Local structure, nucleation sites and crystallization behavior and their effects on magnetic properties of Fe81SixB10P8−xCu1 (x = 0~8)

In this work, an attempt has been made to reveal critical factors dominating the crystallization and soft magnetic properties of Fe81SixB10P8−xCu1 (x = 0, 2, 4, 6 and 8) alloys. Both melt spun and annealed alloys are characterized by differential scanning calorimetry, X-ray diffractometry, Mössbauer spectroscopy, transmission electron microscopy, positron annihilation lifetime spectroscopy and magnetometry. The changes in magnetic interaction between Fe atoms and chemical homogeneity can well explain the variation of magnetic properties of Fe81SixB10P8−xCu1 amorphous alloys. The density of nucleation sites in the amorphous precursors decreases in the substitution of P by Si. Meanwhile, the precipitated nanograins gradually coarsen, but the inhibiting effect of P on grain growth diminishes causing the increase of the crystallinity. Moreover, various site occupancies of Si are observed in the nanocrystallites and the Si occupancy in bcc Fe decreases the average magnetic moment of nanograins. Without sacrificing amorphous forming ability, we can obtain FeSiBPCu nanocrystalline alloy with excellent soft magnetic properties by optimizing the content of Si and P in the amorphous precursors.

Fe-based nanocrystalline soft magnetic alloys with the amorphous/crystalline composite structures have become a hot topic in both research and application. These alloys usually possess outstanding magnetic properties involving high saturation magnetic flux density (B S ) and low coercivity (H C ) [1][2][3][4] . Current state of the art nanocrystalline alloys mainly include FeSiBNbCu 5 , FeSiBPCu 6 , FeMBCu (M = Zr, Nb, Hf) 7 and their derivatives which contain a handful of early transition metal elements to promote their crystallization 8 . It is noteworthy that soft magnetic properties of these alloys are usually variable even if the atomic percent of magnetic Fe atoms keeps the same. For example, Fe 82.65 Cu 1.35 Si 2 B 14 nanocrystalline alloy shows an H C of 6.5 A/m and a B S of 1.84 T, while the H C and B S of Fe 82.65 Cu 1. 35 Si 5 B 11 are 60 A/m and 1.81 T, respectively 9 . Magnetic properties of nanocrystalline alloys are closely associated with the amount, size and chemical composition of nanograins embedded in the amorphous matrix.
Every kind of metalloid and transition metal element plays an essential role on the nanocrystallization and only with appropriate content can the magnetic properties be improved. It has been proven that moderate addition of Cu can provide nucleation sites for α-Fe and thus refines the primary nanocrystallites due to the formation of Cu clusters 10,11 . Moreover, the differences in the kind and content of solute elements usually result in various structures and chemical compositions of nanograins. For instance, α-Fe phase with bcc structure precipitates out in the Fe 83.3 Si 4 B 8 P 4 Cu 0.7 alloy 12 , but Fe 3 Si phase with DO 3 structure precipitates out in the Fe 73.5 Si 13.5 B 9 Nb 3 Cu 1 alloy 13 . The composition of precipitated nanograins varies with the content of Si even in the same system, such as in FeSiBCu 14 and FeSiBNbCu alloys 15 , respectively. In addition, the co-addition of Cu and P can further refine the nanograins, which results from the interaction between solute atoms forming CuP clusters 16 . The crystallization process tightly contacts with the microstructure of amorphous precursors including short range order, free volume, chemical homogeneity and particularly nucleation sites. Thus comprehensively understanding the transformation of amorphous precursors during the crystallization appears to be significantly important for advanced materials design.
For those nanocrystalline alloys with heterogeneous nucleation, various methods have been proposed to increase nucleation sites for the refinement of nanograins, such as the addition of Cu into the FeZrNbB alloy 17 .
