Buffer layer engineering of L1 FePd thin films with large perpendicular magnetic anisotropy

Ultrathin magnetic films with perpendicular magnetic anisotropy (PMA) are of significant technological interest for emerging magnetic random access memory (MRAM) devices to be applied in the internet of things (IoT), aerospace/defense storage, embedded memory, etc.^1–4 The performance (e.g., ultrahigh-density and ultralow-energy) of the magnetic tunnel junction (MTJ) devices (a key building block of MRAM) switched by spin-transfer-torque mainly depends on the magnetic properties of the ferromagnetic layers, such as the PMA value (K_u), Gilbert damping (α), and thermal stability (Δ = K_uV/k_BT, where V is the volume of the free layer in MTJs, k_B is Boltzmann’s constant, and T is ambient temperature).^5–9 For sub-10 nm node MTJs, a large K_u (∼MJ/m³) and small α values are required to maintain ten-year data storage and low switching current density for next-generation spin memory and logic devices. With the well-known interfacial PMA materials (e.g., Ta/CoFeB/MgO), relatively low K_u and high α¹⁰ make it challenging to satisfy these requirements.^11,12

Recently, L1₀ FePd exhibiting large K_u (∼MJ/m³) and low α (<0.01) has attracted considerable interest for spin memory device applications. These properties have created interest particularly in implementing FePd films for electric-field control of magnetism,^13,14 magnetostrictive transducers,^15,16 ultrahigh density granular storage media,^17,18 and spintronics.^19–25 The magnetic properties of the FePd thin films can be tuned though various growth conditions for specific applications. For spintronics applications, the large K_u and low α values are, in particular, demanded, which mainly rely on a well-controlled orientation of the tetragonal phase.^23,26–29 Moreover, practical spintronic implementations are likely to include a metallic buffering layer between the substrate and the FePd film to function as an electrode or spin Hall channel. While Pt- and Ru-based buffer layers have recently been implemented with highly ordered FePd thin films, alternative buffer materials such as Ir and Rh could also deliver advantageous performance in the static and dynamic properties of FePd or enhanced performance in interfacial Dzyaloshinskii–Moriya interactions^30,31 or spin–orbit torques.³² Importantly, understanding the magnetic properties of FePd films engineered using buffer layers is a significant step toward high-performance FePd-based spintronic devices.

The buffer layers - Cr/(Pt, Ru, Ir, Rh), Mo/Ir, and Ir - are designed to determine whether a closer lattice match between the buffer layer and FePd could co-optimize the degree of L1₀ order and (001) growth orientation along with high perpendicular magnetic anisotropy and low Gilbert damping.^28,29,33,34 A refractory Cr layer serves as an adhesion layer over the MgO substrate and reduces the lattice mismatch between the substrate and the Pt, Ru, Ir, Rh buffer layers. In addition, the refractory Mo adhesion layer and the single Ir buffer layer are also considered due to their high temperature processing stability in spintronic devices.^35,36 All of the buffer layers show highly oriented (001) growth transferred from the (001) MgO substrate and induce highly ordered (001) texture in the overlying FePd thin films, except for the Mo/Ir buffer layers in which the out-of-plane growth directions are Mo(001)/Ir(011)/FePd(011). The Cr/Ru and Cr/Pt buffer layers facilitate the growth of L1₀-ordered FePd thin films with large K_u (0.9 MJ/m³ for Cr/Ru and 1.2 MJ/m³ for Cr/Pt) and low α (0.008 for Cr/Ru and 0.005 for Cr/Pt).

