Tailoring interface alloying and magnetic properties in (111) Permalloy/Pt multilayers

Figure 1 shows the XRD pattern from Py/Pt multilayers deposited on the p-Si (001) substrate with native oxide and 100 nm thermally grown oxide. The vertical dashed lines in black and red indicate (111) peak position for the bulk state Pt (ICDD 00-001-1190) and Py (ICDD 01-071-8324), respectively. It can be seen that a single (111) Py/Pt multilayer peak appears between the bulk Py and Pt peaks. Such a single multilayer peak have previously been reported for ultrathin Fe/V (Pálsson et al., 2008), Fe/Au Rafaja et al. (2002a) and Co/Au (den Broeder et al., 1988) (with 5.4, 12 and 13.1% lattice mismatch, respectively). Rafaja et al. (2002a) also showed that a hard annealing process can result in discontinuous layers that returns the XRD spectra to two separate peaks at the respective bulk peak positions (Rafaja et al., 2002a). Thus, such a decomposition of a multilayer peak into peaks and their shift towards the bulk positions is an indication of discontinuous layers. For instance discontinuity as small as 15% has been detected for Fe (26 Å)/Au (24 Å) (Rafaja et al., 2002a). For thinner layers, such as reported on here, smaller discontinuities should be detectable while thicker ones might already present a peak at the bulk position as strain has relaxed. Thus, our XRD results present characteristics of continuous layers. This indicates that all the Py and Pt layers are under in-plane tensile and compressive strain, respectively. This is expected due to the large 10% lattice mismatch at the interface, that does not relax unless the thickness of the individual layers exceed a few nm in thickness (Rafaja et al., 2002a; Corrêa et al., 2016). To the left of the (111) peak, there are two satellite peaks indicated by $-1$ and $-2$ and the distance to those is proportional to the multilayer period ($\Lambda$). It is worth mentioning that extracting period thickness from XRD at high angles might be tricky due to the fact that satellite peaks are caused by diffraction while XRR fringes are the result of interference as discussed by Pálsson et al. (2008). Asymmetric intensity of satellite peaks, e.g. $+1$ versus $-1$, is a characteristic of strain normal to the film plane Vartanyants et al. (2000). Here we do not observe positive satellite peaks except for a very small $+1$ peak for the multilayer deposited by dcMS at 150 W on both substrates and 50 W dcMS on the 100 nm thermally grown oxide substrate (These are hardly visible on the scale used in Figure 1 but are magnified in insets in Figure 4). Absence of the $+1$ peak is indicative of huge strain normal to the film caused by in-plane epitaxial strains. This strain has been shown to be relaxed by precise annealing, rendering the satellite peaks more symmetric (den Broeder et al., 1988). Thus, one can consider Py/Pt multilayers that exhibit $+1$ peaks to be slightly more relaxed than their counterparts. The sharp peaks in Figure 1 are characteristics of single crystal substrate e.g. the largest peak at 69.9^∘ is the Si (004) peak. The smaller sharp peak that sometimes appears at 34.6^∘ belongs to Si (002) planes which is a structurally forbidden peak, although it sometimes shows a weak presence as discussed by Zaumseil (2015). Due to the lack of Py/Pt (002) and (022) peaks, we conclude that these multilayers have a strong $\langle 111\rangle$ texture normal to the substrate surface. Note that the Ta layer is found to be amorphous and thus it presents no peak in the XRD pattern (Ritchie et al., 2002). It is also worth mentioning that the main (111) peaks observed from the multilayers prepared by HiPIMS deposition of the Py layers are slightly shifted towards smaller angles. Such a shift can be interpreted as a slightly larger lattice spacing normal to the film and a contraction in-plane. The latter can be considered as more pure Py layer, with less alloying with the Pt as will be discussed in more details in the following.

Furthermore, we studied our multilayers using polar mapping, i.e. scanning in-plane ($\phi$) and out-of-plane ($\psi$) angles while $\theta-2\theta$ angle is fixed to the main (111) multilayer or satellite peaks. As an example, the results for the main (111) and $-1$ satellite peaks for a multilayer deposited by dcMS at 150 W on the smooth 100 nm thermally grown oxide are shown in figure 2. All the other (111) pole figures (not shown here) consisted of an intense spot at $\psi=0$, corresponding to the film normal, and a much weaker ring between $\psi=70-80^{\circ}$ corresponding to the (11$\bar{1}$) planes. Note that the relaxed (11$\bar{1}$) plane must appear at $\psi=70.5^{\circ}$. The dots at $\psi=45^{\circ}$ belong to the (113) Si planes. They are an indication of correct alignment of the instrument.