When the amorphous precursors contain insoluble atoms or atom pairs, such as Cu, Nb and Cu-P, they tend to precipitate out and agglomerate to form insoluble clusters at the early stage of annealing process which can serve as nucleation sites for nanograins. However, little is known about the real content of nucleation sites during the crystallization which is usually reflected by the size of nanograins. It is probable to be inaccurate in some cases because some solute atoms such as P can also restrain the growth of nanograins 18 . The difficulty lies in the facts that nucleation sites are too small to be characterized and that one can hardly discern which clusters can effectively serve as nucleation sites. But from another point of view, crystallites precipitate out on nucleation sites at the very beginning of crystallization with the formation of nanovoids at grain boundaries 19 . Positron annihilation lifetime spectroscopy (PALS) is an efficient method to provide precise information about characteristic positron annihilation sites in nanocrystalline alloys, such as interstitial holes, vacancies and grain boundaries 20,21 . This technology can be used to characterize the formation and growth of crystallites, which can effectively reflect the density of nucleation sites during the crystallization. Meanwhile, Mössbauer spectroscopy can precisely identify the local structure around atomic nucleus for its high energy resolution, with which the information about short-range order in the amorphous matrix and chemical compositions of the nanograins can be obtained 22 .
In this work, we analyze the local structure differences in both melt spun and annealed Fe 81 Si x B 10 P 8−x Cu 1 (x = 0, 2, 4, 6 and 8) alloys to reveal the critical factors dominating the crystallization. Specifically, we focus on the microstructure of amorphous precursors, nucleation sites at the beginning of crystallization and crystal phases after the nanocrystallization. Furthermore, we are aiming at understanding the effects of local structure, nucleation sites and crystallization behavior of the amorphous precursors on magnetic properties of nanocrystalline alloys.

Results and Discussion
Thermal properties. As shown in Fig. 1, differential scanning calorimetry (DSC) curves are collected to investigate the thermal dynamic characteristics of Fe 81 Si x B 10 P 8−x Cu 1 (x = 0, 2, 4, 6 and 8) alloys. All these alloys have two conspicuous and separated exothermic peaks marked as T X1 and T X2 , which indicates that the crystallization of Fe 81 Si x B 10 P 8−x Cu 1 alloys occurs in two stages. For Fe 81 Si x B 10 P 8−x Cu 1 alloys with x = 0, 2, 4, 6 and 8, T X1 is 713, 688, 697, 701 and 704 K, respectively. It is worth noting that 2at. % of Si significantly advances the primary crystallization but postpones the secondary crystallization, and then both T X1 and T X2 increase gradually with Si further substituting for P. The temperature interval (ΔT X = T X2 − T X1 ) between the two peaks shows a maximum of about 117 K for Fe 81 Si 4 B 10 P 4 Cu 1 alloy. The smaller ΔT X means the more strict annealing treatment during the crystallization process, especially for Fe 81 B 10 P 8 Cu 1 alloy in which the secondary crystallization practically tends to occur. Thus, appropriate Si superseding P in Fe 81 Si x B 10 P 8−x Cu 1 alloys provides an advantage for the nanocrystallization.
Local structure in amorphous precursors and crystal phases. Figure 2 shows the X-ray diffractometry (XRD) patterns for the melt spun and annealed Fe 81 Si x B 10 P 8−x Cu 1 (x = 0, 2, 4, 6 and 8) alloys. Each diffraction pattern of the melt spun ribbons in Fig. 2(a) exhibiting a typical broad halo without any appreciable crystalline peak indicates the formation of amorphous structure. According to the results of DSC, all Fe 81 Si x B 10 P 8−x alloy ribbons are annealed at their individual T X1 for 5 min. As shown in Fig. 2(b), a weak crystalline peak located at 2θ = 45° corresponding to bcc Fe suggests the beginning of primary crystallization. Figure 2(c) presents the diffraction patterns of the Fe 81 Si x B 10 P 8−x Cu 1 nanocrystalline alloys obtained by annealing their amorphous precursors at 743 K for 5 minutes. In each pattern, there appear three distinct diffraction peaks corresponding to bcc Fe phase. The intensity of these peaks is enhanced as Si content x increases, which implies that the crystallinity of Fe 81 Si x B 10 P 8−x Cu 1 alloys increases with P gradually replaced by Si. Meanwhile, according to the Scherrer equation, the average mean grain size of the α-Fe phase is estimated to increase.
