Held Zeng Fept Thinfilms Jap 2004

JOURNAL OF APPLIED PHYSICS

VOLUME 95, NUMBER 3

1 FEBRUARY 2004

Magnetics of ultrathin FePt nanoparticle films G. A. Held,a) Hao Zeng, and Shouheng Sun IBM TJ Watson Research Center, PO Box 218, Yorktown Heights, New York 10598

共Received 3 September 2003; accepted 27 October 2003兲 We present magnetization data for polymer-mediated, self-assembled films of 4 nm FePt nanoparticles with thicknesses between one and four nanoparticle layers. As synthesized and deposited, the assemblies contain weakly magnetic material whose volume is equivalent to an outer 0.5 nm shell of the nanoparticles. During a 580 °C anneal, this material is incorporated into the magnetic domains of the nanoparticles. The fraction of nanoparticles transformed by annealing into ferromagnetic L1 0 FePt, the structure essential for magnetic storage applications, is found to vary with sample thickness; samples thinner than four nanoparticle layers show no significant fraction of ferromagnetic 共at 300 K兲 L1 0 structured nanoparticles under these annealing conditions, while in films comprised of four nanoparticle layers, less than half of the particles are ferromagnetic at 300 K. Possible causes of these observed results are discussed. © 2004 American Institute of Physics. 关DOI: 10.1063/1.1635669兴

I. INTRODUCTION

have carried out detailed magnetization measurements on assemblies of one, two, three, and four monolayers of FePt particles, both before and after annealing at 580 °C. This temperature was chosen because annealing at higher temperatures is known to result in significant nanoparticle aggregation, while annealing at lower temperatures results in reduced atomic ordering and magnetic coercivity.2,7 Prior to annealing, we find that the FePt/polymer assemblies contain particles which are superparamagnetic at room temperature, as well as a significant amount of material which is not magnetically saturated at applied fields of 70 kOe. This lack of saturation could result from a weakly magnetic outer shell, said weakness being the result of spin canting8 as well as interactions with surface ligands. Following a 30 min anneal, the lack of saturation largely disappears and the principle magnetic constituents of the assembly are superparamagnetic nanoparticles. Further, we find that the fraction of nanoparticles which exhibit hysteresis at 300 K does indeed vary with sample thickness. For one, two, and three nanoparticle layer samples, this fraction is zero, while for four layer samples it is less than one-half.

For the areal density of magnetic recording media to continue increasing at its current rate, it will be necessary to develop magnetic materials which can support increasingly smaller bits.1 Media comprised of highly anisotropic FePt nanoparticles represent one potential candidate for ultra-high density magnetic storage.2 The synthesis of FePt nanoparticles, which exhibit magnetic hysteresis at room temperature, has been reported,2,3 as has the self-assembly of these particles into 120 nm films onto which 共following an anneal兲 magnetic data could be written and read back at moderate linear densities 共5000 flux changes/mm兲. In principle, it should be possible to store data on assemblies of exchangedecoupled FePt nanoparticles at significantly higher densities, provided that the easy axes of the nanopaticles can be ordered4 and assemblies can be made which are only a few nanoparticle layers thick.1 With this goal in mind, a technique whereby a monolayer of nanoparticles is bound to a thin polymer film has been developed.5 This technique enables the depositon of FePt nanoparticles, one monolayer at a time, over large areas. The FePt nanoparticles, as synthesized and deposited, have a chemically disordered face centered cubic 共fcc兲 crystalline structure and are superparamagnetic6 at 300 K 共and, thus, show no magnetic coercivity兲. Only after being annealed do they exhibit an ordered, face-centered tetragonal 共fct兲 L1 0 crystal structure, a phase that is magnetically hard at room temperature and viable as storage media. However, magnetic studies of annealed assemblies have revealed a decrease in coercivity with decreasing film thickness.7 This has significant implications for the viability of these materials as magnetic media—the presence of nonferromagnetic nanoparticles in such an assembly would directly reduce the maximum attainable bit density. To address the issue of thickness dependence of the magnetic properties of ultra-thin nanoparticle assemblies, we

II. EXPERIMENT

We synthesized 4 nm diameter Fe58Pt42 particles by wet chemical processing, using platinum acetylacetonate and iron chloride as precursors and following published protocols.7 Following this synthesis, the Fe58Pt42 particles were deposited onto a substrate using established methods:5,7 A naturally oxidized and double-side-polished Si substrate was immersed in a chloroform solution of polyethylenimine 共PEI, ⬃20 mg/ml兲 for about 30 s, withdrawn from the solution and dipped into ethanol to wash off the excess PEI, and dried. The PEI functionalized substrate was immersed into the hexane dispersion of FePt nanoparticles 共⬃10 mg/ml兲 for 30 s, withdrawn from the dispersion, rinsed with hexane, and dried. This yielded a monolayer of an FePt nanoparticle/PEI

a兲

Electronic mail: [email protected]

