Magnetism in metastable and annealed compositionally complex alloys

Compositionally complex materials (CCMs) present a potential paradigm shift in the design of magnetic materials. These alloys exhibit long-range structural order coupled with limited or no chemical order. As a result, extreme local environments exist with a large variations in the magnetic energy terms, which can manifest large changes in the magnetic behavior. In the current work, the magnetic properties of (Cr, Mn, Fe, Ni) alloys are presented. These materials were prepared by room-temperature combinatorial sputtering, resulting in a range of compositions with a single bcc structural phase and no chemical ordering. The combinatorial growth technique allows CCMs to be prepared outside of their thermodynamically stable phase, enabling the exploration of otherwise inaccessible order. The mixed ferromagnetic and antiferromagnetic interactions in these alloys causes frustrated magnetic behavior, which results in an extremely low coercivity $(<1\mathrm{mT})$, which increases rapidly at 50 K. At low temperatures, the coercivity achieves values of nearly 500 mT, which is comparable to some high-anisotropy magnetic materials. Commensurate with the divergent coercivity is an atypical drop in the temperature dependent magnetization. These effects are explained by a mixed magnetic phase model, consisting of ferro-, antiferro-, and frustrated magnetic regions, and are rationalized by simulations. A machine-learning algorithm is employed to visualize the parameter space and inform the development of subsequent compositions. Annealing the samples at 600 °C orders the sample, more-than doubling the Curie temperature and increasing the saturation magnetization by as much as 5×. Simultaneously, the large coercivities are suppressed, resulting in magnetic behavior that is largely temperature independent over a range of 350 K. The ability to transform from a hard magnet to a soft magnet over a narrow temperature range makes these materials promising for heat-assisted recording technologies.

Figure 1

Composition map showing at.% of (a) Mn, (b) Cr, (c) Ni, and (d) Fe. The circular illustration is representative of the 100 mm diameter wafer. (e) X-ray diffraction of the as-grown wafer measured across the equatorial band of the wafer (five measurements), ranging between (Cr-27%, Mn-43%, Fe-13%, Ni-17%) in red and (Cr-53%, Mn-7%, Fe-31%, Ni-9%) in black.

Figure 2

(a) X-ray diffraction data for the 17 measured points, (b) select measurements with identified peaks, and (c) map of phases on wafer, with measurement locations identified.

Figure 3

Major hysteresis loops measured at $2.5\text{–}400\mathrm{K}$ for ${\mathrm{S}}^{\mathrm{K}}$, (Cr-33%, Mn-12%, Fe-39%, Ni-16%) (a),(b) as-grown and (c),(d) after annealing. Panels on the right are zoomed in views of the shaded blue boxes. (e) Saturation magnetization (left, black, solid) and field cooling magnetization (right, blue, open) for the as-grown and annealed samples. (f) Coercivity of the as-grown and annealed samples and exchange bias of the as-grown sample versus temperature.

Figure 4

Saturation magnetization, M-T plots and major loop coercivity at 2.5 K for (a), (c), (e) the as-grown sample and (b), (d), (f) the annealed sample. Contour lines in (a), (b) indicate the average number of valence electrons, based on the composition.

Figure 5

(a) Saturation magnetization plotted against valence, with a comparison to the Slater-Pauling curve shown in the inset. The linear line fitted to the as-grown data has a reduced chi-squared value of 0.67. Heat maps of (b) the drop in the M-T plots and (c) the major loop exchange bias at 2.5 K. Simulated magnetic ordering in (d) sample A, (e) sample C and (f) sample E, with red representing ferromagnetic clusters and blue antiferromagnetic.

Figure 6

Predicted values of (a), (b) saturation magnetism, (c), (d) coercivity, and (e), (f) bias field across the (Cr, Mn, Fe, Ni) parameter space using Gaussian process regression (left) and random forest regression (right). The red circles denote the approximate boundary of the thin film wafer overlaid onto the parameter space. The black points denote the 18 samples used to train the regression model (numbered 1–19, sample 6 was not measured). The contours show level sets of the predicted values. The black points on the perimeter of the parameter space represent theoretical values added to the training data.