PHYSICAL REVIEW B 95, 094430 (2017)

Three-dimensional characterization of Co/Pd multilayer thin films

using resonant soft x-ray scattering

Samuel Flewett,1,* Durgamadhab Mishra,2 Thiago J. A. Mori,3 Christian M. Günther,4 Juliano C. Denardin,5 Simón Oyarzún,5
Sebastián Michea,5 Dieter Engel,6 Manuel Fohler,2 Tulio C. R. Rocha,3 Alexandra Ovalle F.,1 Leandro T. Núñez A.,1
Bastian Pfau,6 Juan Escrig,5 and Stefan Eisebitt4,6
1
Instituto de Fı́sica, Pontificia Universidad Católica de Valparaı́so, Avenida Universidad 330, Valparaı́so, Chile
2
Helmholtz-Zentrum Berlin, Albert-Einstein-Straße 15, 12489 Berlin, Germany
3
Laboratorio Nacional de Luz Sı́ncrotron–LNLS/CNPEM, Caixa Postal 6192, CEP 13083-970, Campinas SP, Brazil
4
Institut für Optik und Atomare Physik, Technische Universität Berlin, Straße des 17. Juni 135, 10623 Berlin, Germany
5
Departamento de Fı́sica, Universidad de Santiago de Chile, Avenida Ecuador 3493, Estación Central, Santiago, Chile
6
Max-Born-Institut, Max-Born-Straße 2a, 12489 Berlin, Germany
(Received 15 September 2016; revised manuscript received 19 January 2017; published 29 March 2017)

In this work, we take a step toward nanometer resolution magnetic tomography, demonstrating the use of
resonant magnetic x-ray scattering for three-dimensional (3D) characterization of Co/Pd multilayers. Trading
coherence for x-ray flux, we measured magnetic scattering out to full period resolutions of 20 nm for incidence
angles up to 75◦ from normal incidence, which was sufficient for a semiquantitative analysis of the 3D magnetic
domain structure of such films. For the analysis of such patterns, we developed a scattering model based upon
multiple plane propagation of the beam through the sample, taking into account the effects of multiple scattering
not accounted for with Born or Rytov approximation based models. Our results demonstrate that 3D statistical
characterization of nanoscale magnetic features is currently possible, and it is expected that the advent of high
brilliance sources such as MAX IV in Sweden and SIRIUS in Brazil will provide the coherent flux necessary to
permit the generalization of this work to nanometer resolution magnetic tomography.