Figure 3 shows the XRR result of a Py/Pt multilayer deposited by dcMS and HiPIMS on (001) p-Si with native oxide and 100 nm thermally grown oxide. The values of $\alpha$, $\beta$ and $\Lambda$ indicated in the figure are inversely proportional to the total thickness, Ta under-layer thickness and period thicknesses, respectively. Here, the Ta acts as a buffer layer to decrease substrate roughness, while also promoting texture. It can be seen that the $\beta$ fringes from the Ta layer can be detected only when utilizing a smooth substrate (100 nm thermally grown oxide). However, for the native oxide substrate, these fringes disappear due to high roughness at the SiO₂/Ta interface. Using both substrates the $\Lambda$ fringes can be detected up to 10^∘ which essentially means that the Ta layer can successfully improve surface roughness enough for growing delicate multilayers. For the case of HiPIMS deposition on the 100 nm thermally grown oxide, the amplitude of the $\alpha$ fringes decay much slower with the angle of incidence and they are visible up to $2\theta=7.3^{\circ}$. This means that HiPIMS deposition of the Py layers gives a smoother surface for the whole stack. As mentioned earlier, surface roughness commonly increases with the number of periods. It appears here, however, that HiPIMS deposition contributes to smoothening of the Py layers and consequently smoother interfaces, resulting in a smoother top surface for the multilayer stacks.

It has been shown that the discontinuity of individual layers can cause a shift of the foremost Kiessig fringe, indicated by vertical dashed line, towards smaller angles (Rafaja et al., 2002b). We observed a negligible shift (0.005^∘) to the left for dcMS deposition case at 50 W on the 100 nm thermally grown oxide substrate. On the other hand, multilayers deposited by HiPIMS on different substrates and pressures present 0.01 – 0.02^∘ shift towards higher angles. Thus, based on this criteria there is lower probability of discontinuous layers within multilayers prepared by HiPIMS than by dcMS.

To determine alloying of the interfaces, we fit the XRR results carefully. The XRR fit gives the thickness and the density of the individual layers, and gives the interface roughness quantitatively. It has been shown that the XRR technique is a reliable method to study multilayers i.e. thickness and discontinuity of each layer as well as interface roughness (den Broeder et al., 1988; Chason and Mayer, 1997; Moreau-Luchaire, 2016; Azzawi et al., 2016). It is worth mentioning that the interface roughness ($\delta$) is not an absolute value and changes in relation with the roughness of the whole stack (Baumbach and Mikulik, 1999) (cf. $\delta$ in figure 3 (b) that is proportional to $\Lambda$ and $\alpha$ intensity). For this reason, during the fitting process we gave more weight to the density and thickness and thus sacrificed the precision of the interface roughness. It is also worth mentioning that thanks to the large probing area of the XRR method (on the order of millimeters), surface roughness larger than film thickness is detectable (Foll, 2019, § 3.2.1). We use the density of each layer to determine the interface mixing considering the fact that Pt layers shrink and Py layers expand in the film plane (cf. XRD results). Thus, for an ideal Pt layer it is expected that its density increases above the bulk value of 21.45 g/cm³. On the other hand, substitution of Pt with Ni and Fe that occurs at a mixed interface reduces the density due to much lower mass of the substituent. Using the same principle, we expect a density that is lower than 8.72 g/cm³ for a pure Py layer due to lattice strain, while diffusion of Pt into the layer will counteract the decrease in density. As pointed out by Baumbach and Mikulik (1999), for rough multilayers, such as the ones deposited on the native oxide, it is important to understand the interface morphology, e.g. stepped versus Gaussian and its replication distance, before the fitting process. We apply the same fitting procedure to multilayers deposited on both substrates that may lead to inaccurate results for the native oxide.

The results of the fitting are summarized in table 1. Utilizing dcMS deposition on native oxide gives higher density for the Py layer than when it is deposited by the HiPIMS counterpart. The higher density indicates the existence of Pt in the Py layer which is evidence of a diffused interface. Although utilizing HiPIMS when depositing on native oxide can reduce the Pt diffusion, the value 8.98 g/cm³ is still above 8.72 g/cm³ (the bulk value), and the single layer density observed in our previous works (8.6 – 8.7 g/cm³ (Kateb and Ingvarsson, 2017; Kateb et al., 2018)). The density of the Pt layers is higher than the bulk value of 21.45 g/cm³ for the dcMS deposited cases as expected from the compressive strain. However, the Pt layer density is lower than the bulk value for the HiPIMS deposited counterparts which is evidence of Ni and Fe diffusion into Pt.

