Optical control of magnetization dynamics in Gd–Fe–Co films with different compositions

Ultrafast manipulation of magnetization in ferrimagnetic thin films by femtosecond laser pulses has been intensively studied, both experimentally and theoretically. The magnetization dynamics has gradually become understood, and other physical phenomena have been discovered in related studies, inspiring strong interest in nonequilibrium magnetism. Circularly polarized light elicits direction-selective magnetization rotation or switching in ferrimagnetic alloys,^1–6) which has attracted much attention for its scientific significance and application potential. Even linearly polarized light triggers clear magnetization reversal in 4f–3d ferrimagnetic alloys.^7,8)

However, the responses of materials to laser pulses are not limited to magnetization switching. The electric and magnetic field vectors of laser beams can interact strongly with electrons in materials, generating coherent magnons^9–12) and plasmons.¹³⁾ Laser-triggered superdiffusive spin currents have transiently modified the equilibrium spin populations in specially designed magnetic structures.^14,15) These examples show that ultrafast pulsed lasers can not only control the magnetic states in custom-designed materials and structures, but can also answer fundamental research questions.

In this paper, we expose Gd–Fe–Co ferrimagnetic films with different compositions and multilayer structures to femtosecond laser pulses and systematically study their magnetization responses. Figure 1 schematizes the temperature-dependent spin dynamics around the angular-momentum-compensation point in ferrimagnetic systems.¹⁶⁾ When the angular-momentum-compensation temperature T_A is below or above the experimental temperature T_exp, we expect prolonged magnetization precession or rapid magnetization switching, respectively. In a Gd–Fe–Co film with a Gd content of 22% [T_A < T_exp (room temperature, RT)], we observed anomalous wavelike modulation. In a detailed analysis, the wavelike magnetic modulation was explained well by intense resonant precession with weak damping, accompanied by a heat-induced lateral shift in the precession as an extrinsic perturbation.

Zoom In Zoom Out Reset image size

Perpendicularly magnetized Gd–Fe–Co thin films were grown by magnetron sputtering. The designed multilayer structure was Ta (1 nm)/Ru (1 nm)/Gd–Fe–Co (20 nm)/Ru (20 nm)/Ta (5 nm)/glass (substrate). For comparison, we tested Gd₂₂Fe₇₀Co₈ and Gd₂₆Fe₆₆Co₈ films with and without a 5-nm-thick Si₃N₄ heat-blocking layer grown between the Gd₂₆Fe₆₆Co₈ (20 nm) and Ru (20 nm) layers (hereafter, these three samples are referred to as Gd22%, Gd26%, and Gd26%–Si₃N₄, respectively). Time-resolved photoemission electron microscopy (PEEM) experiments were performed at the BL25SU soft X-ray beamline at SPring-8 using the X-ray magnetic circular dichroism (XMCD) effect at the Gd M₅ absorption edge. The experimental setup for the time-resolved XMCD-PEEM measurement is similar to that used in our previous study,¹⁷⁾ and a timing diagram of the pump–probe measurements is presented in Fig. S1 in the online supplementary data at http://stacks.iop.org/APEX/10/103002/mmedia. Linearly polarized laser pulses (λ = 800 nm, 120 fs, 5 kHz) were incident at 60° from the sample surface normal. The laser energy was set to ∼4 µJ per pulse, and the radiation spot on the sample surface was elliptical with an approximate size of 125 × 250 µm² (full width at half-maximum), which corresponds to a power density of ∼16 mJ/cm². After each pump–probe cycle, the magnetization of the sample was initialized by a magnetic field pulse (XMCD-PEEM observations were performed under zero magnetic field). Static XMCD-PEEM imaging at room temperature confirmed that in the Gd26% and Gd22% samples, the Gd magnetization was aligned parallel and antiparallel to the direction of the applied magnetic field, respectively (data not shown). This indicates that the magnetization-compensation temperatures (T_M) were above and below RT in Gd26% and Gd22%, respectively (to compare the dynamics on the same gray-scale contrast, we applied the magnetic field pulse to the Gd26% and Gd22% samples in opposite directions). Apart from one comparative experiment [shown later in Fig. 4(b)], all the experiments were performed at RT.