0021-8979/2004/95(3)/1481/4/$22.00

1481

© 2004 American Institute of Physics

Downloaded 08 Mar 2004 to 129.107.14.178. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

1482

J. Appl. Phys., Vol. 95, No. 3, 1 February 2004

Held, Zeng, and Sun

which are magnetically saturated above ⬃20 kOe. RESULTS AND DISCUSSION

FIG. 1. 共a兲 Magnetic moment/area vs applied field for unannealed, polymermediated three layer thick FePt nanoparticle assembly at 10 K. Solid line is diamagnetic contribution of Si substrate; dashed line is best fit to slope of the data for 兩 H applied兩 ⬎20 kOe. 共b兲 Magnetization signal 共a兲 with contribution from Si substrate 共solid line兲 subtracted. 共c兲 Magnetization signal 共a兲 with linear fit to high field data 共dotted line兲 subtracted. Solid line is best fit of data to Langevin function 共see text兲. 共d兲, 共e兲, 共f兲 are equivalent to 共a兲, 共b兲, 共c兲 but with data taken at 300 K. Data taken at increasing magnetic fields shown as open squares and data taken at decreasing fields shown as solid squares.

assembly. By repeating the procedure, multilayer assemblies of FePt nanoparticles with thicknesses of two, three, and four nanoparticle layers were readily prepared. Magnetization measurements were taken with a Quantum Design MPMS Squid magnetometer in reciprocating sample measurement mode. The sample substrates were lodged into the center of a straw, with the plane of the sample parallel to the direction of the applied field. Figures 1共a兲 and 1共d兲 show magnetization data taken at 10 and 300 K, respectively, from a three nanoparticle thick unannealed FePt/polymer assembly. At most values of the applied field, the signal appears dominated by the diamagnetic response of the Si substrate. The solid lines shown in the figures correspond to the diamagnetic signal, as determined by measuring plain Si substrates, whereas the dotted lines are best fits to the slopes of these data in their regions of linear response, 兩 H applied兩 ⬎20 kOe. Figures 1共b兲 and 1共e兲 show the data of Figs. 1共a兲 and 1共d兲 following subtraction of the measured diamagnetic signal, while Figs. 1共c兲 and 1共f兲 show the same data, but with the linear fits to the high field data 关i.e., the dotted lines in Figs. 1共a兲 and 1共d兲兴 subtracted. It follows that the data in Figs. 1共b兲 and 1共e兲 show the full magnetization of the FePt/polymer assembly, whereas the data in Figs. 1共c兲 and 1共f兲 show the magnetization only of those components

It is clear from Figs. 1共b兲 and 1共e兲 that the magnetization of the sample has not begun to saturate at even the highest applied fields 共70 kOe兲. This indicates that some spins are not aligned with the nanoparticle cores—possibly due to surface canting8 or interactions between metal ions and the surface ligands 共for example, the formation of a surface oxide layer兲. We fit the nonhysteretic regions of the data in Fig. 1共c兲 to the Langevin paramagnetic function, which is known to describe the magnetization of superparamagnetic particles:6 M (x)⫽N ␮ 关 coth x⫺(1/x) 兴 , where x⫽ ␮ H/k B T, N is the number of nanoparticles, ␮ is the magnetic moment of an individual nanoparticle, H is the applied field, k B is Boltzmann’s constant, and T is the absolute temperature. From the best fit value for ␮ (1.1⫻10⫺18 emu), we can obtain an approximate value for the average size of those magnetic domains already saturated at 20 kOe 共presumably the nanoparticle cores兲 at 10 K. Assuming a saturation magnetization of 1030 G 关the observed value for disordered FePt 共Ref. 9兲兴, the fit to Fig. 1共c兲 yields an average domain size of 1.3 nm. This value is probably an underestimate; the Langevin function does not include the effects of magnetoanisotropy, clearly present, as demonstrated by the observed hysteresis. Note that an assembly of particles, each with a magnetic moment of 1.1⫻10⫺18 emu, will reach 90% of saturation at an applied field of 12 kOe at 10 K, whereas at 300 K a field of 350 kOe would be required. This 30-fold difference in the required applied field readily accounts for the largely unsaturated magnetic response observed at 300 K 关Fig. 1共e兲兴. Plots of magnetization as a function of applied field following an anneal at 580 °C for 30 min under N 2 are shown in Fig. 2; data taken at 10 K for assemblies comprised of one, two, three, and four nanoparticle layers are shown as Figs. 2共a兲, 2共b兲, 2共c兲, and 2共d兲, respectively, while equivalent measurements taken at 300 K are shown as Figs. 2共e兲, 2共f兲, 2共g兲, and 2共h兲. These measurements show that, postanneal, the magnetizations of all of the samples are largely saturated by 70 kOe, suggesting that most of the spins have now been incorporated magnetically into the nanoparticles. This conclusion is further supported by fitting the 300 K data from the annealed three layer sample 关Fig. 2共g兲兴 to a Langevin paramagnetic function. Again assuming a saturation magnetization of 1030 G, the best fit value for the diameter of the average superparamagnetic particle is 4 nm. This value compares well with the diameter of 4 nm observed by transmission electron microscopy 共not shown兲, indicating that, postanneal, each nanoparticle is comprised of a single magnetic domain. Comparing the saturation moments observed in Figs. 1共c兲 and 2共c兲, one can estimate the volume of the magnetic material incorporated into the single domain nanoparticles through annealing. The saturated magnetic moment/cm2 obtained from the Langevin fit to Fig. 1共c兲 is 44 ␮emu/cm2. The equivalent fit to Fig. 2共c兲 yields 103 ␮emu/cm2. Assuming the postanneal magnetic domain size to be 4 nm, the volume of the material incorporated into these domains as a

