Study of reorientation processes in L10-ordered FePt thin films.

We report on the development of structural and magnetic order in epitaxially grown L10 FePt thin films. Upon annealing, the easy axis of magnetization changes from the out-of-plain into the in-plain direction. We found that the overall fraction of reoriented domains first increases but after certain time decreases before achieving a saturated state. The results are based on conversion electron Mössbauer spectroscopy studies and confirm Monte Carlo simulations in nano-layered FePt. We present a modified version of the Johnson-Mehl-Avrami (JMA) model adequately describing the experimental findings. Two dynamical processes, the first being a 2D-growth, dominate the initial state of sample annealing and the second being a 3D-growth, dominate the late stage close to saturation. From an Arrhenius plots of JMA coefficients for both processes we extracted the activation energies of the underlying dynamics which are 1.5(1) eV for disordering and 0.8(2) eV for ordering.

Introduction

While FePt prepared as a thin film is one of the main candidates for the next generation data storage devices, its physical properties regarding ordering dynamics were mainly investigated in the bulk. In the L10 phase the magnetocrystalline uniaxial anisotropy in the range between [1], [2] 7 J cm−3 and 80 J cm−3 is the reason for its hard ferromagnetic constitution of about 0.5 meV anisotropy per 3d-atom. FePt is therefore suitable for perpendicular magnetic data storage media applications, which offer potential for increased data storage densities with respect to the currently still common in-plane or parallel magnetization media [3], [4]. Recent research is already dealing with practical designs for hard discs based on a new technique [5], [6]. The impact on data storage devices makes a detailed knowledge of the dynamics and diffusion processes essential. Only then one can selectively design alloys that guarantee an acceptable thermal stability of the permanent magnetization in very small volumes enabling scale reduction of the memory bits in the recording media.

In the last years various experiments with bulk L10 FePt devoted to diffusion [7], [8], [9] and ordering dynamics studies [10], [11], [13] have been performed. On the contrary, the results we present here deal with the ordering mechanism of the epitaxial FePt thin film system. The L10 structure may grow in bulk or in thin film samples with three different structural orientations of the c-axis as shown in Fig. 1. Our recent Monte Carlo (MC) simulations [10], [11], [12], [13] predicted that the c-variant of a single-crystalline L10 structure with out-of-plane easy axis of thin film magnetization yielding technologically desired magnetic axis is unstable in its atomic stacking. The mono atomic planes spontaneously reorient creating a-variant domains with the easy axis of the magnetic field in the film plane. The underlying atomic-scale mechanism is as follows: some iron atoms from the surface iron layer replace platinum atoms in the layer beneath, see Fig. 2 of Kozlowski et al. [13]. The effect of spontaneous L10 c-variant → a-variant reorientation in stoichiometric FePt structure limited by free (001) iron surfaces was achieved by the Monte Carlo method implementing vacancy mechanism of atomic jumps [11], [12], [13]. Nearest neighbor (NN) and next nearest neighbor (NNN) pair-interaction energies addressing FePt were deduced from ab-initio calculations combined with a cluster-expansion procedure [14], [15].

Fig. 1

Different variants of the L10 structure (one should imagine the surface of the thin film as a horizontal plane): the c-variant with c-axis ⊥ to the surface of the thin film (a), or a-variant with c-axis in plane of the thin film (b) and (c).

Fig. 2

XRD spectra of as-prepared samples and samples annealed at 773 K, 848 K and 898 K. The spectra are shifted vertically for better visibility.

Here we present the experimental evidence of the reorientation within c-variant ordered L10 structures of FePt thin films. The rearrangements of ordered domains are caused by successive annealing at temperatures between 773 K and 898 K following order/order transformations between different L10 variants in the FePt thin film: c-variant (c), a-variant (a), and a random, disordered r-variant (r). We further report on the effect which was not observed in the MC simulations, i.e. partial recovery of the c-variant in the epitaxial thin film after long annealing times. One may consider this process as partly self-healing of the FePt film driven by the ordering free energy overpowering the surface free energy. Both processes originate from an initial disordering generated by (fast) surface diffusion [16] followed by a partial recovery of the long-range order generated by (slow) bulk diffusion. A simple kinematical model describes this phenomenon yielding activation energies for ordering and disordering.