Figure 2(a) shows a static XMCD-PEEM image of the initially homogeneously magnetized Gd26% sample after in situ excitation by a single ultrashort laser pulse with a fluence of ∼16 mJ/cm². The magnetization was clearly reversed from the initial spin-down (black region) direction to the spin-up (white) direction in an elliptical area around the center of the excitation spot. Because the T_A value of this sample exceeds RT, the magnetization precession should be abruptly damped; that is, the local magnetization vector should quickly settle in the reversed direction.¹⁶⁾ Figure 2(b) shows the result for the Gd22% sample under the same excitation conditions. In this sample, there is no homogeneous magnetization reversal, but a fine-grained multidomain pattern formed, suggesting that most of the excitation energy was converted into intense spin precessions rather than magnetization reversal.

Zoom In Zoom Out Reset image size

Figures 2(c) and 2(d) show time-dependent XMCD-PEEM images of the Gd26% and Gd22% samples, respectively. We define Δt as the delay time after laser irradiation. In the Gd26% sample [Fig. 2(c)], the magnetization generally settled to its final state approximately 1 ns after excitation. However, in the Gd22% sample [Fig. 2(d)], a packet of wavelike magnetization modulation propagated isotropically along the radial direction from 800 ps to 5 ns post-irradiation (Movie S2 in the online supplementary data at http://stacks.iop.org/APEX/10/103002/mmedia). Note that no sign of a wavelike feature appeared in the Gd26% sample.

Figure 3(a) traces the XMCD intensities of the Gd22% sample at Δt = 1.5 ns. The profile was traced along line AO [marked at the top and in the inset of Fig. 3(a)]. Also plotted are the net magnetization distribution (M_net) excluding the modulating component (blue dashed line) and the sinusoidal curve fitted to the data (red solid line). The fitting parameters [wavelengths (λ) and XMCD signal amplitudes (|A_XMCD|)] in regions I–III of Fig. 3(a) are summarized in Table I. The spatial variation in the parameters reflects the inhomogeneity in the laser-beam intensity profile. However, the XMCD modulation amplitudes (|A_XMCD|) are surprisingly large overall, ranging from 20 to 32% of M_net.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figures 3(b) and 3(c) show complete profiles of the evolving space- and time-resolved magnetization and the wave shape simulated from this set of transient curves, respectively. In the calculated oscillation of the XMCD intensity, precessional spin motion was assumed. The XMCD signal was detected as a projected component of the magnetization parallel to the SR direction [see Fig. 3(d)]. For simplicity, we assumed that the precession frequency, wavelength, and angle of the waves were spatially and temporally uniform. The effects of the finite temporal resolution and the relevant sampling points are presented in Fig. S3 in the online supplementary data at http://stacks.iop.org/APEX/10/103002/mmedia. In Fig. 3(b), the wave frequency appears to be between 100 MHz and 1 GHz (period = 1–10 ns). However, considering an aliasing effect due to discrete and unevenly spaced sampling, a much higher frequency was needed to reproduce the features of the experimental profiles [sampling points are indicated by blue dots in Fig. 3(b)]. After varying the frequency parameter in the waveform simulations, we found that settings of 6.3, 8.3, 10.3, ... GHz rather than of 100 MHz–1 GHz yielded strong agreement between the simulated and experimental waveforms [the waveform at an 8.5 GHz precession frequency is shown in Fig. 3(c)]. These frequencies are consistent with the ferromagnetic resonance (FMR) frequencies expected in ferrimagnetic systems.¹⁶⁾ The precession angle simply estimated from the modulation of the XMCD intensity was approximately 20° (±10°). However, when accounting for the finite temporal resolution of the experiment (50 ps), which reduced the amplitude of the 8.3 GHz waves, the precession angle approximately doubled [to ∼40° (±20°)].

