of the different reference wafers is negligible. Further it can be seen, that the values after the first two polishing steps are quite stable. The deviation for the concentrations from step two on are 6 % for C-A310t, 4 % for C-A301t and 4 % for C-A210t. These values are above the reported repeatability for the protocol, probably a result of the skipped hydrofluoric acid etch or due to the surface change due to polishing. Nevertheless, these results are quite acceptable. However, for C-A301t we note a great difference between as grown and polished values. There is little difference between the two values obtained for the two reference crystals, so this deviation must be due to the spectrum of C-A301t. Compared to the spectrum of C-A210t, which has the same thickness, the absolute values in the neighbourhood of the peak are similar, the baseline is similar, only the absorbance peak is higher. Hence, we conclude that there must be really some more oxygen in the sample. Maybe on the surface. The last thing to check, are the full width at half maximum (FWHM) of the peaks. In our case, they were rarely below 32 cm⁻¹, the median was 33 cm⁻¹. Following the protocol strictly, only 20 % of our measurements would be accepted. However, as the distance between two data points is 1 cm⁻¹, at least a tolerance of 2 cm⁻¹ must be demanded. If this is taken into account, all of our measurements can be considered to be properly done.

The other part was to compare the measurements with resolution 8 cm⁻¹ and 4 cm⁻¹. The values obtained by the calculations of the oxygen protocol are shown in Figure 5.3b. For the raw samples, no measurements have been carried out with resolution 8 cm⁻¹. One can see, that the measurements with resolution 8 cm⁻¹ lead to systematically lower oxygen content.

The deviation is below 2 % for steps 2 to 4 and around 6 % for step 5.

As the spectra also contain information about the carbon concentration, we can try to

5.2 Thinner samples

calculate the carbon concentration according to protocol SEMI MF1391-0704. Of course we must not rely on the absolute values, because thickness and resolution do not conform to the protocol requirements, but we can check, whether there is a stable relationship, which only needs a calibration. For the calculations, we used the option to tune the normalization factor visually, to get a well isolated band. The concentrations are shown in Figure5.3c. First of all, we note, that the variation between the different steps as well as between the two reference slices is larger than in the oxygen case. The values for the as grown wafers in general do not coincide with the values for the polished wafers. The results for the carbon concentration and the oxygen concentration at resolution 4 cm⁻¹ have in common, that the value after step 5 increases with respect to the preceding step, but for the carbon content, this effect is more pronounced. The reason can be seen in Figure5.3eand5.3f. For this thickness and resolution, some oscillations in the spectrum appear. Their amplitude is added to the existing absorption peaks. In the case of oxygen, the amplitude is smaller than the peak height of the oxygen absorption band, which leads to a small increase in oxygen value. In the case of carbon, the amplitude is of the same height as the carbon absorption peak. Therefore, the nominal carbon concentration is doubled. For resolution 8 cm⁻¹, the oscillation vanishes, so the carbon concentration is approximately maintained, as can be seen in Figure 5.3d. However, based on this data, the carbon concentration cannot be measured reliably. A problem can be, that besides the low resolution, the sample holds little carbon and is very thin, which makes the absorption peak very weak.

The oscillations, which appeared in the last etching step, were the reason, why no further etching was carried out. In the next step, the oscillations would have covered the whole spectrum and hided all information. Therefore, we investigate in the next section, what the reason for these oscillation is and how you can get rid of it.