Theory

Kinematical model

The experimental data were evaluated by the Johnson–Mehl–Avrami (JMA) model. The same model was applied already to describe L10-ordering in FePt after sputter deposition [17]. In our investigation we found that the ordering is driven by different kinds of dynamics for short and long time scales. The JMA-model was adapted to account for these differences and to fulfill various constraints characteristic for thin films:

initial and final fractions of all variants need not to be 1 or 0 (i.e. the dynamics do not operate in the whole thin film).

transformation between following variants are included in the model: c → (a,r), (a,r) → c, a → r, r → a.

different ordering dynamics account at different time scales, i.e. different Avrami exponents (AE) are applied.

The first and simplest approach contains of two different time scales, e1(t) for long times and e2(t) for short times, both with corresponding AE’s. The model consisting only of two time scales works sufficiently well for all measured data.Here t is the time, τ the characteristic time of the process and n the corresponding AE. The time dependence of the amount of c-variant is then determined by the sum of a decaying fraction d(t) and a growing fraction g(t)c0 is the saturation value for t → ∞. The model allows that only a fraction c2 reorients, while the recovering of the c-variant dominates only in a fraction c1 of the film, i.e. the starting volume of the transformation is limited because a part of the film remains c-variant. According to equation (2) the fractions for the a-variant and the random r-variant were determined taking into account that a and r can be formed only by fractions a1 and r1 from the amount which was rearranged from the c-variant and vice versa, thus a1 + r1 = 1. For the a-variant and r-variant the growing fractions g(t) and g(t) and the decaying fractions d(t) and d(t) can be determined including also the initial fractions of a and r, a0 and r0, respectively. This finally gives the total fractions of a and r:In the simplest approach the formation and loss of a and r have the same weights a1 and r1, respectively. Further, if a constant mass flow is assumed a → r or r → a for all time scales, only the rate a1 changes. To make the model flexible in describing rates of the formation or annihilation of a or r, a non-constant mass flow f between a and r was introduced (again for the simplest case of two different time scales):Note that the model can only describe net-flows, i.e. the result of a → r for short times is equivalent to the fact that the flow c → r is larger than c → a.

For short times a part (1 − f1) of the growing a-variant is forming the r-variant with corresponding fractions expressed by

For long times a part (1 − f2) of the decaying a-variant is not forming c but r

Monte Carlo simulations

The structure simulated in [11], [12], [13] was a nano-layered L10 AB-binary stoichiometric thin film of 403 cubic fcc cells with periodic boundary conditions. A single vacancy was introduced into the system by emptying one site chosen at random. Glauber dynamics with a vacancy mechanism of atomic jumps [14] have been used in the simulations. Pair-interactions of Fe- and Pt-atoms in two co-ordination shells were evaluated on the basis of “ab-initio” calculations combined with the Cluster-Expansion technique [15]. The simulated isothermal “order–order” kinetics at temperatures in equilibrium Tequ followed sudden changes of temperature from a starting temperature Tst to Tequ. Among the monitored parameters were the Bragg-Williams-type long-range order (LRO) parameter (η) dedicated to particular L10 superstructure variants, specific short-range order parameters and frequencies of particular atomic jumps. The simulated system showed a discontinuous “order–disorder” transition at a temperature T close to the experimental one. Thin layers of FePt with different variants of the L10 superstructure were simulated by removing periodic boundary conditions in direction parallel (c-variant) or perpendicular (a- and b-variant) to the c-axis. The layer thickness was varied in such a way that the vacancy concentration was constant. For thin layers 0 K → Tequ relaxations were performed and the atoms were forced to remain in the initial box. The film surfaces were composed one of Fe-atoms and one of Pt-atoms, each an interface to vacuum. No substrate and no magnetic energy were introduced.

