Three-stage ultrafast demagnetization dynamics in a monolayer ferromagnet
Ultrafast demagnetization triggered by shining a femtosecond laser pulse onto a ferromagnetic transition-metal sample has been extensively studied since its discovery. In spite of numerous investigations, the mechanism underlying light-induced demagnetization could not yet be clearly identified. A variety of different microscopic models has been put forward over the past years to explain how an ultrashort laser pulse could modify the magnetic system within a few hundred femtoseconds after laser excitation. Most are based on ultrafast spin-flip scattering of some kind, such as Elliott-Yafet electron-phonon spin-flip scattering, electron-magnon spin-flip scattering, and Coulomb exchange spin-flip scattering. Other possibilities are direct laser-induced or relativistic electromagnetic-radiation-induced spin-flips. Regardless of the diverse nature of these mechanisms, all scenarios adopt the same first step in the excitation process, namely direct absorption of a femtosecond laser pulse in the ferromagnetic film. Conversely, in this work we explore a different process that leads to very efficient ultrafast demagnetization, namely transport of laser-excited non-spin-polarized electrons into the ferromagnet.
Recently, it has been predicted that spin-dependent transport of laser-excited hot electrons may play a central role in ultrafast demagnetization: through superdiffusive transport the spin polarization is displaced out of a ferromagnetic Ni film and transferred into an adjacent non-magnetic Al layer. The resulting demagnetization in the Ni film is ultrafast and sufficiently large to explain the experimental data. To experimentally separate the effects of the demagnetization due to direct laser excitation and transport of hot non-equilibrium (NEQ) electrons we reverse here the role of the two layers and investigate how hot electrons that are generated in a non-magnetic layer influence the adjacent ferromagnet.
We use a femtosecond laser pulse to generate hot NEQ electrons in a 30-nm-thick Au layer, which covers a 15 nm ferromagnetic Ni film, on an Al substrate with a thin Pt seed layer (see the sketch in Fig. 1). The Au layer thickness is designed specifically to absorb nearly all of the laser pump pulse that generates the hot electrons: 90% of the laser pulse energy is deposited in the 30 nm Au layer. Only an estimated 7% of the laser energy reaches the Ni film compared with 70% that reach Ni in the reference sample (see the respective intensity profiles in Fig. 1b). As a consequence, optical methods working in the visible photon energy range suffer from the same extinction length effects and are thus less suited to detect changes of magnetization in the underlying Ni film. Recent experimental work either constrained the total sample thickness, or used excitation from the back side through a transparent substrate.
To circumvent these limitations, we probe the Ni magnetic state by transmitting soft X-rays through the sample and recording the temporal evolution of the magnetic moment by means of time-resolved X-ray magnetic circular dichroism (XMCD; see Methods). This ensures a reliable measurement of the Ni magnetization underneath the Au layer. For comparison, we also measure the demagnetization caused by direct femtosecond optical excitation of the Ni layer, using a 20 nm Ni layer on an Al foil, covered with a thin Pt layer to prevent oxidation. The measurements are complemented by superdiffusive transport theory to calculate the laser-induced demagnetization in the two heterostructures. Doing so, we prove that hot, non-thermal electrons travelling from Au into the Ni film are causing the observed magnetization decrease on the ultrafast timescale.
In Fig. 2 we show the measured ultrafast magnetization dynamics of the Ni magnetic moment for the Au/Ni structure and compare them with the Ni reference film. Both samples were measured quasi-simultaneously, frequently changing back and forth between them during the measurements. This ensures that sample and reference have the same point of zero time delay. Two laser pump fluences were used, resulting in different grades of demagnetization. In the high-fluence setting shown in Fig. 2a, the Au/Ni sample was pumped with an incident fluence of 33 mJ cm, and the Ni reference with 10 mJ cm to achieve similar quenching amplitudes for the magnetization. The higher pumping fluence for Au/Ni is necessary because of the much higher reflectance of Au compared with the Pt cap on the Ni reference: R = 0.88, R = 0.52, measured on the very same samples under experimental conditions (35° angle between laser beam and sample surface normal, p-polarization, wavelength 780 nm). The absorbed fluence then becomes 4.0 mJ cm and 4.8 mJ cm for the Au/Ni and Ni reference samples, respectively.
