Tailoring the soft magnetic properties of sputtered multilayers by microstructure engineering for high frequency applications

The continuous trend towards miniaturization with increased functionality for internet-based mobile systems increasingly requires that various high-performance devices (e.g. III-V electronic/optoelectronic devices, low-power sensors, RF filters, micro- and nanoscale transducers, inductors and transformers, MEMS actuators, etc.) be combined with silicon complementary metal-oxide-semiconductor (Si-CMOS) technology in a monolithic integrated circuit (IC) fabrication process.^1–3 As far as the CMOS-integrated magnetic and electro-magnetic devices are concerned, soft magnetic layers (typically several microns thick) with tunable uniaxial anisotropy are key for applications requiring operation from high to ultra-high frequencies (from a few MHz to several GHz) and low power consumption.^4,5 These films should have high saturation magnetization (M_s) and electrical resistivity (ρ), and a well-defined anisotropy with a specific in-plane EA and anisotropy constant (K_u).⁶ The AC permeability (μ) perpendicular to EA remains largely constant up to the ferromagnetic resonance (FMR) frequency of the material (⁠$fr∼Hk4πMs$⁠, where H_k is the anisotropy field).⁷ Thus, by tuning M_s and H_k with appropriate material choices one can adapt to different frequency requirements (e.g. GHz applications typically require films with H_k>20 Oe). Moreover, to avoid hysteresis loss (i.e. coercivity H_c≲1 Oe) these films are laminated with intermediate non-magnetic interlayers, which not only stop the crystallite growth, but also reduce the Eddy current loss.

In the case of magnetron sputtering, the in-plane uniaxial magnetic anisotropy is strongly affected by the film microstructure/stress that can be controlled by process parameters (e.g. temperature, pressure, power, etc.).^8–11 Moreover, it is well known that the amount of disorder at the surfaces and interfaces, which is also influenced by process parameters, can greatly disturb the soft magnetic properties of thin films, such as coercivity, magnetic domain structure, magnetization reversal and magnetic anisotropy.^12–17 Therefore, when tailoring the soft magnetic multilayers for high-frequency operation it is important to predict and control the microstructure and surface/interface roughness.

In this paper, we present a systematic study on the interdependence of structural and magnetic properties of Ni_78.5Fe_21.5, Co_91.5Ta_4.5Zr₄ and Fe₅₂Co₂₈B₂₀ soft magnetic thin films laminated with SiO₂, Al₂O₃, AlN, and Ta₂O₅ dielectric interlayers, sputtered in an industrial, high-throughput magnetron sputtering system. In view of their large saturation magnetizations, e.g. ∼(0.8-1) T for Ni_78.5Fe_21.5, ∼(1.5-1.7) T for Co_91.5Ta_4.5Zr₄ and ∼(1.8-2) T for Fe₅₂Co₂₈B₂₀, these soft magnetic materials currently receive much attention for their potential for high frequency applications.^18–25 

We prepare multilayers based on the Ni_78.5Fe_21.5, Co_91.5Ta_4.5Zr₄ and Fe₅₂Co₂₈B₂₀ soft magnetic alloys with different interlayer materials, interlayer thickness, total film thickness, process parameters and angular distribution of the sputtered material. More specifically, we deposit these films on 8” bare Si wafers and Si/200nm-thermal-SiO₂ wafers in the high-throughput Evatec LLS-EVO-II sputter system. The system is equipped with five cathodes placed in a vacuum chamber (base pressure <10^-8 mbar), a moveable shutter and a rotating substrate cage housing that allows nine 8” wafers to be processed in the same batch (see Fig. S1 in the supplementary material for the schematics of this system). The magnetic sublayers, which represent the greater fraction of the soft magnetic multilayer stack, are deposited by DC and pulsed DC (duty cycle 40% unless otherwise specified) sputtering at ∼(0.5-2.0)×10^-3 mbar from long-life (nominal 250 kWh) targets, which allow ∼1000 8” wafers to be sputtered with 1 μm thick soft magnetic material. The non-magnetic (AlN, Ta₂O₅, Al₂O₃, SiO₂) interlayers are deposited by RF or reactive DC sputtering from monoblock Al, Ta, Al₂O₃ and SiO₂ targets at ∼5×10^-3 mbar.

To induce in-plane magnetic anisotropy one option is to resort to modifications to sputter geometry allowed by the LLS-EVO-II system. By reducing the speed with which the Si wafers mounted on the rotating cage move into a position facing the sputter cathodes, we can increase the fraction of time during which there is an oblique incidence of the sputtered material, which leads to a horizontal linear order in the growing films. By using horizontal or vertical collimators (HC, VC) with different aspect ratios a:b (collimator plate height: collimator plate separation), this effect can be enhanced or reduced, giving additional freedom to engineer the magnetic anisotropy. The other option for inducing uniaxial anisotropy is by depositing the films in the presence of a linear magnetic field parallel to the wafer plane, which is designed such that the magnetic field of the magnetron located behind the opposite target is not perturbed.