A further study with Mössbauer spectroscopy provides more detailed information about the microstructure of as-quenched and nanocrystalline Fe 81 Si x B 10 P 8−x Cu 1 (x = 0, 2, 4, 6 and 8) alloys. As shown in Fig. 3(a), the wide and asymmetrical sextets demonstrate the amorphous nature of all the melt spun alloys 23 . The hyperfine parameters of fitted Mössbauer spectra are given in Table 1. Isomer shift (IS) refers to the observed shift of the resonance spectrum, quadrupole splitting (QS) results from the asymmetrical distribution of charge around atomic nucleus, and isomer shift change (DTI) is a coupling parameter that ensures the observed asymmetry of the experimental spectrum. The average magnetic hyperfine field (B hf,a ) is a parameter to describe the average hyperfine interaction and proportional to the spin-exchange interaction between magnetic atoms in each sample. It is noted that both Fe 81 B 10 P 8 Cu 1 and Fe 81 Si 4 B 10 P 4 Cu 1 show relatively high B hf,a compared with the other amorphous alloys. As reported early, the strengthened magnetic interaction can be ascribed to an increased degree of order in the topological structure 24 .
Hyperfine field distributions obtained from the Mössbauer spectra of melt spun Fe 81 Si x B 10 P 8−x Cu 1 alloys are depicted in Fig. 3(b). Magnetic hyperfine field distributions are usually decomposed into several components adopting Gaussian distributions to investigate the chemical short range orders in the structural relaxation or the crystal phases in the crystallization [25][26][27] . However, it is less accurate for melt spun metallic glasses where excess free volume and residual stress distort the coordination environment of Fe atoms. Nevertheless, all the hyperfine field distributions distinctly separate into two regions in Fig. 3(b) which suggests the coexistence of two distinguished zones in the amorphous structure of melt spun ribbons, namely the low field region (8 T to 15T) with weakened magnetic interactions and the high field region (15 T to 35 T) with enhanced magnetic interactions 28 . The low field region indicates Fe-deficient zones with plentiful B atoms and a little Cu atoms occupying in the neighbor shell of Fe atoms, while the high field region implies Fe enriched ones 29,30 . Therefore, the relative area of the low field region in Fig. 3(b) can be utilized to evaluate the chemical homogeneity of the amorphous structure. The maximum area ratio of the low field region in Fe 81 Si 2 B 10 P 6 Cu 1 alloy manifests a high degree of inhomogeneity in the amorphous structure which will promote the crystallization in the annealing, while the minimum value of that in Fe 81 B 10 P 8 Cu 1 alloy leads to a high degree of homogeneity and a high T x1 , which is consistent with the results from DSC. Figure 4 displays Mössbauer spectra and the corresponding hyperfine field distributions of Fe 81 Si x B 10 P 8−x Cu 1 nanocrystalline alloys obtained by annealing their amorphous precursors at 743 K for 5 min. The split sextets superimposed upon the broadened spectra in Fig. 4(a) and the emergence of split peaks at about 33 T corresponding to bcc Fe in Fig. 4(b) characterize the microstructure with nanograins embedded in the residual  Similarly, A7 and A6 configuration mean the bcc Fe structure with seven and six nearest-neighbor Fe atoms, respectively. Therefore, to further explore the specific site occupancy in the nanocrystallites of Fe 81 Si x B 10 P 8−x Cu 1 alloys, the hyperfine field distributions are precisely decomposed into several components by Gaussian distributions. The gray histograms in Fig. 4(b) represent the actual hyperfine field distributions, while the dotted black