All films with a stack of MgO (001) sub./buffer layer/FePd (8 nm)/Ru (2 nm)/Ta (3 nm) were deposited at 350 °C using ultrahigh vacuum dc magnetron sputtering in a chamber whose base pressure is below 3 × 10⁻⁷ Pa. The Ar working pressure is ∼0.40 Pa during sputter deposition of the layers. The 1.5 in. diameter sputtering targets were operated at a constant power of 80 W except FePd, which was operated at 60 W. Samples were prepared using a combined buffer layer, including a 15-nm Cr layer, followed by a 4-nm layer of either Ir, Pt, Rh, or Ru, and an additional specimen was grown on top of an 15-nm Mo layer, followed by a 4-nm Ir layer. One specimen with only a 15-nm Ir buffer layer was produced for comparison. For this specimen only, the Ir buffer layer was grown at 500 °C before the temperature was reduced to 350 °C for growth of the subsequent layers. All samples were in situ post-annealed at 500 °C for 30 min, which has been demonstrated to be adequate for Pd-buffered FePd films with the high-ordered L1₀-phase.²⁹ The magnetic and structural properties of all samples were characterized by broadband ferromagnetic resonance (FMR) spectroscopy, vibrating sample magnetometry (VSM), magneto-optic Kerr effect (MOKE), and x-ray diffraction (XRD).

Figure 1 shows the crystallographic properties of all the FePd films characterized by out-of-plane and in-plane XRD measurements measured by a Rigaku Smartlab diffractometer in a parallel beam geometry and using a 14-kW rotating anode source (Cu Kα).³⁷ The peaks at approximately 24° and 49° correspond to the out-of-plane (001) and (002) peaks for L1₀ FePd, which are superlattice and fundamental peaks, respectively. As shown in Fig. 1(a), clear (001) and (002) peaks are observed for the FePd samples with the Ir, Cr/Ru, Cr/Pt, Cr/Rh, and Cr/Ir buffer layers, indicating that the FePd thin films possess the L1₀ phase. However, the (001) and (002) peaks are absent for the FePd thin film with the Mo/Ir buffer layer, and the (022) Ir and FePd reflections were present along with the (002) Mo peak reflection for this specimen. The absence of either Ir or FePd peaks associated with the (001) or (002) family with the Mo/Ir buffer layer is consistent with a 45° rotation of the Ir/FePd bilayers (001) axis away from the surface normal, which is shown in supplementary material, Fig. S3, with x-ray diffraction under a 45° chi tilt of the specimen away from the surface normal. Figure 1(b) shows the in-plane XRD patterns, in which the (200) epitaxial peak is observed in the FePd thin films with Cr/Pt, Cr/Ru, Cr/Ir, Cr/Rh and Ir buffer layers. Based on the out-of-plane and in-plane XRD results, the ordering parameter (S) and lattice parameters (a and c) of all the FePd thin films are estimated by fitting the peak positions and integrated intensity to pseudo-Voigt functions,^29,38,39 as summarized in Table I along with the one-sigma fitting uncertainty. Here, S is equal to the square root of the integrated peak intensity ratio (I₁₀₀/I₂₀₀) divided by the theoretical intensity ratio. It appears that the Cr/Pt buffer layer gives the highest S ∼ 0.82, followed by Cr/Ru (S ∼ 0.68), Cr/Ir (S ∼ 0.64), Ir (S ∼ 0.54), and Cr/Rh (S ∼ 0.47).

In order to investigate the magnetic properties of the FePd thin films induced by the different buffer layers, the magnetic hysteresis (M-H) loops of all the samples were measured, as shown in Fig. 2. All samples have an apparent out-of-plane easy axis, except for the Mo/Ir buffer layer, consistent with the preferred out-of-plane magnetization direction dictated by the (001) orientation. All the FePd samples show a large out-of-plane remanence (⁠$Mr⊥/MS$ > 0.8) except the FePd with the Ir buffer layer, which shows relatively lower remanence and correspondingly low hard axis saturation field (0.7 T) combined with the largest out-of-plane lattice parameter (c = 0.3762 nm) in the series. The in-plane magnetic hysteresis loops all show an inflection point around 0.7 T but with varying amounts of hysteresis in the low field range that correlate somewhat with the degree of L1₀ ordering in each sample. Furthermore, the FePd thin film with a Cr/Ir buffer layer possesses large remanence and a hard axis saturation field (1.2 T), indicating stronger PMA energy than the Ir buffered FePd sample, while maintaining a shorter c axis (c = 0.3708 nm). This marked qualitative difference in the magnetization behavior is despite both samples having moderate order parameters—0.54 vs 0.68 for the Ir and Cr/Ir buffer layers, respectively.