Magnetic properties of the sputtered multilayers, e.g. H_c, H_k and the angular dispersion of EA (D), are mapped on the 8” wafers by means of MOKE. Surface morphology and root-mean-square roughness (R_q) are measured by AFM using Park-Systems-NX20 and Bruker-Dimension-Icon3 microscopes. The degree of crystallinity is probed by grazing incidence X-ray diffraction (GIXRD) using a Bruker-D8-Discover diffractometer. Layer thicknesses and surface/interface roughness are determined by XRR with specular scans and RSMs that are collected at various azimuths between 0° (∥EA) to 90°(⊥EA) using a Rigaku-SmartLab-3kW diffractometer.

Figure 1 shows the surface morphology probed by AFM of various soft magnetic multilayers sputtered in the LLS-EVO-II system. The DC-sputtered NiFe multilayers with Ta₂O₅ and AlN interlayers exhibit a random, granular microstructure with no visible order (Fig. 1(a)). GIXRD measurements reveal the crystal structure of these films as face-centered-cubic (fcc) with almost no preferential texture. The R_q of the 1 μm thick NiFe multilayers with 2 nm, 4 nm and 8 nm thick Ta₂O₅ interlayers are 1.89 nm, 1.92 nm and 2.01 nm, respectively. Moreover, the NiFe multilayers with AlN interlayers exhibit similar surface morphology and roughness. In contrast, the multilayers based on CoTaZr and FeCoB alloys exhibit surface morphologies consisting of ripples elongated perpendicular to the incident flux direction (Fig. 1(b–e)). These surface ripples are a consequence of the self-steering effect during deposition, and they usually develop for oblique incidence deposition at small incidence angles.^26,27 No crystallographic phases are present in the GIXRD patterns, which indicates that these multilayers are amorphous. We find that the total thickness of the film not only affects the roughness but also surface morphology (Fig. 1(b,d)) where by increasing the thickness of the CoTaZr/Al₂O₃ multilayers from 1 μm to 4 μm the elongated surface ripples become more pronounced. One-dimensional (1D) AFM scans reveal the periodicity of the ripples as ∼22.5 nm.

Surface morphology of the multilayer films can be controlled by using collimators for both the magnetic and non-magnetic interlayers. Thus, by sputtering the magnetic layers with a 2:1 HC (sample S1 in Table I), the surface ripples become more evident (Fig. 1(e)); periodicity of the ripples determined from 1D-AFM scans is 35 nm. In this case, the horizontal oblique incidence effect caused by the substrate rotation is enhanced, whereas the vertical oblique incidence contribution from the elongated target is suppressed. The difference between the ripple periodicity for the CoTaZr/Al₂O₃ and FeCoB/AlN multilayers could be due to the different surface (Al₂O₃ vs. AlN buffer layer) and energy of the incoming particles (DC vs. pulsed DC).^28,29 When in addition to the 2:1 HC for the FeCoB layers we use a 1:1 VC for the AlN interlayer (sample S2 in Table I), the surface morphology no longer exhibits elongated ripples (Fig. 1(f)) even though surface roughness remains unchanged (∼1 nm).

Additionally, we observe that the process pressure and the type of sputter process (e.g. DC, pulsed DC) has a large impact on the roughness and surface morphology (Fig. 1(g–i) and Table I), and we expect this to also greatly affect the magnetic properties. We should only point out that when a very low process pressure of 5×10^-4 mbar is used for the FeCoB layers, the surface roughness of the 1 μm thick FeCoB/AlN multilayer is reduced by a factor of 7.7 (i.e. from 1.15 nm to 0.15 nm), and the surface morphology no longer exhibits ripples.