lines represent the fitting results obtained from the corresponding colored components. The excellent match between them confirms that the Gaussian fitting results can provide accurate structural information of the FeSiBPCu nanocrystalline alloys. Peaks at hyperfine fields of about 33 T, 31.5 T and 28.5 T can be attributed to A8, A7 and A6 configurations, respectively, as suggested by previous studies 14,33 . The other peaks in Fig. 4(b) can be assigned to Fe atoms with various coordination surroundings in the residual amorphous matrix. It is noteworthy that only nanograins with A8 configuration precipitate out in the Fe 81 B 10 P 8 Cu 1 alloy because of the lack of Si, but for the rest alloys containing Si, both A8 and A7 configurations appear in the crystal phases resulting in the broadening or splitting of peaks. When x reaches 6, all the three configurations coexist in the bcc Fe crystal phases. The area ratio of each sub-peak and the sum of those corresponding to bcc Fe are shown in Table 2. With P gradually replaced by Si in the Fe 81 Si x B 10 P 8−x Cu 1 alloys, Si atoms tend to occupy more nearest sites of Fe atoms in bcc Fe structure, which results in the nanocrystallites diversified and the crystallinity of alloy ribbons increasing. It can be attributed to two aspects. On one hand, Si is preferentially spread into bcc nanocrystallites compared with amorphous matrix during the primary crystallization 14,32 . On the other hand, the decrease of P in the amorphous matrix weakens its inhibiting effect on grain growth 34 .   Table 2. Area ratios of sub-peaks corresponding to A8, A7, A6 and bcc Fe in hyperfine field distributions of nanocrystalline Fe 81 Si x B 10 P 8−x Cu 1 (x = 0~ 8) ribbons.
As shown in Fig. 5, the bright field images of transmission electron microscopy (TEM) and the corresponding selected area electron diffraction (SAED) patterns prove the amorphous/nanocrystalline composite structures of the 743 K annealed Fe 81 B 10 P 8 Cu 1 , Fe 81 Si 4 B 10 P 4 Cu 1 and Fe 81 Si 8 B 10 Cu 1 ribbons, and the average grain sizes are 20 nm, 25 nm and 36 nm, respectively. It suggests that, accompanied with the crystallinity increasing, the nanograins coarsen with P replaced by Si, which agrees well with the XRD observation. It is reported that CuP clusters have the ability of reducing the atomic migration and pining down the growth of α-Fe 35 . Therefore, by replacing P with Si in the Fe 81 Si x B 10 P 8−x Cu 1 alloys, except the inhibiting effect of P on grain growth diminishing, the density of CuP clusters may also decrease which is expected to cause the increase of grain size.  Fig. 6. Due to the formation of various free volume defects during the rapid solidification, positron lifetime spectra from all the samples can be best fitted to three lifetime components (τ 1 , τ 2 and τ 3 ) representing three different vacancy defects whose respective intensities are defined as I 1 , I 2 and I 3 . The relative errors of all the components and their intensities remain in the limit of ±0.5 %. The shortest lifetime τ 1 can be ascribed to the positron lifetime in interstitial holes in the polyhedral packing model which best describes several metallic glasses 36 . The intermediate lifetime τ 2 is related to the positron annihilation in nanovoids in the amorphous matrix and at grain boundaries formed during the annealing 37 . The longest lifetime τ 3 with values of more than 3 ns and intensities of less than 3 % is associated with the ortho-positronium annihilation occurring at the surface between the source and samples, as well as the internal surfaces between the ribbon sheets in the sandwiched samples 38 , and thus it is not given here.