Downloaded 08 Mar 2004 to 129.107.14.178. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

J. Appl. Phys., Vol. 95, No. 3, 1 February 2004

Held, Zeng, and Sun

1483

FIG. 2. Magnetic moment/area vs applied field measured at 10 K for polymer-mediated FePt nanoparticle assemblies annealed at 580 °C under N 2 for 30 min 共a兲 one, 共b兲 two, 共c兲 three, and 共d兲 four nanoparticle layers thick. 共e兲, 共f兲, 共g兲, and 共h兲 show equivalent data taken at 300 K. The diamagnetization of the Si substrate has been subtracted from all the data. The solid line in 共g兲 is a best fit to the Langevin function. The solid line in 共h兲 is a superposition of a Langevin paramagnetic function and the hysteretic lineshape predicted for a collection of randomly oriented Stoner–Wohlfarth particles. Data taken at increasing magnetic fields shown as open squares and data taken at decreasing fields shown as solid squares.

result of the anneal is equivalent to the outermost 0.5 nm of these particles. Put differently, our data is consistent with, but does not prove, a model whereby the nanoparticles as synthesized have a 0.5 nm outer shell of weakly ferromagnetic 共e.g., canted or ligand bound兲 material which, during the annealing process, is incorporated into domains comprised of entire nanoparticles. The data in Fig. 2 show that the saturation moment per unit area of the nanoparticle assemblies increases monotonically 共and more or less linearly兲 with number of deposited layers, both at 10 and 300 K. This indicates that all of the layers are comprised of ferro- and super-paramagentic nanoparticles. The samples comprised of one, two, and three nanoparticle layers all exhibit the behavior expected for blocked superparamagnetic particles at 10 K, but show little hysteresis at 300 K. This is consistent with particles comprised of disordered fcc FePt. It is not consistent, however, with well-ordered 4 nm L1 0 FePt nanoparticles, as these would exhibit hysteresis at 300 K.10 The observed corecivity increases abruptly between the three and four layer thick assemblies 关Figs. 2共c兲 and 2共d兲兴. This indicates that the four layer assembly contains at least some particles with significantly greater magnetoanisotropy than the thinner samples— presumably L1 0 FePt. However, the observation that the coercivity in Fig. 2共d兲 is only 7.6 kOe suggests that the L1 0 phase is only partially ordered.10 Further, the observation of hysteresis in the 300 K data of Fig. 2共h兲 supports the presence of some L1 0 FePt particles in the four layer film. Note, however, in Fig. 2共h兲, that the width of the hysteresis curve observed at finite magnetic moments is greater than the coercivity at zero moment. This unusual behavior is consistent with an assembly comprised of both ferro- and superparamagnetic particles. The solid line in Fig. 2共h兲 is a best fit of the data to a superposition of the Langevin function for 4 nm fcc FePt nanoparticles and a curve which describes the hysteretic behavior of a collection of randomly oriented Stoner–Wohlfarth particles.11 The coercivity of the Stoner– Wohlfarth curve and the values of the saturation moments of