The saturation magnetization (M_s) of each of the FePd films within this series was calculated based on the M-H loops, as shown in Table I. We found that the M_s value varied between 1000 kA/m and 1100 kA/m for all the samples except for the markedly lower value (880 kA/m) for the Cr/Rh buffered sample. To understand the origin of the low M_s value observed in the Cr/Rh buffered FePd thin film, we carried out specular x-ray reflectivity measurements of the FePd samples with Cr/Rh, Mo/Ir, and Ir buffer layers, presented in Figs. S5, S1, and S4, respectively. The layer structure model for each sample is given in Tables S1–S3, respectively. In the Rh-buffered film we observe significantly more accelerated intensity decay, indicative of higher overall roughness. The roughness may also reflect intermixing at the Rh/FePd interface rendering a magnetically dead layer that could explain the apparent low saturation magnetization.

To better understand the variation in the out-of-plane coercivity across samples, polar magneto-optic Kerr effect (MOKE) imaging of the magnetic domains is shown in Fig. 3. Images were taken at the out-of-plane coercive field determined by the M-H loops shown in Fig. 2. The magnetic domain preparation was carried out by applying a moderate negative field (−0.25 T) for ∼1 min and changed to the positive coercive field for each sample. A background image was acquired at the negative field in saturation. The acquired image at the coercive field was enhanced for magnetic domain contrast by subtracting the background image. The Cr/Ru and Cr/Pt buffered FePd samples exhibit larger domains at coercivity than the Cr/Ir and Ir buffered samples, respectively, which is consistent with the softer magnetic properties (lower coercivity) of the Cr/Ru and Cr/Pt buffered samples relative to the Cr/Ir buffered samples. The Ir buffered sample shows significant lateral inhomogeneity and small domains consistent with the large coercivity observed in the macroscopic M-H loop. Interestingly, the Cr/Rh buffered sample exhibits relatively large domains at coercivity, despite exhibiting a relatively moderate coercive field. As suggested above, this distinctive magnetic domain configuration could result from intermixing of the Cr/Rh buffered FePd sample.

FMR measurements were used to quantify the anisotropy field and α with the different buffers. For each sample, absorption was measured as a function of applied field over a range of frequencies to ascertain both the FMR field vs frequency (f) spectrum and the linewidth vs frequency spectrum. Figure 4(a) shows representative absorption curves for the Cr/Ru buffered FePd sample at four different frequencies. The line shapes shown in Fig. 4(a) are each fit with the derivative of a Lorentzian line shape, from which the resonance field (H_res, the zero-crossing) and linewidth (ΔH, the full-width at half maximum of the integrated intensity) are obtained. The FMR relationship between the microwave excitation frequency and H_res for fields applied normal to the thin film plane and ΔH vs f relationship are given by^40,41

Here, H_res is the resonance field, γ = gµ_B/h is the gyromagnetic ratio, g is the spectroscopic g-factor, μ_B is the Bohr magneton, h is the Planck constant, μ₀ is the vacuum permeability, $μ0Heff⊥=μ0HK−μ0MS$ is the effective anisotropy field comprising the perpendicular magnetic anisotropy field and the demagnetizing field, α is the Gilbert damping constant, and ∆H₀ is the inhomogeneous linewidth broadening. Figure 4(b) shows the frequency-dependence of the FMR field with applied out-of-plane external fields. The effective anisotropy field of each sample was evaluated by fitting the frequency vs resonance field data in Fig. 4(b) to the linear expression in Eq. (1) and is shown in Table I with the one-sigma uncertainty from the best-fit parameters. The M_s values were used to calculate the demagnetization field and K_u for each sample shown in Table I. One can see that S correlates with K_u [Fig. S12(a), plots K_u vs S]. Interestingly, the Cr/Ru buffered sample (0.36 T) has a significantly lower anisotropy field, despite having an order parameter falling between the Cr/Pt buffered and the other samples. One immediately notes that the Cr/Ru buffered sample has a significantly smaller c and higher degree of tetragonality than the other samples (c = 0.365 nm, c/a = 0.93), which correlates with higher L1₀ order and higher anisotropy due to the influence of strain on L1₀ ordering during the growth process.⁴² The relatively low anisotropy in this stacking structure may instead be related to diffusion of Fe out of the FePd layer in Cr/Ru/FePd structures shown recently by Zhang et al.⁴³ 