Two processes were observed [13]: (i) homogeneous disordering i.e. generation of antisites within the volume of the layer (as found for bulk-calculations [13]); (ii) formation of L10-ordered a- and b-variant domains due to the jumps of surface Fe-atoms on Pt antisites. The resulting ordering process nucleated heterogeneously and selectively on the Fe-surface and advanced discontinuously inward the layer. The Pt-surface stayed, however, unaffected. For the simulation of very thin films of about 10 unit cells thickness, the reaction front reached the free Pt-surface which resulted in an additional energy gain from the formation of a- and b-variant domains at the Pt-surface (the energy gain was due to the transformation/reorientation of the first layer of c-variant L10 unit cells at the Fe-surface). The thin film froze in this configuration, which then is the final saturated state in thermodynamical equilibrium. Only for thick films the reaction front does not reach the free Pt-surface and the propagation depth saturates depending on the simulation temperature.

Monte Carlo and ab-initio simulations of Fe–Pt intermetallic alloys are not numerous but two recent papers should be commented. Misumi et al. [18] have studied equilibrium configurations in bulk FePt. Interesting is the derivation and application of four-body potentials on the basis of a combination of ab-initio and cluster-expansion calculations. Although explored is a lattice model, the way the potentials are calculated allows to account for off-lattice effects. However, as the direct-exchange MC algorithm is used, no information on ordering kinetics is provided. In addition, the simulations start from a disordered fcc structure, so observed is a formation of a system of L10 domains. In general, the paper focuses on the correctness of the order–disorder transition temperature resulting from the simulations.

Very important results relevant for our studies are presented by Dannenberg et al. [19]. Ab-initio energies of L10 FePt-surfaces with diverse orientations are calculated and it is shown that the (100) surface (i.e. a-variant) is energetically favourable with respect to the (001) (i.e. c-variant) surface. This is a relevant support for our experimental observation of the reorientation phenomena.

Experimental results

Samples and characterization

Experimentally this ideal structure is represented by a thin film sample of c-variant L10-ordered FePt grown on MgO. The real structure can of course never provide the same results as the Monte Carlo simulation with two free surfaces. Further, during the preparation the atoms may migrate, segregate and diffuse. Due to the presence of the substrate the films show also epitaxial strain enhancing the tetragonal distortion of the unit cell close to the substrate. The best approximation of free surface was a thin film without cover. To ensure that the reorientation effect originates genuinely from formation of a-variant domains at the surface, Pt covered reference samples were also produced. All FePt thin films were prepared by molecular beam epitaxy (MBE). The pressure was better than 1 × 10−10 mbar and the substrate temperature was 623 K during preparation. A first set of samples consisted of 57FePt(500 Å)/MgO(001) and a second identical set was covered in-situ by 20 Å of platinum. Both sets were produced simultaneously by co-evaporation of Fe (0.041 Å s−1) and Pt (0.053 Å s−1). Additionally an independent third uncapped sample set was identically grown to check the reproducibility of the result of the sample one.

For more details on the quality of the sample inclusive TEM pictures, see Laenens et al. [20]. The thickness of 500  Å is about three times larger than the forty unit cells thick layer for MC simulations. Nevertheless, the thicker sample was chosen to guarantee a relaxed, flat film of well ordered L10 structure [9], [20], [21]. The samples were characterized regarding their chemical and magnetic order using conversion electron Mössbauer spectroscopy (CEMS), X-ray diffraction (XRD), and vibrating sample magnetometer (VSM) [20]. All methods give clear evidence of an almost perfect c-variant L10-ordered FePt thin film.

