Layered structure analysis of GMR multilayers by X-ray reflectometry using the anomalous dispersion effect

As a basic layered structure for giant magnetoresistive (GMR) heads, NiFe/Cu/NiFe/Ta/Si substrate was measured by X-ray reflectometry at Cu Kα, Cu Kβ and Cu K-absorption-edge energies. The accuracy of both the Cu thickness and the interface width between the upper NiFe and the Cu layers was found to improve in the order Cu Kα < Cu Kβ < Cu K-edge. The final thickness and interface width values obtained from Cu Kβ reflectivity are in good agreement with those from the Cu K-edge. The anomalous-dispersion effect is useful in the more accurate analysis of the layered structure of transition metal multilayers because it causes a large difference in the refractive indices of specific elements near the absorption edge. The Kβ X-rays, which can be produced from conventional X-ray sources, are also available for the accurate analysis of reflectivity measurements.

1. Introduction

Giant magnetoresistive (GMR) spin valve heads have been investigated for high-recording-density rigid disk drives because of their high sensitivity in reading magnetic records (Dieny et al., 1991). The layered structure of the heads consists of two ferromagnetic layers separated by a noble metal spacer of a few nm thickness. Their magnetic properties, such as the magnetoresistance and interlayer coupling between the two ferromagnetic layers, strongly depend on the thickness and the interfacial roughness of each layer. Therefore, a precise structural characterization of the GMR multilayers is important for producing good heads and for improving their magnetic properties.

The X-ray reflectivity technique is a powerful tool for investigating layer thickness, electron density and interface roughness. Huang et al. (1992) applied it to GMR multilayers using the Cu Kα line from a conventional X-ray source. However, in transition metal multilayers, such as NiFe/Cu/NiFe GMR multilayers, the difference in the refractive index between NiFe and Cu at Cu Kα energy is too small to analyse precisely the layered structure, because of the lower intensity of specular X-rays reflected from NiFe/Cu interfaces. The refractive index is a strong energy-dependent variable and rapidly changes near the absorption edge of the material. Using this anomalous-dispersion effect to enhance the X-rays reflected from the interfaces, Bai et al. (1996) measured the composition profile on an Fe/Cr superlattice from the reflectivities around the K-edge of Fe and Cr. Usami et al. (1997) reported reflectivity measurements of NiFe/Cu/NiFe multilayers using a Cu Kβ line from a conventional X-ray source.

In this report, using a synchrotron radiation facility, the availability of reflectivity measurements using the dispersion effect was studied for the precise layered structure analysis of transition metal multilayers. Reflectivities of NiFe/Cu/NiFe multilayers were measured at Cu Kα, Cu Kβ and Cu K-edge energies. The accuracy of the layered structure analysis for each X-ray energy was investigated.

2. Experimental

The Fresnel reflection coefficient F_i,j between layers i and j is expressed as

where g = (n ² − cos²θ)^1/2, n = 1 − δ − iβ is the refractive index and θ is the grazing-incidence angle; the last approximation uses β ≃ 0 and θ ² 2δ. Because the smaller value of δ_i near the absorption edge of the i material increases the X-ray intensity from the interface between the layers i and j, one can analyse the layered structure of the multilayers with high accuracy. For example, in NiFe/Cu/NiFe multilayers, (δ_NiFe − δ_Cu)/δ_Cu can be calculated as 3, 13 and 30% at Cu Kα, Cu Kβ and Cu K-edge energies, respectively.

The Ni₈₁Fe₁₉(10)/Cu(10)/Ni₈₁Fe₁₉(10)/Ta(10)/Si substrate sample was deposited by RF magnetron sputtering. The numbers in parentheses are the nominal thicknesses in nm calculated from the deposit condition. Reflectivities of the sample were measured at BL8C2 at the KEK Photon Factory, Japan. X-rays monochromated through an Si(111) double-crystal monochromator were used to undertake the measurements. The X-ray wavelengths were 0.13805 nm (Cu K-edge), 0.1392 nm (Cu Kβ) and 0.1540 nm (Cu Kα). The incident X-ray intensity was typically 10 Mcounts s⁻¹ and the exposure time was 3 s per point. The reflectivity data were collected using the θ–2θ scanning technique and analysed by the least-squares method, which uses the reflectivity formula and includes interfacial effects due to roughness and/or interdiffusion (Parratt, 1954; Névot & Croce, 1980). The values of δ, thickness (t) and interface width (σ) for each layer were refined by minimizing χ²,

where I_exp and I_cal are the experimental and calculated reflectivity intensities, respectively. To evaluate the reliability of the least-squares refinement analysis, the reliability factor R was calculated from

A fitting model was used, containing an oxidized surface of NiFe and an interface layer between the Ta and Si substrates, because the fitting model containing the two layers drastically decreased the value of R.