Note that the XMCD signals reflect both the out-of-plane and in-plane components of the magnetization [Fig. 3(d)]. Further, the XMCD-PEEM contrast exhibits a perfectly uniform concentric distribution. To match this observation, the precession phase of the magnetization should be identical along the circumferential direction [Fig. 3(e)]. In the absence of a magnetic field, the initial magnetization is directed perpendicular to the sample plane; hence, we infer that the incident direction of an obliquely incident laser pulse will determine the direction of the initial in-plane torque applied to the magnetization (the experimental geometry is detailed in Fig. S1 in the online supplementary data at http://stacks.iop.org/APEX/10/103002/mmedia). The wave front propagation in Fig. 2(d) is reminiscent of that of coherent magnons, but the nonzero momentum (traveling of the wave fronts) is likely produced by extrinsic factors such as lateral shifting of the precession phase induced by the radial heat gradient. External momentum sources are inferred from (i) the large wavelength (on the order of 10 µm), which exceeds that of magnons generated in metallic systems, (ii) the limitation of the observed modulation to the irradiated area, and (iii) the contraction and inward movement of the wave fronts toward the center of the spot (note that backward-volume magnetostatic waves⁹⁾ are ruled out by the experimental geometry, which directs the magnetization perpendicular to the propagation direction of the waves). Considerations (ii) and (iii) strongly suggest that the wave fronts are driven by thermal and effective-field gradients created within the illuminated area. Shrinkage of the waving region suggests that the precession initially relaxes at the edge of the irradiated spot, where the system is promptly cooled. This picture is consistent with the results in Table I. The precession (characterized by |A_XMCD| normalized by M_net) is less intense at the edge of the spot (region III) than in the central area (region I). The smaller wavelength in region III than in region I is attributed to the steep effective field created by adjacent magnetization standing perpendicular to the surface.

Although our simulation can consistently explain the wavelike modulations of the XMCD contrast by assuming (FMR-like) high-frequency magnetization precession, it does not uniquely determine the exact magnetization dynamics. To overcome this limitation, we further verified that the present magnetic modulation originates from the precessional motion (Fig. 1). Figure 4(a) shows the domain pattern of the Gd26%–Si₃N₄ sample excited in the same way as the Gd26% and Gd22% samples. The domain shape is complex and quite different from that of Gd26%, despite the identical composition of the two films. The pattern somewhat resembles that of the Gd22% sample [Fig. 2(b)] but develops no wavelike feature during the post-excitation sequence of nonequilibrium states [see Fig. 4(c)]. These peculiar results imply that the domain structures in Gd22% and Gd26%–Si₃N₄ are created by completely different mechanisms. In the Gd26%–Si₃N₄ sample, laser-induced heat diffuses very slowly through the Si₃N₄ heat-blocking layer, forming random domains. On the other hand, the Gd22% sample [Fig. 2(b)] possesses a fine domain structure originating from significant spin precession (as explained in Fig. 1). Figure 4(b) shows the domain structure of the Gd22% sample at a temperature much lower than the sample's T_A (namely, at 100 K). The film shows clear magnetic reversal, analogous to that observed in the Gd26% sample. This proves that the properties of the magnetization dynamics are determined by T_A relative to T_exp.¹⁶⁾ From these results and discussions, we conclude that the resonant precession inherent in low-T_A systems is essential to generating anomalous spin dynamics, such as those shown in Fig. 2(d).

In conclusion, the magnetic domain structures and spin dynamics of the Gd–Fe–Co layer in multilayer thin films can be controlled by changing the alloy composition of the Gd–Fe–Co layers or the multilayer structure of the film. Comparing our results with those of Ref. 18, which reported monotonic magnetization reversal even in Gd22% samples, we confirmed that the magnetization response of the sample depends not only on the T_A value of its Gd–Fe–Co layer, but also on its multilayer structure. Although the quasi-spin-waves observed in our experiments may not qualify as coherent magnons,^9–12) we achieved an unprecedentedly large precession angle of the induced waves (on the order of 10°). Considering that low-frequency precession is allowed within the uncertainty of our present experiment, and knowing that anomalous behavior occurs in samples with specific heat conduction, experimental and theoretical searches for spin–phonon coupling^19,20) may also be worthwhile. Detailed understanding of the diverse material responses to optical excitation is expected to provide novel information, enabling the design of magnetic memories or ultrafast on-chip communication using spin waves.

Acknowledgment