It was expected that using the smooth 100 nm thermally grown oxide substrate would effectively enhance the sharpness of the interfaces. However, for Py layers deposited by dcMS at 50 W, the densities indicate a negligible improvement compared to the native oxide, i.e. there is still diffusion of Pt into the Py layer. Depositing at 150 W with dcMS, the Py and Pt densities reach 8.72 and 20.69 – 21.31 g/cm³, respectively. Thus, using a smooth substrate, an increase in dcMS power reduces the diffusion of Pt into Py but it causes diffusion of Ni and Fe into Pt. On the other hand, HiPIMS deposition gives the lowest Py density of 8.61 – 8.68 g/cm³, indicating the least diffusion of Pt into Py, accompanied by a high density of 23.54 g/cm³ for the Pt layers, indicating negligible diffusion of Ni and Fe into Pt.

For working gas pressure of 0.25 Pa, dcMS deposition at 50 W on a substrate with 100 nm thermally grown oxide gives higher density for both Pt and Py than a deposition at a pressure of 0.4 Pa. This means even more diffusion of Pt into Py but little diffusion of Ni and Fe into Pt. For 150 W dcMS deposition the Py density does not change with the working gas pressure but the Pt density increases, i.e. there is less diffusion of Ni and Fe compared to at 0.4 Pa. We also observed that there is a noticeable change in $\Lambda$ ($t_{\rm Py}+t_{\rm Pt}$) with pressure changes for 50 W dcMS. This means that at low power dcMS operation the deposition rate, and consequently the thickness of individual layers, are very sensitive to the variation in the working gas pressure. Thus, depositing with dcMS at low power, provides an extremely narrow and inflexible pressure window, which is undesirable. The latter can be improved by depositing at 150 W dcMS, i.e. a limited variation in $\Lambda$ in the 0.25 – 0.4 Pa pressure range. Using HiPIMS, there is a negligible change in the Py density and the Pt density is maintained at the lower pressure. However, the multilayer period becomes smaller, which indicates that lowering the pressure decreases the deposition rate of Py. This is probably due to the delay between voltage and current onset which shows larger change with pressure variation below 0.33 Pa (Kateb et al., 2018).

A question that may arise here is how an increase in sputter power by going from 50 to 150 W dcMS, and a further increase in instantaneous power by going to HiPIMS deposition decreases the interface alloying. Regarding the alloying by energetic ions, it has been shown that ions with bombarding energy of a few hundred eV can only penetrate into a few nm depth if the mass density is lower than 80% of theoretical mass density (Müller, 1986a, b). Using HiPIMS for the Py deposition leads to a higher film mass density which makes such penetration rather impossible. However, thermal spikes associated with energetic collisions might in-fact redistribute atoms at the interface (Kateb et al., 2019c). This can be considered as inherent alloying. Thus, higher power here does not contribute to interface alloying in the conventional way of diminishing of the interface through bombardment and displacement of sublayer atoms. Quite the opposite, at higher powers the top surface becomes smoother and there is no increase in roughness with increased number of periods. We have recently shown, using an atomistic simulation, that the increase in the ionized flux fraction in fact reduces surface roughness of a single metallic layer (Kateb et al., 2019c, 2020). The lower roughness is evident from the extent of $\alpha$ fringes at higher incident angles in Figure 3 (b). A similar conclusion has been made by comparison of PLD with MBE. Although MBE grown layers are atomically sharp (1 Å) in terms of alloying, the interface roughness might become tens of nm Johnson et al. (1996). On the other hand highly ionized deposition methods such as PLD and HiPIMS provide smooth surfaces and their interface alloying can be controlled e.g. by the correct choice of pulse length or other deposition parameters.