As a consequence of the excitation by the femtosecond laser pulse, the magnetic moment is quenched very efficiently in both samples, dropping down to ≈20% of its original value. The drop proceeds on the sub-picosecond timescale with the indirectly excited Ni film underneath the Au layer reacting slightly slower than the directly excited Ni reference. To quantify the speed of demagnetization, we fit the experimental data points with an exponential decay, plotted as lines in Fig. 2 (for details, see Methods). The resulting decay times τ are 330±40 fs for the Au/Ni structure and 140±10 fs for the Ni reference in the high-fluence scenario plotted in Fig. 2a. Note that the onset of the demagnetization process in the Au/Ni sample seems to be slightly delayed by 25±30 fs with respect to the Ni reference. As this value lies within the experimental error, we need further evidence before commenting on its origin.
A very similar behaviour is found when reducing the excitation power (Fig. 2b). Here we excited the Au/Ni sample with 23 mJ cm and the Ni reference with 7 mJ cm, resulting in a demagnetization of ≈50%. The fitted time constants τ are 400±160 fs for Au/Ni and 185±30 fs for Ni, slightly longer than the ones measured with higher laser fluence (the larger error bars are due to a shorter integration time during this measurement). The fits to the data in Fig. 2b include a second exponential time constant τ = 1 ps to account for the recovery of the magnetic signal. Another difference from the data collected with high pump fluence concerns the onset of the demagnetization: from the fits in Fig. 2b we now get a clear delay of 70±40 fs for the Au/Ni sample, confirming our initial assumption. This delayed response of the Ni magnetization in the Au/Ni sample points towards some kind of transport mechanism playing a vital role during the demagnetization process.
The surprising finding is that the demagnetization of the Au/Ni sample is as efficient as that of the Ni reference sample, even though essentially no light reaches the Ni film in the Au/Ni structure. This contradicts present knowledge on the origin of the demagnetization process, and raises the key question of how such an ultrafast demagnetization can be accomplished if almost no optical excitations are created in the magnet. We anticipate that hot electrons travelling from the Au film into the Ni film substantially cause the demagnetization of Ni on an ultrafast timescale. Consistent with this expectation is the slightly slower time constant found for the Au/Ni sample as compared with the Ni reference sample, and the delay in the onset of demagnetization (Fig. 2). Note that these two observations also exclude that the demagnetization is caused by laser light shining through an Au layer that would not be completely opaque for the pump pulse.
To examine in detail the role of hot electron currents in the observed demagnetization process, we perform computational modelling of the measurements using superdiffusive transport theory. In this description NEQ electrons are initially created through laser excitation, whereupon they start to move ballistically through the material. These hot, non-thermal electrons undergo a thermalization process by inelastic scattering, losing energy and generating cascade electrons. These scattering processes are fully treated in the superdiffusive transport model. We emphasize that within the transport calculations we assume all scattering processes to be spin conserving; that is, there is no channel of ultrafast dissipation of spin momentum invoked. The only source of magnetization dynamics is spin transport by superdiffusion. Furthermore, the full layered structure of both samples is taken into account in the calculations, as well as the absorption profile and time structure of the pump laser pulse (for further details, see Methods).
Solving the superdiffusive spin-transport equation numerically we obtain the temporal evolution of the magnetization M(z,t) in the Ni layer along the z axis normal to the layers. The spatially averaged time-resolved magnetization of the Ni layers of both studied systems is shown in Fig. 3. Here we have plotted the time-dependent magnetizations computed for different fluences, requesting an almost equal demagnetization for both systems, as was done in the experiments. In the theoretical modelling this requirement is achieved for an incident fluence of 33 mJ cm for the Au/Ni sample and 13 mJ cm for the Ni reference film, corresponding to 4.0 and 6.3 mJ cm absorbed pump fluence, respectively. Note that the theoretical model assumes a simple linear response to the absorbed fluence, but there could exist nonlinear effects that come into play. Fitting the theoretical curves with the same exponential function as used for the experimental data (see Methods) we obtain a decay time of 300 fs for the Au/Ni structure and 250 fs for the Ni reference. The theoretical magnetization decay of the Au/Ni sample agrees well with the experiment, but that of the Ni sample is larger. This overestimation might be due to several approximations made in the model. In particular, ab initio-calculated electron velocities and lifetimes have to be extrapolated to the low-energy range, leading to an uncertainty of ≈20% in both amplitude and decay time (see refs 8, 15 for a discussion). Any uncertainty in the decay times following from fitting to the exponential function lies well below this range.
