THE DETECTION OF 0.5AT% BORON IN Ni₃Al USING PARALLEL ENERGY LOSS SPECTROSCOPY

Steel Research Centre, NKK Corporation, Kawasaki 210, Japan.

*now at: Department of Materials Science and Metallurgy, Pembroke St, Cambridge, CB2 3QZ, UK.

Recent work has suggested that the improvement in ductility when boron is added to the inherently brittle Ni₃Al is caused by the segregation of boron to the grain boundaries and that this seems to be associated with the presence of disordered grain boundary "phases".^1,2,3 In order to clarify the rôle of boron at Ni₃Al boundaries at the resolution of transmission electron microscopy we have developed a method for detecting concentrations as low as 0.5at% of boron in Ni₃Al using parallel energy loss spectroscopy.

Specimens of Ni-24at%Al with and without 0.5at% B were electropolished in a 20% perchloric acid/ethanol solution and examined at 120 kV in a Philips 420T electron microscope equipped with a Gatan parallel energy loss spectrometer. Figure 1a shows part of the loss spectrum from the Ni₃Al-0.5%B matrix after background subtraction. Provided a high enough count rate can be obtained, the major problem for detecting small edges in energy loss spectra is channel to channel variations in the gain, which for figure 1a produce a noise level (2[sigma]) of ~0.3%. Such noise is caused by non-uniformities in the fibre optics transmitting light to the diodes, as well as intrinsic gain variations between individual diodes. The standard procedure for removing this noise is to divide by a previously collected "gain spectrum" obtained with the diodes uniformly illuminated and the result of this procedure is shown in figure 1b. It can be seen that the noise is only reduced by a factor of about 2, because the width and position of the illuminated region of the scintillator is not exactly the same for the gain spectrum and the Ni₃Al spectrum. First difference spectra and averaging many spectra shifted by one channel relative to each other have been used to overcome such gain variations^4,5 but edge areas cannot be obtained easily from a first difference spectrum and many spectra (eg ~60) must be acquired for the averaging method to reduce the noise significantly. Figure 1c shows the result of averaging a series of 8 Ni₃Al-0.5%B spectra shifted by from 9 to 35 channels with respect to each other. Although the noise is reduced by root(8) to about 0.1% compared to figure 1a, this is still much greater than the expected electron shot noise for these spectra at the boron edge of 2[sigma]~0.003%. However, the residual channel gain noise can be removed by the iterative averaging method shown diagrammatically in figure 2, for the case of 2 spectra shifted with respect to each other. Firstly the original spectra (figure 2a) are calibrated and aligned with respect to energy loss, then averaged over corresponding energy losses (figure 2b). An estimate of the channel gain variations can be made from the original spectra by dividing each original spectrum (figure 2a) by the average spectrum (figure 2b). These channel gain variation estimates are then averaged over corresponding channels to give a better estimate of the gain variations (figure 2c). Each original spectrum can then be corrected for channel gain variations by dividing by this new gain spectrum (figure 2c) to give less noisy spectra (figure 2d). The process is then iterated, with the newly corrected spectra (figure 2d) averaged again to form a new average spectrum (like figure 2b), each time obtaining successively better estimates of first the average spectrum, then the gain spectrum. Because the spectra extend over a finite range of channels, the low and high eV ends will be averaged over fewer spectra and thus the greatest noise reduction will be towards the middle of the energy range.

In practice only 2 or 3 iterations of this process are necessary before no significant improvement in the average spectrum occurs and the result after 3 iterations of the 8 Ni₃Al spectra is shown in figure 1d. The noise at the boron edge has been reduced by a factor of ~20 to ~0.005%, which compares favourably with the expected noise of 0.003%. Unfortunately, the boron edge is hidden beneath the extended fine structure of the Ni M and Al L edges at 68 and 74 eV respectively and it can only be made visible by applying a second difference filter, whose effect is to enhance the high frequency components of the spectrum (especially the noise!), as shown in figure 3a. The boron edge is now visible as a shoulder at 188 eV that is not present in similar spectra from Ni₃Al containing no boron (figure 3b). It should be pointed out that this shoulder has been consistently observed in 3 series of Ni₃Al-0.5%B spectra, yet never observed from Ni₃Al with no B. Given that the iterative averaging method removes almost all of the channel gain variation noise, detecting the higher concentration of boron expected at grain boundaries is "only" a matter of obtaining sufficient counts from a given probe size to reduce the electron shot noise to acceptable levels.⁶

References

1. C. T. Liu et al., Acta Metall. 33 (1985) 213.
2. J. A. Horton and M. K. Miller, Acta Metall. 35 (1987) 133
3. I. Baker and E. M. Schulson, Scripta Metall. 23 (1989) 1883
4. H. Shuman and P. Kruit, Rev. Sci. Instrum. 56 (1985) 231
5. H. Shuman and A. P. Somlyo, Ultramicrosc. 21 (1987) 23
6. We would like to thank Professor S Hanada for provision of the Ni₃Al specimens and one of us (CBB) acknowledges the support of the Japan Key Technology Centre.

Fig. 1.--Energy loss spectra from Ni₃Al-0.5at%B after background subtraction. a) One spectrum acquired for 100s at 0.5 eV/channel and b) after dividing by gain spectrum. c) Average of 8 spectra similar to a) shifted by from 9 to 35 channels with respect to each other. d) After iterative averaging.

Fig. 2.--Iterative averaging for 2 shifted spectra assuming 2 channels have a higher gain than the rest. a) Original spectra. b) After averaging over corresponding energy losses to give a less noisy spectrum. c) Channel gain spectrum obtained by dividing original spectra a) by average spectrum b) and averaging over corresponding channels. d) Original spectra a) divided by average channel gain spectrum c).

Fig. 3.--Second difference spectra obtained using the inset filter (figures in channels). a) Ni₃Al-0.5%B with the boron edge at 188 eV arrowed. b) Ni₃Al with no B.