The long-range order (LRO) of the samples was measured by XRD using Cu-K radiation. Spectra for the as-prepared state as well as for the samples annealed at 773 K, 848 K and 898 K are shown in Fig. 2. Well pronounced (001) and (002) reflections indicate a well ordered crystalline L10 structure for all thin film samples. The c-axis lattice constant of the as-prepared sample as calculated from the position of the Bragg peaks is cexp = 3.723(2) Å. This corresponds to the ideal value of the c-axis in the L10 FePt structure (cid = 3.7212(3) Å)[22], [23]. From the intensities of the (001) and (002) reflections we estimate an LRO parameter to be 0.77(5) for the as-prepared sample. The LRO parameter increases after first annealing steps up to values larger than 0.9. Magnetization curves (not shown) measured with the external field perpendicular to the film surface by means of VSM reproduced the values from the CEMS measurements quite well. It was, however, not possible to conclusively correlate the VSM data with the fraction of differently oriented variants. The last feature was measured with high accuracy by CEMS.

Mössbauer spectroscopy

We performed stepwise annealing in vacuum of 2 × 10−6 mbar or better at 773 K, 808 K, 848 K and 898 K (a different sample at each temperature) followed by CEMS characterization at room temperature. To double-check the stability of the specimen during annealing procedure, XRD and high resolution XRD (HRXRD) were aplied. For the CEMS measurements we used a flowing-gas detector (He–CH4) and a standard Mössbauer drive and electronics. The source was 50 mCi 57Co in Rh-matrix and the calibration was done with a foil of high-purity natural iron.

The spectrum of the as-prepared sample is shown in Fig. 3. (a). Intensities of CEMS resonances were fitted with the software including the hyperfine field distribution (RECOIL [24]). Intensities of Mössbauer lines, especially the intensity of the second and fifth line contains information about the orientation of the incident k vector of the 14.4 keV γ-quanta with respect to the vector of the magnetic hyperfine field. Detailed analysis of the CEMS spectra allows therefore to determine the fraction of different L10 variants in the structure of the FePt thin film samples: c-variant , a-variant as well as the random distribution (with a randomly oriented magnetic field vector). Enhanced intensities of the second and fifth Mössbauer lines are an evidence of the lower c-variant content and, consequently, higher a-variant (or random domains) level and vice versa. The changes in the relative intensities of 2nd and 5th line are visible comparing e.g. Mössbauer spectra in Fig. 3(a–c). A larger fraction of very broad lines (the lines are broad due to large hyperfine field distribution) in Fig. 3(b) compared to (a) or (c) are evident. We want to point out that these variants are identical crystalline L10-ordered domains in the FePt thin film, but of different orientation of the c-axis (i.e. the magnetization vector), compare Fig. 1. The reference sample of set one and two (an as-prepared specimen) was composed of 90(1)% of c-variant, 3.5(8)% a-variant and 4(1)% r-variant. A small quadrupole doublet in the center of the spectra with about 2–4% of the total spectral area results from surface oxidation during annealing, see Fig. 3. Its fraction remains constant for all annealing steps. The corresponding hyperfine parameters are shown in Table 1. The line width of 0.15 mm s−1 was measured for the as-prepared sample and confirms the high degree of order in the thin films.

Fig. 3

CEMS spectra of the specimen annealed at 773 K after annealing for (a) 0 h (as-prepared), (b) 8 h and (c) 36 h and annealed at 898 K for (d) 1.3 h and (e) 3 h. The solid lines are the fit with the RECOIL (24) software. (f) shows the hyperfine field distribution of all variants for the as-prepared state (solid lines) compared to the sample annealed at 773 K for 8 h (dashed lines).

Table 1

Typical hyperfine field parameters from the CEMS analysis. Γ gives the HWHM of the Mössbauer lines, IS the isomer-shift, QS the quadrupole-splitting, H the hyperfine magnetic field and σ the Gaussian width of the magnetic field distribution.