To unveil how the femtosecond demagnetization takes place in the Au/Ni we show in Fig. 4 the computed time- and position-dependent magnetization change ΔM(z,t) for both configurations. In the Au/Ni sample the laser light is absorbed mainly in the Au layer. The excited hot electrons move fast towards the Ni layer. The thus-generated NEQ electron current is not spin polarized; that is, it consists of equal amounts of spin-majority and -minority electrons. Once they reach the Ni layer, they start to behave distinctly according to their spin alignment with the Ni magnetization. Owing to their good transport properties, spin-majority electrons do not scatter much and can traverse the Ni layer into the Al substrate. Conversely, spin-minority electrons possess poorer transport properties and become, to a first approximation, trapped within the Ni layer, creating the demagnetization region depicted in Fig. 4a. Also, as the non-thermal spin-minority electrons undergo many inelastic scatterings they create cascade electrons of both minority and majority spin, which induces further demagnetization due to the outflow of spin-majority electrons into the Au and Al.
In the Ni reference sample the laser light excites electrons directly in the ferromagnetic layer. The hot NEQ spin-minority electrons are again much less mobile and effectively trapped near to where they were created. Instead, the mobile spin-majority hot electrons diffuse rapidly out of the Ni layer. A small fraction escapes towards the Pt capping but the main part travels towards the Al substrate. The demagnetization in Ni near the Ni/Pt/Al interface is generated by the back flux of electrons excited by the laser and by cascade electrons created in Al. Note that the NEQ spin transport occurring until electron thermalization has taken place is distinctly different from standard diffusion that would lead only to a small exchange of spin-polarized electrons near a non-magnetic/ferromagnetic interface.
Our result has direct implications on the intense debate about the mechanism of ultrafast laser-induced demagnetization. First, we observe that direct optical excitation is not a precondition for the demagnetization. This excludes that spin-flips, directly induced through the spin-laser field interaction, could be responsible. Rather, in our experimental configuration we find that transport is the dominant factor to explain the demagnetization of the Au/Ni structure. Second, several other spin-flip mechanisms, based on an enhanced electron-quasiparticle scattering of the laser-excited electrons, such as, for example, electron-phonon or electron-magnon scattering, become much less likely as main driving mechanism of demagnetization. A phonon-mediated spin-flip mechanism seemed compelling recently; however, ab initio calculations could not confirm Elliott-Yafet electron-phonon spin-flips as the major driving mechanism of ultrafast demagnetization, and neither could electron-electron spin-flip scattering be confirmed. As only little light reaches the Ni film underneath the Au layer, the spin-polarized electrons and also the phonons (or magnons) in the Ni film remain in near equilibrium conditions; that is, they do not experience a quick rise of the electron temperature. Although such quasiparticle spin-flip scatterings are always present, here they are not expected to be responsible for the ultrafast demagnetization, at least until the Ni film heats up. The absorption of most of the pump pulse in the Au film leads to a rise of the electron temperature in Au. This laser-induced heating of the Au has to transfer to the Ni layer. Heat propagates with the speed of sound, which is 3.2 nm ps in Au (ref. 20) and ~5.6 nm ps in Ni (ref. 21), implying that conventional heating of the Ni film would commence only after several picoseconds. In our experiment it could still be that hot NEQ electrons reach the magnetic layer fast enough to rapidly heat the phonons and magnons in it. Recent experiments on Ni however showed that this heating requires 1-2 ps for the phonons. The observed demagnetization in Au/Ni however proceeds much faster (within ≈0.5 ps), which deems a purely thermally driven demagnetization unlikely for the Au/Ni system. Our observations thus pin down the observed ultrafast demagnetization as a transport effect; that is, on the femtosecond timescale spins are displaced out of the magnetic layer. On longer timescales of several picoseconds thermal spin diffusion and spin-flip scattering processes come into play that will lead to a subsequent magnetization relaxation and re-magnetization.
We come to the conclusion that spin transport rather than spin-flips plays the major role in magnetic processes on the femtosecond timescale. Our experiments have shown that a pulse of NEQ hot electrons generated in a non-magnetic layer can efficiently accomplish ultrafast demagnetization of a ferromagnetic film. In strong contrast to existing knowledge we find that direct opticalexcitation is not a precondition for ultrafast demagnetization, but that the demagnetization through a hot electron current is as efficient as that created through direct laser irradiation. Our observations are accurately explained by the superdiffusive transport theory of hot NEQ electrons and provide decisive arguments for unravelling the mechanism of ultrafast demagnetization.