As shown in Fig. 6(a), for the melt spun alloy ribbons, τ 1 values of about 0.127 ns are shorter than 0.175 ns in the pure iron containing crystal mono-vacancy defects 20 , while τ 2 values ranging from 0.513 ns to 0.537 ns indicate the free volume in nanovoids equivalent to at least five atom vacancy clusters in the amorphous structures 39 . A. P. Srivastava et al. 40 and L. Z. Lu et al. 41   When Fe 81 Si x B 10 P 8−x Cu 1 alloys are annealed at their primary crystallization temperature (T X1 ) for 5 min, τ 1 and τ 2 values in Fig. 6(b) decrease slightly compared with those in Fig. 6(a), which means the defect sizes in the amorphous matrix reduce due to the structure compaction during the annealing. I 2 associated with the content of nanovoids increases drastically after the annealing, which can be attributed to the formation of numerous crystallites. Therefore, the obvious downward trend of I 2 in Fig. 6(b) illustrates that the content of crystallites at the beginning of crystallization declines with P gradual replaced by Si. It is reported that CuP clusters (Cu 3 P like) separate from the amorphous matrix during the annealing and serve as the heterogeneous nucleation sites for crystallites in FeSiBPCu amorphous alloys 42 . More crystallites will precipitate out with more nucleation sites at the very beginning of crystallization, thus the trend of I 2 in Fig. 6(b) reflects the decrease in the density of heterogeneous nucleation sites in Fe 81 Si x B 10 P 8−x Cu 1 alloys, which can be attributed to the reduction of beneficial CuP clusters in the amorphous matrix due to the amount of P decreasing. Figure 6(c) shows τ 1 , τ 2 and their corresponding intensities of the Fe 81 Si x B 10 P 8−x Cu 1 nanocrystalline alloys. The value of τ 1 still has no obvious change after the crystallization, while τ 2 increases visually because of the grain growth. The upward trend of I 2 in Fig. 6(c) is ascribed to the increase of grain boundaries versus x, which suggests that the crystallinity increases with P replaced by Si in the Fe 81 Si x B 10 P 8−x Cu 1 nanocrystalline alloys. It agrees well with the results of XRD and Mössbauer spectroscopy.
Magnetic properties. Figure 7 shows B-H loops and Si content dependence of B S and H C of the Fe 81 Si x B 10 P 8−x Cu 1 (x = 0, 2, 4, 6 and 8) melt spun alloys. It can be found that both Fe 81 B 10 P 8 Cu 1 and Fe 81 Si 4 B 10 P 4 Cu 1 melt spun alloys exhibit relatively high B S , which is consistent with the result of B hf,a . On the other hand, H C initially increases to a maximum in the Fe 81 Si 2 B 10 P 6 Cu 1 alloy and subsequently decreases, which can be attributed to the variation in magnetic anisotropy associated with the homogeneity of the amorphous matrix observed by Mössbauer spectra. Figure 8 shows where R is the crystallinity, B SC and B SA are the saturation magnetic flux densities of the crystalline and amorphous phases, respectively. Therefore, B S strongly depends on the crystallinity R because B SC is larger than B SA . On Si substituting P, due to the replacement of Fe by nonmagnetic Si in bcc structure, the average magnetic moment of nanocrystallites inevitably decreases. However, the increase of crystallinity dominates with Si substituting for P in the Fe 81 Si x B 10 P 8−x Cu 1 alloys, which results in the improvement of B S from 1.67 T to 1.75 T. As H C ∝ D 6 (D is the average grain size) 44 , the increase of H C from 9.13 A/m to 36.92 A/m with x increasing can be attributed to the grain growth due to the inhibiting effect of P and CuP clusters on grain growth diminishing. The Fe 81 Si 4 B 10 P 4 Cu 1 alloy, whether melt spun or nanocrystallized, exhibits simultaneously high B S and low H C compared with the others. It implies that the appropriate content ratio of Si and P in the Fe 81 Si x B 10 P 8−x Cu 1 alloys can not only increase topological structure and homogenize the amorphous matrix, but also provide adequate heterogeneous nucleation sites and promote the crystallinity, all of which contribute to the enhancement of magnetic properties.