Site	Γ [mm s−1]	IS [mm s−1]	QS [mm s−1]	H [T]	σH [T]
c-variant	0.15(1)	0.29(1)	0.14(1)	27.3(1)	0.7(3)
a-variant	0.15(1)	0.26(1)	0.12(1)	30.0(5)	1.9(6)
Random dist.	0.15(1)	0.26(2)	0.13(3)	25.7(9)	2.5(7)

We chose samples from sample set one for thermal treatment at three annealing temperatures (773 K, 848 K and 898 K) and one sample of sample set three at a different temperature of 808 K to verify the experimental results and our model. For the uncapped samples we find for the lowest temperature of 773 K an initial very strong decrease of c-variant content (down to 62%), with corresponding increase of a-variant (15%) and random distribution r (19%). The maximally reoriented state was reached after approximately 8 h and the final saturation state after about 20 h as shown in Fig. 4. For annealing times longer than 8 h, the c-variant starts to recover with corresponding decrease of a-variant and r-variant. All fractions of the specimen annealed at 773 K are shown in Fig. 4. The samples annealed at 808 K and 898 K show a similar behavior as the one annealed at 773 K. The fractions of c, a and r saturate for times longer than 12 h at 808 K and 2 h at 898 K, respectively. A different behavior was found at 848 K. The c-variant never recovers at this temperature. Nevertheless, the fractions saturate for times longer than 10 h.

Fig. 4

Time evolution of the c-variant content of the specimens annealed at 773 K (•), 808 K (▴) and 848 K (♦). The data for annealing at 898 K is shown in the inset (•). The a-variant and random distribution of the sample annealed at 773 K are shown as (▪) and (▾), respectively. The dashed lines shows the growing fraction g(t) and decaying fraction d(t) of equation (2) for a fit to the data of the sample annealed at 773 K. All solid lines represent fits to the CEMS data.

For the sample set capped with 2 nm Pt the formation of a-variant is almost completely suppressed, as can clearly be seen in Fig. 5. According to the MC simulation this is due to the completely missing free Fe-surface where the reorientation is initialized. The suppression-effect is smaller at higher temperatures (see graph for 898 K in Fig. 5). The reason for this is the higher mobility of atoms: a random formation of a-variant domains in the topmost layers of the FePt corresponds to the formation of Fe-antisites on Pt lattice sites.

Fig. 5

Time evolution of the c-variant content of the specimens annealed at 773 K, left picture and 898 K, right picture. In each picture, (•) shows the first annealing steps for the uncapped sample set, whereas red (▪) shows the time evolution for the c-variant of the Pt-capped sample (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article).

Discussion

For all annealing temperatures 773 K, 808 K, 848 K and 898 K we find a fast decay of the c-variant – τ2 – (i.e. fast formation of a-variant and random distribution r) and a slow recovery of the c-variant – τ1 – (i.e. decay of a-variant and r-variant). Our kinematical model describes the experimental results satisfactorily. For the sample annealed at 848 K the dynamics contains an additional slowly decaying component – τ3. The origin of this behavior is different from the results at other temperatures and in two sample sets and is an objective of further investigations. We find two different AE-values for short and long times, see Fig. 6. The τ-values show Arrhenius behavior, see Fig. 7. All resulting parameters are summarized in Table 2.

Fig. 6

(a) Avrami exponents n2 for short times (▪) representing disordering and n1 for long times (•) representing ordering. The horizontal lines at n = 2 and n = 3 indicate 2D- and 3D-growth, respectively. (b) Flow rates for short times (▪) and for long times (•) indicate changes of volume fractions of different variants. The line at f = 1 indicates no flow between a and r-variant.

Fig. 7

Arrhenius plot for the reorientation dynamics. (▪) shows the logarithm of τ2 corresponding to short time dynamics (disordering or formation of a-variant), (•) shows the logarithm of τ1 corresponding to long time dynamics (ordering, i.e. formation of c-variant).

Table 2

Fitted parameters of the JMA-model.