To obtain insight about the roughness of the buried interfaces, we perform XRR measurements. Figure 2(a) compares the specular XRR scans of 80nm-CoTaZr/4nm-Al₂O₃ multilayers with 6, 12 and 48 periods, corresponding to total thicknesses of ∼0.5μm, 1μm and 4μm, respectively. For comparison, the specular scan of 80nm-NiFe/4nm-Ta₂O₅ multilayer with 12 periods (total thickness ∼1μm) is also shown. We determine the individual layer thickness and the interface roughness by fitting the specular scans with a model that assumes the layer roughness to increase from the lowest interface to the top surface; an example of such a fit is presented in Fig. 2a for the CoTaZr/Al₂O₃ multilayer with 6 periods. Consequently, for the 6-period CoTaZr/Al₂O₃ multilayer XRR provides layer thicknesses/roughnesses of 81.1 nm/∼0.4 nm (CoTaZr) and 4.5 nm/∼0.2 nm (Al₂O₃), consistent with AFM (R_q=0.27 nm). Moreover, we find the individual layer roughness to increase from ∼0.2 nm at the lowest interface to ∼0.4 nm, ∼0.6 nm and ∼1 nm at the top surface for the 6, 12 and 48-period structures, respectively. For comparison, the interface roughness of the NiFe/Ta₂O₅ multilayer with 4 nm thick Ta₂O₅ interlayers is much higher, e.g. ∼1.8 nm, in very good agreement with AFM (1.89 nm). In the case of FeCoB-based multilayers with different laminating materials, by fitting the specular XRR scans in Fig. 2(b) we find the Al₂O₃ interlayers to provide the smoothest FeCoB interfaces (<0.3 nm) where the roughness of the dielectric interlayers is ∼0.4 nm, which is also in good agreement with AFM (R_q=0.38 nm). Furthermore, we also investigate the effect of interlayer thickness on surface and interface roughness. Thus, for FeCoB/AlN multilayers XRR shows that the interface roughness of AlN interlayers increases from ∼0.4 nm to ∼0.6 nm when their thickness increases from 4 nm to 20 nm; interface roughness of the top FeCoB layers is ∼0.6 nm. Similar results are obtained for NiFe/Ta₂O₅ multilayers with Ta₂O₅ interlayers.

The XRR technique is used to also probe the diffuse scattering (i.e. off-specular region in the reciprocal space), which can give unique information about the correlation of interfacial roughness between successive layers.^30–32 Figure 2(c,d) shows RSMs of 48×(80nm-CoZrTa/4nm-Al₂O₃) multilayer for two azimuths, e.g. perpendicular and parallel to the surface ripples, respectively. Consequently, the anisotropy of the multilayer stack is confirmed by the more intense diffused scattering for the measurement with the X-ray beam path along the surface ripples, e.g. for Q_z=0.2238 Å^-1 the intensity of the diffused scattering increases by a factor ∼2.

Figure 3(a,b) shows the MOKE hysteresis loops for ∼1 μm thick 12×(80nm-NiFe/4nm-AlN), 12×(80nm-CoTaZr/4nm-AlN) and 10×(100nm-FeCoB/4nm-AlN) multilayers for EA and HA orientations; the corresponding H_c and H_k values are 0.08 Oe/0.12 Oe/0.20 Oe and 8.5 Oe/17.1 Oe/25.5 Oe, respectively. Moreover, we see that D rapidly deteriorates, while H_c/H_k increases/decreases when the interlayer thickness increases. This is consistent with the results discussed in the previous section, where roughness is larger for the multilayers with thicker interlayers. Furthermore, increasing the total thickness of the multilayer stack only slightly deteriorates the soft magnetic properties, as shown in Fig. 3(h–j) for 48×(80nm-CoTaZr/Al₂O₃). Thus, in spite of the complex, 4μm thick structure, consisting of almost 100 interfaces, we obtain superior soft magnetic properties with excellent uniformity across the 8” wafer, e.g. D<±2°, H_c∼0.2 Oe and H_k∼17.1 Oe.

Finally, we want to emphasize the key role of process conditions in controlling the soft magnetic properties of sputtered multilayers. Figure 3(f,g) compares the hysteresis loops for EA and HA of FeCoB/AlN multilayers sputtered with different process conditions (see also Table I). Magnetic anisotropy in these structures is induced by the aligning magnetic field and the oblique incidence inherent to the LLS-EVO-II system, the latter being adjusted by means of linear collimators. The soft magnetic properties of these structures are summarized in Table I. Thus, by choosing the appropriate process parameters one can tune H_k from ∼29 Oe to ∼120 Oe, while H_c<0.9 Oe and D goes between ±1.5° and ±4°.

In summary, we performed a systematic study of the structural and magnetic properties of high quality, soft magnetic multilayers based on the Ni_78.5Fe_21.5, Co_91.5Ta_4.5Zr₄ and Fe₅₂Co₂₈B₂₀ alloys, deposited with the industrial, high-throughput Evatec LL-EVO-II magnetron sputtering system. We showed how soft magnetic properties of these multilayered structures can be adjusted by means of microstructure engineering. We provide an industrial sputtering solution for high-quality soft magnetic multilayers with tunable magnetic properties for applications requiring operation from high to ultra-high frequencies.

See supplementary material for schematics of the LLS-EVO-II tool.

We are grateful to Silvan Wüthrich, Maurus Tschirky and Balz Gubser for continued support and stimulating discussions, Heinz Felzer and Arnold Ammann for technical assistance, and the Swiss Scanning Probe Microscopy User Lab at Empa for providing access to their facilities. We also acknowledge financial support from the Swiss Federal Commision for Technology and Innovation (CTI-Project Nr. 18940.2 PFNM-NM). One author wishes to acknowledge the project ED1.1.00/02.0068. The research was also financially supported by the MEYS of the Czech Republic under the project CEITEC 2020 (Project Nr. LQ1601).