Conclusions
The microstructure of Fe 81 Si x B 10 P 8−x Cu 1 (x = 0, 2, 4, 6 and 8) alloys and their crystallization mechanism are investigated by Mössbauer spectroscopy and positron annihilation technique. The topological structure and chemical homogeneity vary a lot even if the Fe content remains the same. Meanwhile, the reduction of crystallites at the early stage of crystallization suggests that the density of heterogeneous nucleation sites decreases with the increase of x during the annealing, which can reflect the variation in the content of beneficial CuP clusters in the amorphous matrix. Although the grains gradually coarsen with Si substituting for P, the inhibiting effect of CuP clusters and P atoms on the grain growth diminishes leading to the increase of crystallinity. Furthermore, Si tends to occupy lattice positions in bcc Fe, which results in the average magnetic moment of nanograins decreasing, but its contribution to B S seems to be far less than that from the increase of the crystallinity. Tuning the content of Si and P appropriately in the FeSiBPCu alloys can not only improve the topological structure to strengthen the magnetic interaction in the amorphous matrix and the chemical homogeneity to decrease the magnetic anisotropy, but also optimizes effective CuP clusters during the crystallization to refine nanograins and finally promotes the crystallinity, which is an efficient method to improve synthetical magnetic properties without sacrificing the glass-forming ability.

Methods
The Fe 81 Si x B 10 P 8−x Cu 1 (x = 0, 2, 4, 6 and 8) alloy ingots were prepared by induction melting mixtures of industrial raw materials Fe (99.90%), Cu (99.99%), Si-Fe (Si: 99.59%, Fe: 0.27%), P-Fe (P: 26.11%, Fe: 73.80%), B-Fe (B: 17%, Fe: 82.90%) in a Ti-deoxidant argon atmosphere. To ensure the homogeneity of chemical components, they were remelted four times with electromagnetic mixing. Amorphous ribbons with a cross section of about 0.02 × 2.5 mm 2 were then produced by a melt-spinning technique on a single-roller copper-wheel under argon atmosphere with a surface velocity of about 40 m/s. Thermal properties of melt-spun ribbons were measured by DSC at a heating rate of 20 K/min under high purity argon flow. To investigate the density of heterogeneous nucleation sites at the beginning of primary crystallization, the Fe 81 Si x B 10 P 8−x Cu 1 (x = 0, 2, 4, 6 and 8) amorphous ribbons sandwiched by two quartz plates were sealed in a high vacuum furnace and isothermally annealed at their individual primary crystallization temperature T x1 obtained from DSC curves for 5 min by furnace cooling. Nanocrystalline ribbons were obtained by annealing the melt spun alloy ribbons at 743 K for 5 min. All the annealing is carried out with a heating rate of 1 K/s. The microstructure of melt spun and annealed ribbons was characterized using both XRD with Cu Kα radiation (λ = 0.154056 nm) and room-temperature Mössbauer spectroscopy with 57 Co as γ-ray source. Velocity calibration of Mössbauer spectroscopy was accomplished by an α-Fe foil of 25 μm thickness. With assuming a distribution of hyperfine fields, Mössbauer spectra were analyzed by a least-square fitting procedure with NORMOS program 45 . TEM specimens were prepared adopting ion thinning at both sides of ribbons and then investigated by Tecnai 12 at 120 kV. PALS was employed to obtain detailed information about the free volume in melt spun and annealed ribbons. Specimens for PALS measurement were composed of two stacks of 10 layers of the sample ribbon with the 22 Na source sandwiched between them to ensure the annihilation of positrons within the volume of samples. Positron annihilation lifetime spectra had 10 6 counts and the time resolution of the spectrometer was 230 ps. Each positron lifetime was fitted in three components by using LT software version 9 46 . H C and B S of samples were measured by DC B-H loop tracer.