To address the question of how switching properties and in-plane anisotropy is affected by intermixing we measured the magnetic hysteresis loops by MOKE. The MOKE experiments do not yield absolute values of magnetization, so the traces have been normalized to the saturation value in each case. Prior to these measurements, the direction of the magnetic easy axes had been determined by measuring hysteresis loops with the applied field at different angles in the film plane. The legend in each subfigure indicates the axis along which magnetic field is applied, and substrate type. For the 50 W dcMS deposited stack shown in figure 4 (a), the substrate with native oxide gives an open hard axis trace and a well defined easy axis, while utilizing a smooth 100 nm thermally grown oxide substrate, enhances the squareness of the easy axis and presents a perfectly closed and almost linear hard axis trace. For the 150 W dcMS deposited stack, as shown in figure 4 (b), both substrates result in very well defined hard and easy axes. In this case, the 100 nm thermally grown oxide substrate results in a smaller anisotropy field, $H_{\rm k}$, but it also leads to a small hysteresis in the middle of the hard axis. This is barely visible in the figure, but is distinguished by a slope change in the hard axis loop at a around $\pm$4 Oe.

Figure 4 (c) shows the magnetic response of the HiPIMS deposited stack which exhibits an open hard axis and square easy axis regardless of the substrate type (different smoothness). Note that these loops were obtained after examining different angles in the film plane. Thus they present the hardest and easiest directions in the film plane. The coercivity in this case is slightly higher than for both the dcMS counterparts. In general, opening in hard axis loops can be caused by domains whose magnetization directions deviate slightly from the easy axis direction, for whatever reason. This could be brought on by grain structure, surface roughness, or by interfacial strain. It has been shown that alloying Py with Pt can cause an increase in coercivity but the reports on the anisotropy field are controversial (Chen et al., 1991; Ingvarsson et al., 2004). This is not the case here since we have shown by fitting the XRR results that there is minimum alloying for multilayers prepared by HiPIMS deposition compared to their dcMS counterparts on the same substrate. The insets in Figure 4 display a blown up portion of the XRD data in figure 1, where the $+1$ satellite peak is expected, between 47.5 – 48^∘. As stated above, strain relaxation is accompanied by the appearance of the $+1$ satellite peak, and more symmetric satellite peaks in general. It is interesting to note that the open hard axis loops occur when there is no discernible $+1$ satellite peak, i.e. in the most strained films, HiPIMS deposited Py (Figure 4 (c)), or when it is barely detectable as in dcMS deposited Py at 50 W on native oxide and dcMS deposited Py at 150 W on 100 nm thermal oxide. The samples in which there are hints of the $+1$ satellite peak all have perfectly closed hard axis loops. This suggests a magnetoelastic response to the high epitaxial strain at the sharp Py/Pt interfaces, i.e. inverse magnetostriction, is responsible for the opening in the hard axis loops and increased coercivity. In dcMS deposited counterparts, such a strain is relaxed by e.g. diffusion of Pt with larger atomic radius into stretched Py or vice versa. Except for the case of 50 W dcMS deposition on native oxide, which similar to HiPIMS does not present $+1$ satellite peak (cf. Figure 1), and is more strained. A summary of the anisotropy field ($H_{\rm k}$) and coercivity ($H_{\rm c}$) of multilayers prepared by both methods and different pressure can be found in table 2.

Furthermore, we studied the effect of strain by substituting Pt with Cu and Cu₅₀Pt₅₀ at. %. Cu has a lattice constant of 3.61 Å close to that of Py (3.54 Å) while Pt has a much larger lattice constant of 3.92 Å. Figure 5 shows the MOKE response of Py/Cu, Py/CuPt, and Py/Pt, multilayers deposited on the substrate with 100 nm thermally grown oxide. All the samples were deposited using identical conditions i.e. using HiPIMS for Py deposition and dcMS for NM layer deposition at a constant deposition rate. It can be seen that for the Py/Cu multilayers, very soft, but also very well defined uniaxial anisotropy is obtained. However, as the lattice constant of the NM increases upon substitution of Cu by Pt, the increase in strain at the interface leads to a higher coercivity and opening up in the hard axis trace. This is in agreement with the results for the single layer Py (Ohtani et al., 2013; Kim and Silva, 1996; Hollingworth et al., 2003a, b). It has been shown that both polycrystalline (but textured) and single crystal (111) Py present almost isotropic magnetization in the film plane due to the strain (Ohtani et al., 2013). It is worth mentioning that there exists a report on the correlation between magnetostriction coefficient and surface roughness (Kim and Silva, 1996). In the latter study, the surface roughness changed as a consequence of change in layer thickness. Later, it was shown that surface roughness is rather unimportant compared to the effect of interfacial strain (Hollingworth et al., 2003a, b).