3. Results and discussion

Fig. 1 shows the experimental and calculated reflectivities of the NiFe/Cu/NiFe/Ta/Si sample obtained with the Cu Kα, Cu Kβ and Cu K-edge X-rays. Each R factor was less than 1% and the refined reflectivity curves closely match the experimental data. The refined δ, t and σ values are listed in Table 1. The refined thickness measured at Cu Kβ was close to that at the Cu K-edge, within 0.1 nm, whereas the thickness at Cu Kα was different by about 0.3 nm. Similarly, the refined interface widths at Cu Kβ and at the Cu K-edge were in good agreement with each other, but the interface width at Cu Kα differed by up to 0.2 nm. Moreover, note that the interface width of the upper interface of Cu was larger by about 0.3 nm than that of the lower interface, whereas the interface width of NiFe did not change at the upper or lower interface. This is due to the larger size of the Cu grain due to crystal growth.

Table 1
Refined δ, thickness (t) and interface width (σ) of NiFe/Cu/NiFe/Ta/Si multilayers

Assumption: each δ of the upper and lower NiFe layer was the same.

	δ × 10^−6			t (nm)			σ (nm)
	Cu Kα	Cu Kβ	Cu K-edge	Cu Kα	Cu Kβ	Cu K-edge	Cu Kα	Cu Kβ	Cu K-edge
Oxide	9.00	12.57	14.08	1.24	1.36	1.43	0.72	0.80	0.83
NiFe	23.67	20.16	19.91	11.03	10.70	10.57	0.98	0.81	0.76
Cu	24.42	17.78	15.38	9.85	10.21	10.15	0.70	0.76	0.77
NiFe	23.67	20.16	19.91	12.03	11.74	11.68	0.40	0.49	0.50
Ta	38.55	30.68	30.18	10.67	10.68	10.69	0.42	0.44	0.47
Interface layer	9.39	8.00	7.80	1.57	1.39	1.39	0.40	0.37	0.41

Figure 1 Experimental and calculated reflectivities of the NiFe(10 nm)/Cu(10)/NiFe(10)/Ta(10)/Si multilayers measured for Cu Kα, Cu Kβ and Cu K-edge X-rays.

In order to investigate the accuracy of the layered structure analysis for each X-ray wavelength, the Cu thickness (t_Cu) was kept fixed to a value offset from the optimum (t_Cu,opt), other parameters were refined again and the fitting reliability was examined. Fig. 2(a) shows the χ² distribution, i.e. the difference of χ² from the minimum value (χ²_min) versus the offset value of the Cu thickness from the optimum. A similar analysis for the interface width between the Cu and upper NiFe layer (σ_Cu) and δ_Cu was undertaken and the results are shown in Figs. 2(b) and 2(c). As |t_Cu − t_Cu,opt| is larger in Fig. 2(a), χ² − χ²_min from the Cu K-edge measurement increases but χ² − χ²_min from the Cu Kα measurement almost never changes. This indicates that t_Cu from the Cu K-edge measurement can be refined more accurately than that from the Cu Kα measurement. The accuracy of the refined t_Cu was found to improve in the following order: Cu Kα < Cu Kβ < Cu K-edge X-ray energy. Similarly, for the interface width, σ_Cu refined from the Cu K-edge measurement was the most accurate of all the measurements. In contrast, the χ² distributions of δ_Cu are the same in all the measurements and the accuracy of δ_Cu is equivalent in every measurement. These results reveal that the anomalous-dispersion effect is highly accurate in analysing the layered structure of the transition metal multilayers. Because the refined parameters obtained from the Cu Kβ reflectivity were in good agreement with those from the Cu K-edge, the Kβ X-rays can also be harnessed for reflectivity measurements for more accurate analysis. This is important for controlling the deposit of the GMR multilayers based on reflectivity measurements, because the Kβ X-rays can easily be produced by conventional X-ray sources.

Figure 2 χ^2 distribution versus parameters of the Cu layer obtained by least-squares methods for the reflectivities shown in Fig. 1; (a) versus Cu thickness (tCu) offset from optimum, (b) versus interface width (σCu) between upper NiFe and Cu layer, (c) versus δ of Cu.