T [K]	n1	n2	τ1 [103 s]	τ2 [103 s]	f1	f2
773	3.2(3)	1.6(2)	44(1)	16.2(8)	0.86(5)	1.2(1)
808	3.2(1)	1.6(1)	20(2)	5.7(5)	1.5(1)	2.5(3)
848	3.2(6)	1.9(7)	20(2)	5(1)	0.9(2)	0.2(1)
898	3.1(9)	1.0(7)	7.0(4)	0.7(4)	0.8(1)	0.1(2)

The results we achieved by means of CEMS measurements show to our best knowledge the reorientation effect described in our kinematical model. However, one has to carefully take into account other effects possibly causing similar changes in the CEMS spectra. First of all oxidation effects have to be investigated carefully. As the uncapped FePt thin films were exposed to air between thermal-treatment steps, oxidation is not unrealistic. The CEMS technique is however very sensitive to oxidation effects due to characteristic changes of the hyperfine field values. We analyzed the spectra taking care on the arising Fe–O compounds, but only paramagnetic quadrupole doublet with a total spectral fraction of about 2–4% arising after the first annealing step could be identified. Stability against oxidation was already proven in [9]. Further, formation of anti-phase boundaries (APB’s) could cause a change in the CEMS patterns. The formation of a-variant domains in an initially purely c-variant sample is only possible by formation of many defects. Nevertheless, we can clearly attribute the arising sextets to a-variant, c-variant and disordered (i.e. small, randomly oriented domains) areas of the sample. APB’s would certainly manifest mainly in the disordered region and only a minor fraction will contribute to the interface of a- and c-variant domains. We therefore conlude that the main contribution to our CEMS results originates from a structural reorientation (i.e. order/order transformation) of L10-ordered domains within the FePt thin films.

Avrami exponents

The AE-values are temperature independent within error bars as shown in Fig. 6. We find as average values for long and short ordering timesrespectively. The AE can be interpreted as a measure of the growth dimensionality of domains [25]. Following this approach, n2 describes the dynamics at the start of the reorientation process where the initial formation of a and r takes place. n2 is close to the value of two. The growth can be interpreted as 2D-like with a constant seed-rate (e.g. the first monolayer is already completely a-variant) or 1D with a growing number of seeds (e.g. in small channels into the depth of the thin film while new a-variant domains are forming within the first monolayer). Seeds then would be Fe-antisites in the second-topmost monolayer. Thinking of a pure topmost Fe-layer in a real experimental setup with samples grown by MBE at elevated substrate temperatures is clearly unrealistic. Nevertheless, we have an indirect proof for the prevailing of an Fe-terminated surface. The preparation and heat treatment of the Pt-capped sample set two was in the same way as for the uncapped sample sets one and three. The only possible difference explaining the complete missing of an ordering effect in sample set two, while at the same time one finds the well pronounced effect in sample set one and the reference sample set three, is at least enhanced concentration of surface terminating Fe-atoms. While sample set two has definitely no surface Fe-atoms due to 20 Å platinum capping, statistically in the sample sets one and three due to the preparation at high temperature and therefore enhanced diffusivities near the surface, see e.g. [16], a reasonable amount of surface atoms may be Fe-atoms. These atoms are available as seeds to start the reorientation processes.

n1 is close to the value of three for all annealing temperatures. In the JMA-model this corresponds to 3D-like growth of the c-variant. Thus the c-variant is either formed in the whole volume of a-variant (and r-variant) and with a random distribution and a constant number of seeds (3D-growth) or is formed at interfaces between c-variant and a-variant (and r-variant) with a non-constant number of seeds (2D-growth). The values for n1 = 3.17(5) and n2 = 1.5(4) reproduce the MC result, that the growth starts from the surface and then proceed into the depth of the thin film. They give information about the dimensions of growth and seed-rate. This result answers an open question arising from other MC simulations, which suggested ordering mechanisms in FePt in terms of surface Pt-precipitation [26], or in terms of surface induced disorder [27].

Flow rates

The values describing the flow rates are shown in Table 2 and in Fig. 6(b). We find for low temperatures f2 < 1 (i.e. a → r for the initial stage of the ordering dynamics or for annealing times short compared to the saturation time) and f1 > 1 (i.e. r → a for long annealing times comparable to the saturation time). This corresponds to the fact that for short times the formation of the random distribution is dominating over the formation of a-variant (disordering generated by the fast surface diffusion), while for long times the formation of a-variant dominates over the formation of the r-variant (ordering generated by slow bulk diffusion). For the two highest temperatures we find f < 1 (i = 1, 2) corresponding to an increase of disorder, since the thermal formation of defects is supported. This means, r grows always at the cost of a, disorder increases.