the two superimposed curves are the only adjustable parameters in the fit; the resulting contributions of the superparamagnetic and ferromagnetic curves to the solid line are 60 and 40%, respectively. The agreement between the data and the solid line in Fig. 2共h兲 demonstrates that the observed shape of the magnetization data at 300 K can be readily modeled by assuming contributions to the magnetization from both blocked and unblocked superparamagnetic particles, with a contribution of 40% from the ferromagnetic L1 0 FePt nanoparticles in this assembly. Note that the data at 10 K 关Fig. 2共d兲兴 do not have the two phase lineshape of the 300 K data. This is consistent with a distribution in the sizes of the magnetic domains—at very low temperatures the coercivity is largely independent of particle size whereas, at higher temperatures, large particles 共or aggregates兲 will remain ferromagnetic, while smaller ones 共with the same magnetoanisotropy兲 become superparamagnetic.1,10 In this work we have observed a clear dependence of the magnetic behavior of annealed assemblies of FePt nanoparticles on sample thickness—a result with potentially significant technological consequences. While it is possible that this thickness dependence could result either from the upper layers of nanoparticles becoming oxidized or the lower layers forming a silicide structure, neither of these possibilities are consistent with x-ray diffraction measurements,12 which show no evidence for either iron oxides or silicides following an anneal at 580 °C. Another possibility is that the degree of interparticle aggregation increases with sample thickness. A recent report indicates that larger FePt particles are more likely to order during an anneal.13 Thus, if the four layer sample has a greater fraction of aggregated particles, it could be these aggregated clusters which order at 580 °C and subsequently contribute a ferromagnetic component to the magnetization. In this case, it may be necessary to anneal above 580 °C to sufficiently order all of the nanoparticles. The choice of ligands stabilizing the particles would then also have to be modified, so as to prevent aggregation.14 Then, with this encapsulation of the particles in an appropriate ma-

Downloaded 08 Mar 2004 to 129.107.14.178. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp

1484

J. Appl. Phys., Vol. 95, No. 3, 1 February 2004

trix, it should be possible to prepare thin nanoparticle assemblies which would indeed be usable as magnetic media. ACKNOWLEDGMENTS

We thank C. Murray and T. Thomson for helpful discussions. D. Weller and A. Moser, IEEE Trans. Magn. 35, 4423 共1999兲. S. H. Sun, C. B. Murray, D. Weller, L. Folks, and A. Moser, Science 287, 1989 共2000兲. 3 S. Stappert, B. Rellinghaus, M. Acet, and E. F. Wassermann, J. Cryst. Growth 252, 440 共2003兲; S. Kang, J. W. Harrell, and D. E. Nikles, Nano Lett. 2, 1033 共2002兲. 4 Y. Huang, H. Okumura, G. C. Hadjipanayis, and D. Weller, J. Magn. Magn. Mater. 242, 317 共2002兲. 5 S. H. Sun, S. Anders, H. F. Hamann, J. U. Thiele, J. E. E. Baglin, T. Thomson, E. E. Fullerton, C. B. Murray, and B. D. Terris, J. Am. Chem. Soc. 124, 2884 共2002兲. 6 C. P. Bean and J. D. Livingston, J. Appl. Phys. 30, 120S 共1959兲. 7 S. H. Sun, S. Anders, T. Thomson, J. E. E. Baglin, M. F. Toney, H. F. 1 2

Held, Zeng, and Sun Hamann, C. B. Murray, and B. D. Terris, J. Phys. Chem. B 107, 5419 共2003兲. 8 M. P. Morales, S. Veintemillas-Verdaguer, M. I. Montero, C. J. Serna, A. Roig, L. Casas, B. Martinez, and F. Sandiumenge, Chem. Mater. 11, 3058 共1999兲; B. Stahl, J. Ellrich, R. Theissmann, M. Ghafari, S. Bhattacharya, H. Hahn, N. S. Gajbhiye, D. Kramer, R. N. Viswanath, J. Weissmuller, and H. Gleiter, Phys. Rev. B 67, 014 422 共2003兲. 9 R. Hayn and V. Drchal, Phys. Rev. B 58, 4341 共1998兲. Estimates of the radii of magnetic domains scale as the cube root of magnetization, and are thus relatively insensitive to the precise value assumed for the magnetization. 10 J. Garcı´a´-Otero, A. J. Garcı´a-Bastida, and J. Rivas, J. Magn. Magn. Mater. 189, 377 共1998兲. Assuming a magnetoanistotropy of 7.5 ⫻10⫺12 ergs/cm3 and a magnetization of 1100 G for L1 0 FePt particles results in predicted coercivities of 62.5 and 31.4 kOe at 10 and 300 K, respectively. 11 E. C. Stoner and E. P. Wohlfarth, Trans. R. Soc. A240, 599 共1948兲. 12 S. Anders, M. F. Toney, T. Thomson, R. F. C. Farrow, J. U. Thiele, B. D. Terris, S. Sun, and C. B. Murray, J. Appl. Phys. 93, 6299 共2003兲. 13 Y. K. Takahashi, T. Ohkubo, M. Ohnuma, and K. Hono, J. Appl. Phys. 93, 7166 共2003兲. 14 S. Momose, H. Kodama, N. Ihara, T. Uzumaki, and A. Tanaka, Jpn. J. Appl. Phys., Part 2 42, L1252 共2003兲.

Downloaded 08 Mar 2004 to 129.107.14.178. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp