2. 10D304-2 Eid et al. J. Appl. Phys. 97, 10D304 ͑2005͒
exchange bias systems where typically TC ӷ TB. Here, we
demonstrate the exchange biasing of the Ga1−xMnxAs layer
by an overgrown antiferromagnetic MnO layer both with
TC ϳ TB and TC Ͼ TB.
Low temperature MBE growth is performed in an Ap-
plied EPI 930 system equipped with Ga, Mn, and As effusion
cells. “Epiready” semi-insulating GaAs ͑100͒ substrates are
deoxidized using the standard protocol, by heating to
ϳ580 ° C with an As flux impinging on the surface. A
100 nm thick GaAs buffer layer is grown after the deoxidiza-
tion. Then, samples are cooled to ϳ250 ° C for the growth of
a 5 nm thick, low-temperature GaAs layer, followed by a
10 nm thick Ga1−xMnxAs layer ͑x ϳ 0.06͒. Growth is per- FIG. 1. Magnetization as a function of temperature and applied magnetic
formed under group V rich conditions with an As:Ga beam field ͑hysteresis loops͒ for sample Ga0.94Mn0.06As ͑10 nm͒ / MnO ͑8 nm͒
equivalent pressure ratio of ϳ12: 1. After the Ga1−xMnxAs grown using second indium mounting protocol field and field cooled at H
= 1000 Oe from T = 200 K to T = 10 K. ͑a͒ Low-field magnetization versus
growth, samples are transferred in situ to an adjoining ultra- temperature for two pieces from different parts of the same sample ͓indium-
high vacuum ͑UHV͒ buffer chamber and the As cell is free portion ͑no annealing͒ and indium-mounted portion͔ measured at H
cooled to the resting temperature of 110 ° C to avoid forma- = 100 Oe. ͑b͒ Field-cooled hysteresis loop for indium-free portion of
tion of MnAs clusters during the subsequent Mn growth. sample. There is no horizontal shift in the loop and the coercivity is low. ͑c͒
Field-cooled hysteresis loop for indium-mounted portion of sample. The
When the As pressure in the growth chamber decreases to an loop is shifted and has an enhanced coercivity. ͑d͒ Field-cooled hysteresis
acceptable level, the wafers are reintroduced into the cham- loop of an indium-free free portion that was annealed at T = 200 ° C in at-
ber. Then, a Mn capping layer with a thickness of ϳ4- or mosphere for 1 min.
ϳ8 nm is deposited. Mn growth is performed at room tem-
perature in order to prevent interdiffusion and chemical re- heterostructures. TEM is used to further characterize the
action between the Mn and Ga1−xMnxAs layers.15 Even structure of the reacted region. Cross-sectional TEM samples
though the capping layer is expected to be pure Mn are prepared by chemical mechanical polishing, dimpling,
͑99.999% source purity͒, the Mn overlayer rapidly oxidizes and ion milling using 2.7 keV Ar+. The TEM is performed
when the samples are removed from the UHV chamber. using a Philips CM30 transmission electron microscope un-
The growth mode and surface reconstruction are moni- der an operating voltage of 300 kV.
tored in situ by reflection high-energy electron diffraction In order to examine the effect of postgrowth annealing,
͑RHEED͒ at 12 keV. The thickness of the Ga1−xMnxAs layer two protocols were designed to mount the wafers to the
is calculated from RHEED oscillations, while the thickness sample holders. In the first protocol, indium covers the entire
of the Mn layer is estimated from RHEED oscillations of bottom surface of the wafer. In the second protocol, only two
MnAs ͑whose growth rate is mainly determined by the stick- edges of the sample are attached with indium, leaving the
ing coefficient of Mn͒ and verified using TEM, RBS, and middle part suspended. Samples of the first kind have to be
XRR measurements. The Mn concentration in our annealed at ϳ220 ° C for a few minutes in order to melt the
Ga1−xMnxAs is x ϳ 0.06, estimated from electron probe mi- indium and remove the sample from the block. For the sec-
croanalysis of earlier calibration samples grown using simi- ond type of sample, the center portion can be directly re-
lar Ga and Mn fluxes. The RHEED pattern during the growth moved by cleaving without any heating, while the indium-
of the Ga1−xMnxAs layer has a streaky 1 ϫ 2 surface recon- bonded edges require a short thermal anneal. Hence, we can
struction suggesting the good crystalline quality of the layer. systematically study the effect of the short annealing in-
During the Mn growth, the RHEED pattern consists of sharp, curred during removal from the wafer holders, as well as
elongated streaks and its symmetry is suggestive of the sta- subsequent ex situ annealing for the identical sample. We
bilization of a cubic phase of Mn.7,16. will show that annealing has significant effects upon the cap-
Magnetization measurements are performed using a ping Mn layer due to the high reactivity of Mn with oxygen.
commercial SQUID. Samples are measured with the mag- Figure 1͑a͒ shows a magnetization-versus-temperature
netic field in plane along the ͓110͔ direction as a function of curve for a sample with TC ϳ 90 K. Data are shown for two
both temperature and applied magnetic field. The surface and pieces from the same wafer grown using the second mount-
subsurface composition is examined by XPS and RBS. The ing protocol. One piece is from the indium-free portion of
former measurements are performed using a Kratos Analyti- the wafer and is not heated after removal from the UHV
cal Axis Ultra system. The photoelectrons are excited using chamber. Another is from the indium-bonded portion and
monochromatic Al K␣ x rays ͑with a photon energy of hence undergoes a rapid thermal anneal to ϳ220 ° C during
1486.6 eV͒. For depth profiling, the samples are ion milled sample removal. The low background magnetization at tem-
using 4 keV Ar+. RBS is performed using 1.4 and 2.3 MeV peratures above TC indicates that the sample is of good qual-
and 20 C of integrated charges of He+ ions with both nor- ity without large Mn2As, GaMn, or MnAs clusters. Although
mal and glancing angle detector geometries, corresponding we observe no difference in the TC of the indium-free and
to scattering angles of 165 deg and 108 deg, respectively. indium-mounted portions of the sample, we do note that the
Both random and ͗100͘ channeling measurements are con- former has a smaller low-temperature saturated moment
ducted to determine the composition and depth profile of the compared to the latter. Figures 1͑b͒ and 1͑c͒ show the mag-
Downloaded 05 May 2005 to 146.186.190.234. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
3. 10D304-3 Eid et al. J. Appl. Phys. 97, 10D304 ͑2005͒
FIG. 3. RBS channeling spectra of Mn/ Ga1−xMnxAs before ͑solid line͒ and
after ͑dotted line͒ postgrowth annealing in air using 2.3 MeV He+ beams
with glancing angle geometry.
this assignment. These satellite excitations are typical for
MnO and are not present in either Mn2O3 or MnO2.17–19
Both their shape and position remain constant with depth,
while only their intensity decreases due to a decrease in Mn
content. Thus, the annealed film is nearly uniformly oxidized
FIG. 2. Mn 2p XPS spectra acquired as a function of depth for the indium-
with MnO being the dominant form of Mn throughout the
free portion of the Mn/ Ga1−xMnxAs heterostructure ͑a͒ annealed in atmo- layer.
sphere at 200 ° C for 1 min and ͑b͒ as-grown. Data acquired while simulta- Figure 2͑b͒ shows high-resolution Mn 2p spectra for the
neously sputtering away the free surface of the sample using 4 keV Ar+; as-grown piece. In contrast, the Mn 2p3/2 line from the as-
sputtering time is proportional to depth below the free surface of the sample.
grown piece exhibits a low binding energy shoulder after
60 s and shifts to 639 eV after 90 s. Ar sputtering. Satellite
netization ͑M͒ of the bilayer as a function of the applied lines also disappear at this point. This clearly indicates that
magnetic field ͑H͒ after the samples were cooled to the mea- while the surface layers of the as-grown piece are oxidized,
suring temperature ͑T = 10 K͒ in the presence of an external metallic Mn0 dominates in the bottom layers. The metallic
magnetic field of 1 kOe. Figure 1͑b͒ is the magnetization of Mn0 bonded Mn would be consistent with the bottom layers
an indium-free part of the wafer. The magnetization curve is being either elemental Mn or Mn in a metallically bonded
symmetric about the zero applied field, indicating the ab- compound such as MnGa or Mn2As. The latter scenario is
sence of exchange bias. Figure 1͑c͒ shows a shifted hyster- consistent with earlier studies of Mn grown on GaAs. Jin et
esis loop measured for an indium-mounted portion of the al.17 reported the formation of a Mn2As-type Mn–Ga–As in-
sample. Finally, Fig. 1͑d͒ shows a hysteresis loop of an terfacial layer during Mn growth on GaAs at 400 K. In ad-
indium-free portion of the sample that was intentionally an- dition, Hilton et al..16 found that an epitaxial Mn0.6Ga0.2As0.2
nealed in atmosphere at 200 ° C for 1 min. The center of the layer consisting of tetragonal Mn2As and MnGa formed be-
hysteresis loop is also shifted from zero. These results tween Mn and GaAs as a result of solid-state interfacial re-
demonstrate that a certain amount of annealing is neces- actions during annealing. At first glance, it may appear sur-
sary to create exchange bias in the Mn/ Ga1−xMnxAs prising to propose the presence of an interfacial reacted layer
heterostructures. even in samples that have never been heated above room
To further understand these results, we perform depth- temperature. Our results are, however, consistent with recent
dependent XPS studies on the indium-free portion of the in situ XPS studies showing that Mn growth on GaAs at
wafer. This is accomplished by acquiring XPS data while temperatures as low as 95 ° C leads to the formation of an 11
simultaneously sputtering away the free surface of the monolayer thick Mn0.6Ga0.2As0.2 interfacial reacted layer.20
sample. In such measurements, time is proportional to depth Simulations of the random RBS spectra confirm the for-
below the surface. Figure 2͑a͒ depicts high-resolution Mn 2p mation of MnO with no detectable Ga or As at the surface.
spectra for the piece annealed in atmosphere at 200 ° C for The overlapped interface surface peaks for Ga and As in the
1 min. The Mn 2p3/2 line from the annealed piece is centered glancing angle detector ͗100͘ channeling RBS spectra in Fig.
at ϳ641.0 eV and its position is in agreement with the bind- 3 correspond to more Ga and As ͑ϳ5 ϫ 1015 atoms/ cm2͒
ing energy of Mn2+, indicating the formation of MnO ͑me- than that expected for an abrupt interface ͑ϳ1 – 2
tallic Mn0 has a 2p3/2 line at 639 eV, while lines from ϫ 1015 atoms/ cm2͒. This clearly indicates the presence of an
Mn2O3 ͑Mn3+͒ and MnO2 ͑Mn4+͒ have binding energies of interfacial reacted layer. The increase in the amount of Ga
ϳ641.7 and ϳ642.5 eV, respectively͒. The two satellite lines and As would correspond to a ϳ2 nm thick Mn0.6Ga0.2As0.2
spaced by 5.5 ev from 2p3/2 and 2p1/2 lines further support if there were no ion channeling in the layer. Since this re-
Downloaded 05 May 2005 to 146.186.190.234. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
4. 10D304-4 Eid et al. J. Appl. Phys. 97, 10D304 ͑2005͒
FIG. 5. XRR measurements for the indium-free portion of a
Mn͑10 nm͒ / Ga1−xMnxAs͑10 nm͒ heterostructure ͑a͒ as-grown and ͑b͒ an-
nealed in atmosphere at 200 ° C for 10 min. Solid lines correspond to the
data, while the dotted lines are fits to an oxide/metal/semiconductor model
described in the text. These samples were not exchange biased; the increased
thickness of the AFM layer was chosen to improve signal to noise in the
XRR measurement. The Ga1−xMnxAs layer was grown under similar condi-
tions as those of exchange biased samples and so has a nominally similar
Mn concentration.
Further support for the proposed reaction model comes
from XRR measurements performed on a 10 nm thick
FIG. 4. Cross-sectional TEM micrographs of Mn/ Ga1−xMnxAs heterostruc-
Ga1−xMnxAs sample capped with a Mn layer that is nomi-
tures mounted by the first mounting protocol during growth: ͑a͒ showed no nally 10 nm thick. These samples were not exchange biased;
exchange bias and ͑b͒ showed exchange bias. We note that these measure- rather, the increased thickness was chosen to be able to ef-
ments are unable to show any observable contrast between the GaAs buffer fectively probe the grown bilayer structures with XRR tech-
layer and the thin Ga1−xMnxAs layer due to the low Mn concentration ͑
ϳ6 % ͒. niques. Figure 5͑a͒ depicts the XRR spectrum for the as-
grown indium-free part of the wafer ͑solid line͒. It indicates
the presence of two thin layers with different electron den-
acted layer is grown epitaxially,16 some channeling may be sity. The figure also shows our current attempts at fitting the
expected, and, therefore, the reacted layer may actually be XRR data ͑dotted line͒ assuming an oxide/metal/
thicker. semiconductor trilayer structure. The thickness of the oxide
Figure 4͑a͒ shows a dark horizontal band ͑ϳ2.3 nm layer increases, while that of the interfacial metallic layer
thick͒ at the interface in the cross-sectional TEM micrograph decreases when the sample is annealed in atmosphere at
of a sample fabricated using the first mounting protocol 200 ° C. Finally, a uniform oxide film is formed, as shown in
͑complete In mounting͒. This sample does not show ex- Fig. 5͑b͒. Figure 6 schematically shows our proposed model
change bias, suggesting that the interfacial reacted layer may of the sample structure before and after postgrowth anneal-
consist of Mn0.6Ga0.2As0.2. This is consistent with the RBS ing.
channeling results of the unannealed samples which were We now examine the temperature and magnetic field-
mounted using the second protocol and which do not show dependent properties of the exchange bias in our
exchange bias. Energy dispersive spectrometry ͑EDS͒ in the Ga1−xMnxAs: MnO bilayer structures. Figures 7͑a͒–7͑d͒
TEM confirms the surface layer as MnOx with a thickness of
ϳ9 nm. Upon postgrowth annealing in air, the samples
mounted with the second protocol exhibit exchange bias and
the RBS channeling interfacial Ga and As peaks increase
slightly ͑ϳ1 ϫ 1015 atoms/ cm2͒. A thin, bright, horizontal
line is observed at the interface by cross-sectional TEM ͓Fig.
4͑b͔͒ for a sample that shows exchange bias. The change in
contrast is consistent with a decrease in density, and the in-
crease in the channeled Ga and As interfacial yields with the
FIG. 6. Schematic drawings of Mn/ Ga1−xMnxAs MBE-grown heterostruc-
decrease in channeling as a result of oxidation of the inter- tures; ͑a͒ as-grown in the MBE chamber; ͑b͒ after removal from the UHV
facial Mn–Ga–As layer. system and exposed to air; and ͑c͒ after post growth annealing in air.
Downloaded 05 May 2005 to 146.186.190.234. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
5. 10D304-5 Eid et al. J. Appl. Phys. 97, 10D304 ͑2005͒
FIG. 7. Hysteresis loops indicating the role of the MnO cap in producing
exchange bias and the behavior of the bias with direction of the cooling
field. Measurements were made at T = 10 K using sample
Ga0.92Mn0.08As ͑10 nm͒ / MnO ͑4 nm͒ grown using the first indium mount- FIG. 8. Exchange field, HE = −͑HC− − HC+͒ / 2, and coercive field, HC
ing protocol. ͑a͒ and ͑b͒ loops for field-cooled measurements ͑H = ͑−HC− + HC+͒ / 2, as a function of temperature for field cooling at H
= + 2500 Oe and H = −2500 Oe, respectively͒. ͑c͒ Zero-field cooled hyster- = 2500 Oe from T = 200 K. ͑a͒ Sample Ga0.92Mn0.08As ͑10 nm͒ /
esis loop. ͑d͒ Field-cooled hysteresis loop ͑H = 1000 Oe͒ for an uncapped MnO ͑4 nm͒ grown using first indium mounting protocol. ͑b͒ Sample
sample. Ga0.94Mn0.06As͑10 nm͒ / MnO͑8 nm͒ grown using second indium mounting
protocol.
show the hysteresis loops of exchange biased and unbiased
samples. Figures 7͑a͒ and 7͑b͒ show that the hysteresis loop yond HE approaching zero as T approaches TC. We note that
is shifted to the left or the right depending on the direction of recent studies of Co/ CoO bilayers have shown a sign rever-
sal in HE for exchange biased systems;21 however, we be-
the cooling field. In Fig. 7͑a͒ ͓Fig. 7͑b͔͒, the sample was
lieve that the data shown in Fig. 8͑b͒ are likely skewed by a
cooled to the measuring temperature in a field of +2500 Oe
small remnant field in the magnetometer. As we discuss in
͓−2500 Oe͔. The hysteresis loops are clearly shifted opposite
the next paragraph, small remnant fields are able to cause
to the direction of the cooling field as is common for
changes in the exchange and coercive fields in these hetero-
exchange-biased systems.8 Ideally the zero-field cooled
structures.
capped sample, Fig. 7͑c͒, would be exactly centered about
Finally, we show in Figures 9͑a͒ and 9͑b͒ the depen-
zero, but we still see a small shift because we cannot elimi-
dence of HE and HC on the cooling field for the two respec-
nate the field coming from the magnetization of the ferro-
tive samples examined in Fig. 8. Both samples show that a
magnetic layer. In Fig. 7͑d͒ there is no shift in the hysteresis
cooling field of only a few Oe is sufficient to create exchange
loop of the uncapped sample, indicating no exchange bias.
Finally, it is also important to notice that the hysteresis loops
of the capped sample in Fig. 7 are all wider than the loop of
the uncapped sample displayed in Fig. 7͑d͒. Exchange bias is
known to enhance HC of the ferromagnetic layer as well as
create a shift in the hysteresis loop, HE.
As the temperature of the sample is changed, HE and HC
will change accordingly. Figure 8͑a͒ shows both HE and HC
as a function of temperature for a sample which has
been cooled down in the presence of a magnetic field
of H = + 2500 Oe. The structure of the sample is
Ga0.92Mn0.08As͑10 nm͒ / MnO͑4 nm͒. Low-field measure-
ments of M͑T͒ indicate that the Curie temperature is TC
ϳ 55 K ͑data not shown͒; HE decreases monotonically with
increasing temperature until it becomes zero at TB = 48 K. HC
decreases, goes through a plateau around TB, and then de-
creases monotonically to zero at TC. Figure 8͑b͒ shows the
same quantities for a sample with a different TC. This sample
has an approximate structure of Ga0.94Mn0.06As͑10 nm͒ /
MnO͑8 nm͒ and TC ϳ 90 K ͓see Fig. 1͑a͔͒. HE approaches
zero at the same temperature as the prior sample, indicating FIG. 9. HE and HC as a function of cooling field for measured at T = 10 K.
Horizontal axis is plotted on two different scales and split at H = 1.5 kOe. ͑a͒
that despite the large difference in TC for the two samples,
Sample Ga0.92Mn0.08As ͑10 nm͒ / MnO ͑4 nm͒ grown using the first indium
the blocking temperature is unchanged because it depends on mounting protocol. ͑b͒ Sample Ga0.94Mn0.06As ͑10 nm͒ / MnO ͑8 nm͒
the antiferromagnetic layer only. Likewise, HC extends be- grown using the second indium mounting protocol.
Downloaded 05 May 2005 to 146.186.190.234. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp
6. 10D304-6 Eid et al. J. Appl. Phys. 97, 10D304 ͑2005͒
bias. Only a slight change in HE is observed for cooling berger for the useful discussion and assistance with XPS
fields a few orders of magnitude larger than the minimum measurements. We thank M. S. Angelone for the help with
field to create bias. This is because a small field is needed to the XRR measurements. Work at ORNL was carried out un-
saturate the magnetization of the FMSC layer at TB. The der Contract No. DE-AC05-00OR22725, U. S. Department
magnetization of this layer turn sets the bias; increasing the of Energy.
magnetic field further has no significant effect on the bilayer.
When the external cooling field is small enough ͑H Ͻ 7 Oe͒ 1
H. Ohno, in Semiconductor Spintronics and Quantum Computation, edited
there is almost no exchange bias shift as expected. Alterna- by D. D. Awschalom, D. Loss, and N. Samarth, ͑Springer, Berlin, 2002͒,
tively, HC does not approach zero for any external magnetic p. 1.
2
S. A. Wolf, D. D. Awschalom, R. A. Buhrman, J. M. Daughton, S. von
field and changes very slightly with field.
Molnar, M. L. Roukes, A. Y. Chtchelkanova, and D. M. Treger, Science
In summary, we have grown a set of GaxMn1−xAs/ MnO 294, 1488 ͑2001͒.
heterostructures that exhibits exchange bias and an enhance- 3
Y. Ohno, D. K. Young, B. Beschoten, F. Matsukura, H. Ohno, and D. D.
ment of coercivity. We have studied the dependence of the Awschalom, Nature ͑London͒ 402, 790 ͑1999͒.
4
coercivity and exchange field on temperature and cooling H. Ohno, N. Akiba, F. Matsukura, A. Shen, K. Ohtani, and Y. Ohno, Appl.
Phys. Lett. 73, 363 ͑1998͒.
field. Both HC and HE depend dramatically on temperature 5
N. Samarth, in Solid State Physics, edited by H. Ehrenreich and F.
but have a much weaker dependence on the cooling field. Spaepen ͑Elsevier-Academic, San Diego, 2004͒, vol. 58, p. 1.
The blocking temperature does not change from sample to 6
K. F. Eid, M. B. Stone, K. C. Ku, O. Maksimov, P. Schiffer, N. Samarth,
sample while TC varies, most likely due to differences in T. C. Shih, and C. J. Palmstrom, Appl. Phys. Lett. 85, 1556 ͑2004͒.
7
Ga1−xMnxAs growth conditions. Our detailed structural stud- K. F. Eid, O. Maksimov, M. B. Stone, P. Schiffer, N. Samarth, cond-mat/
ies of the capping layer indicate that it oxidizes after the 0409108.
8
J. Nogués and I. Schuller, J. Magn. Magn. Mater. 192, 203 ͑1999͒.
removal from the UHV chamber. However, the oxidation is 9
W. H. Meiklejohn and C. P. Bean, Phys. Rev. 102, 1413 ͑1956͒.
partial, resulting in a formation of MnO / Mn– Ga– As/ 10
G. A. Prinz, J. Magn. Magn. Mater. 200, 110 ͑1999͒.
Ga1−xMnxAs trilayer structure. Since the metallic Mn–Ga–As 11
L. Thomas, J. Luning, A. Scholl, F. Nolting, S. Anders, J. Stohr, and S.
region does not appear to create any significant exchange Parkin, Phys. Rev. Lett. 84, 3462 ͑2000͒.
12
bias, short annealing is necessary to uniformly oxidize this K. Nagasaka, Y. Seyama, L. Varga, Y. Shimizu, and A. Tanaka, J. Appl.
Phys. 89, 6943 ͑2001͒.
interfacial layer to form MnO. These results are important 13
M. S. Jagadeesh and M. S. Seehra, Phys. Rev. B 23, 1185 ͑1981͒.
for enhancing the potential for FMSC for use in spintronics 14
X. W. Wu and C. L. Chien, Phys. Rev. Lett. 81, 2795 ͑1998͒.
devices, the basic understanding of exchange bias, and for 15
J. L. Hilton, B. D. Schultz, S. McKernan, and C. J. Palmstrøm, Appl.
designing new experiments to study the optical and spin Phys. Lett. 84, 3145 ͑2004͒.
16
transport properties in exchange biased FMSC. X. Jin, Y. Chen, X. W. Lin, D. S. Dong, Y. Chen, M. Xu, W. R. Zhu, X.
Wang, X. L. Shen, and L. Li, Appl. Phys. Lett. 70, 2445 ͑1997͒.
17
S. P. Jeng, R. J. Lad, and V. E. Henrich, Phys. Rev. B 43, 11971 ͑1991͒.
ACKNOWLEDGMENTS 18
M. A. Stranick, Surf. Sci. Spectra 6, 31 ͑1999͒.
This research has been supported by the DARPA-SPINS
19
M. A. Stranick, Surf. Sci. Spectra 6, 39 ͑1999͒.
20
J. L. Hilton, B. D. Schultz, S. McKernan, and C. J. Palmstrøm ͑unpub-
program under Grant Numbers N00014-99-1093, -99-1- lished͒.
1005, -00-1-0951, and -01-1-0830, by ONR N00014-99-1- 21
F. Radu, M. Etzkorn, R. Siebrecht, T. Schmitte, K. Westerholt and H.
0071, and by NSF DMR 01-01318. We thank J. Shallen- Zabel, Phys. Rev. B 67, 134409 ͑2003͒.
Downloaded 05 May 2005 to 146.186.190.234. Redistribution subject to AIP license or copyright, see http://jap.aip.org/jap/copyright.jsp