The Ir and Cr/Rh buffered FePd samples exhibited remarkably broad absorption lines (see Fig. S8) that made it impractical to estimate the FMR linewidth for both specimens, the FMR field for the Ir buffered FePd sample, and consequently the anisotropy field. From the best-fit line to our ΔH vs f data shown in Fig. 4(c), we extract the α values, as shown in Table I. The α value for the samples range from 0.008 to 0.018 but, unlike the K_u values, do not correlate particularly well with S [see Fig. S12(b)]. The data present two scales of α: moderate-high α for Ir, Cr/Rh, and Cr/Ir buffered samples and lower α for Cr/Pt and Cr/Ru buffered samples. The enhanced α in Rh-based alloys⁴⁴ and large spin-pumping from Ir⁴⁵ are consistent with this observation.

We tested this conjecture regarding the dependence of α in FePd on film thickness. A series of Cr/Pt buffered thick FePd samples with 8 nm, 12 nm, and 16 nm were grown (see Sec. 3 in the supplementary material). The anisotropy field increased monotonically from 0.59 T to 0.73 T as the thickness increased. Moreover, α decreases from 0.012 (t_FePd = 8 nm) to 0.005 (t_FePd = 16 nm), which is consistent with an interpretation that interfacial effects were enhancing the measured α of our FePd layers.⁴⁶ Meanwhile, the estimated S of these films decreased slightly over this range, despite the nearly 60% decrease in α in the 16-nm thin film. While the technological advantage of engineering FePd samples with large K_u and low α on metallic buffer layers is evident, we also seek to compare the performance of the samples described here with a comparable FePd thin film grown directly on the MgO(001) substrate. With the empirical data on K_u and α being absent, we produced an additional 8-nm FePd film directly on MgO(001), processed using identical deposition and annealing conditions carried out on the other underlayer sequences. As summarized in Figs. S7 and S11, we estimate S of 0.64, K_u of 0.75 MJ/m³, and α of 0.017, which is comparable with the Cr/Ir buffered sample and unexpectedly large, given the lack of metallic buffer layers to enhance α through spin pumping. Our results indicate that a buffering layer sequence is advantageous for integrating high K_u, low α FePd films into spintronic applications.

In summary, the magnetic and structural properties of a series of 8-nm FePd thin films with different buffer layers were investigated. The largest K_u ∼ 1.2 MJ/m³ was achieved for the Cr/Pt buffered FePd sample, consistent with a high S ∼ 0.82. The Cr/Ru buffered FePd sample shows the lowest α ∼ 0.008, despite having a lower S ∼ 0.64 and a lower K_u ∼ 0.87 MJ/m³. The lower anisotropy is attributed to intermixing of the Ru buffer and capping layers with the FePd layer annealed at 500 °C. The non-monotonic relationship between α and S led to the study of the Cr/Pt buffered samples with different thicknesses of FePd. Our results show a modest decrease in S with increasing FePd thickness but a nearly 60% reduction in α. These results demonstrate that interfacial effects such as spin-pumping can generate significantly higher α in ultrathin FePd layers. The most promising combination of thermal stability (Δ ≈ 107) with ultralow α is observed in a 16-nm FePd with Cr/Pt buffer layer, in which K_u and α were 0.91 MJ/m³ and 0.005, respectively. We have demonstrated buffer layers that produce FePd films with a high level of L1₀ ordering, large K_u, and low α, consistent with the engineering of devices that require sub-10 nm lateral size for ultrafast switching and ultralow-energy spin memory.

Additional structural and magnetic measurements and associated calculations supporting this study can be found in the supplementary material.

This work was funded, in part, by DARPA HR001117S0056-FP-042 “Advanced MTJs for computation in and near random access memory” and NIST. Additional measurement support was provided by the CharFac of the University of Minnesota and the NIST Nanofabrication Facility. We acknowledge fruitful discussions with B. Kirby (NIST) and M. Shi (Rigaku).