The cause for the formation of r is that in the epitaxial film the a-variant has to form while having a different lattice constant (a = 3.8504(8), c = 3.7212(3)). This means that some kind of defects have to be introduced to allow the formation of a-variant in a region of the film which was c-variant before. Further, the a-variant can only be formed by disordering the c-variant L10 structure first via the formation of Fe-antisites on the Pt sublattice. This makes it likely that a front of disordered L10 is located between a- and c-variant and contain many defects. The overall behavior is consistent with the results of epitaxial ordering of FePt for different ordering stages and annealing temperatures [17], [28], [29].

Activation energies

The time constants τ1 and τ2, shown in Fig. 7, follow Arrhenius behavior. From the slopes of the Arrhenius plot we calculated activation energies for the dynamical processes of

At the beginning of the annealing procedure (short times, Q2) the rearrangement takes place via the formation of Fe-antisites on the Pt sublattice. The atomic mechanism, therefore, is a net migration (diffusion) of Fe-atoms along the c-axis of the thin film. We can compare our activation energy to similar results achieved by residual resistivity measurements for order–order dynamics in bulk samples and thin films in the same temperature range which are 1.5(3) eV and 1.8(2) eV, respectively [10], [30]. Moreover, our former studies on iron diffusion along the c-axis in L10-ordered FePt thin films by nuclear resonant scattering [9] give an activation energy of 1.65(30) eV. The values again are the same within the error bars confirm that the underlying dynamical process is disordering. For long annealing times (Q1) the formation of the c-variant corresponds to ordering of the epitaxial L10 structure. We can directly compare our results with the activation energies found for ordering of disordered sputter-deposited FePt thin films [17] of 0.9(2) eV via XRD and 0.6(4) eV via Mössbauer spectroscopy. The results once more correspond well with our value.

The main result of the CEMS investigations, namely the reorientation behavior reproduces the MC simulation result. Nevertheless, the unique and stunning result of the recovering of the c-variant – i.e. a reversal of the reorientation in the epitaxial FePt thin film has been not shown in the MC simulation. The discrepancy between the simulation and the experiment is apparently due to the fact that the real film is a complex system whose characteristics are determined by tetragonal distortion, epitaxy, magnetic stray fields, non-ideal surface termination and the interface to the substrate. All these features have not been accounted for in the MC simulations [13]. However, the main result of the structural reorientation is in perfect agreement with MC simulations and with the CEMS experiment.

Conclusion

Simulations predict that a reorientation of the c-variant to a- and b-variants in L10-ordered FePt occur. The reorientation effect eventually reverts (recovering of the c-variant) and then saturates for long annealing times. We have experimentally verified that the structural rearrangement in FePt thin films also gives rise to the rearrangement of the magnetic field. We found two dynamical mechanisms (ordering and disordering) working simultaneously but with different time constants, dimensions of growth and activation energies. The superposition of both is the origin of the experimentally found kinetic behavior. The activation energies are comparable to energies for ordering and disordering in L10 FePt found in other measurements. We find that:

with respect to the degree of long-range order, the thin film evolves away from the configurational equilibrium because of the dynamical effect originating from the surface and epitaxial interface (disordering);

the reorientation overshoots the equilibrium LRO state and the system partially restores its long-range order. Both mechanisms work at the same time but with different time constants and activation energies.

The reason for the surprising recovery of the c-variant is that the ordering process has a lower activation energy than the disordering process and therefore dominates, an effect which is further enhanced due to epitaxy. The impact of the ordering component is only visible for long annealing times as its time constant is larger than the one of disordering component. The experimentally found behavior is consistent with the results of highly ordered epitaxial FePt layer annealed at higher temperatures. Finally, we have demonstrated that a capping layer suppresses the reorientation effect. This is due to the missing free Fe-surface which initiates the reorientation. The suppression-effect is smaller at higher temperatures due to the higher mobility of atoms.