DOI: 10.1103/PhysRevB.95.094430

I. INTRODUCTION AND STATEMENT OF THE PROBLEM 5 years. In this work, studying the partially coherent scattering
patterns from such samples at different incidence angles, we
The advancement of thin film structure growth and nanopat-
demonstrate that it is possible to study the statistical properties
terning has led to the development of novel magnetic structures
of the samples to such resolutions, and at the same time identify
at the nanoscale [1,2], which promise to play a significant role
and partially solve some of the technical challenges standing
in modern memory-based devices. In particular, researchers
in the way of a high-resolution real-space reconstruction—in
actively wish to explore the control and manipulation of the
particular the difficulties caused by multiple scattering events
magnetization in domains and domain walls for new kinds
within the sample. As such, this article makes a contribution
of memory and logic circuits [3–7]. To achieve tunability in
toward the long-term goal of 3D structure determination, along
these structures, a detailed understanding of the real-space
with the recent work of other groups [15–21].
magnetization distribution inside the domains and domain
The quest for 3D magnetic imaging dates back to the
walls is necessary, which could unravel the correlation between
seminal work of Dürr et al. [14], where a semiquantitative
the physical structure and magnetic configuration of the system
study was performed in three dimensions using reflection
to higher resolution. A lot of effort is therefore being devoted
geometry. A more recent reflection geometry work is that
to imaging these domains with nanometer resolution, and in
of Fin et al. [17]. Samples similar to those studied in this
fact a lot of progress has been made to date in understanding
paper were also investigated. Further techniques have also
two-dimensional (2D) pictures of domains and domain walls.
been suggested, for example the imaging of 100-nm-thick
Examples include the use of techniques such as spin-polarized
cobalt nanowires by electron holography [22] and polarized
low-electron energy microscopy (SPLEEM) [8,9], magnetic
neutron reflectivity (PNR) for the measurement of 20-nm-thick
force microscopy (MFM) [10], spin-polarized scanning tun-
CoPtCr thin films [23]. Neutron tomography was used for 3D
neling microscopy (SP-STM) [11], Kerr microscopy [1], x-ray
magnetic domain imaging [24], however the weaker scattering
magnetic circular dichroism in conjunction with photoemis-
cross section compared to x-rays restricted resolution to
sion electron microscopy (XMCD-PEEM) [12], full field or
the micrometer range. More recent studies using STXM
raster transmission x-ray microscopy (TXM, STXM) [13], and
have extracted limited 3D information [15,16,19], and the
x-ray resonant magnetic scattering in reflection geometry [14].
probable path toward a high-resolution real-space structure
A full three-dimensional (3D) imaging of magnetization is still
is, in our opinion, likely to involve a derivative of STXM
lacking for such systems, however the advent of high brilliance
such as ptychography. Work on such samples using STXM
synchrotron facilities in the next few years suggests that 3D
is, however, still limited by diffraction-limited depth of focus
resolutions of between 10 and 20 nm could be possible within
constraints, which adversely affect the resolution as the sample
thickness is increased. Ptychography is being extensively
developed for 3D structural reconstruction, including chemical
*
samuel.flewett@gmail.com contrast [25,26]. In conjunction with hard x-rays Ptychography

2469-9950/2017/95(9)/094430(11) 094430-1 ©2017 American Physical Society

SAMUEL FLEWETT et al.
FIG. 1. Magnetic force microscopy images of the samples used in
this study. (a) Co(0.5 nm)/Pd (0.8 nm) with 50 repeats, (b) Co(0.8 nm)/
Pd(0.8 nm) with 50 repeats, and (c) and (d) Co(0.8 nm)/Pd(0.8 nm)
with 100 repeats. Parts (a), (b), and (d) were subject to in-plane
demagnetization to produce stripe domains, whereas (c) was subject
to out-of-plane demagnetization, leaving the sample in a worm
domain state.
featuring magnetic contrast was recently demonstrated [20].
This same group has reported magnetic vector tomography of
a magnetic vortex [21], however the magnetic contrast probed
by the hard x-rays used in their work is much weaker than for
soft x-rays, limiting their 3D resolution to around 100 nm.
FIG. 2. An example of simulated stripe domains. In (a), a horizontal slice in the x/y plane is shown through the center of the film in the
case of Mz , and along the surface in the case of Mx and My . In (b), a vertical cross section showing the y/z plane is displayed for both My and
Mz . In both (a) and (b), the color scale is identical, however in (b), the length scale has been adjusted.
094430-2
PHYSICAL REVIEW B 95, 094430 (2017)
As a proof of concept, we use Co/Pd multilayers of
different thicknesses, which form out-of-plane magnetized
domains interspersed by Bloch domain walls due to strong
surface anisotropy [1,10,27,28]. These perpendicular magnetic
anisotropy (PMA) multilayers form worm domains with in-
plane magnetized closure domains [29,30] at the film surface or
interface to a substrate (see Figs. 1 and 2). The worm domains
can be aligned in stripes following an in-plane demagnetization
procedure, and it is possible as well to control the closure
domain size by varying the film thickness, with thicker samples
generally exhibiting larger closure domains [29,30].
This paper is structured in two parts: the first presents the
theoretical model developed for the analysis of the scattering
patterns, and the second discusses the experimental work
performed at the synchrotrons BESSY II and LNLS.
II. MODELING
To model the domain pattern and resulting diffraction, we
adapted the model used by Loh et al. [31] and generalized it
from two to three dimensions. First, a 2D domain pattern was
generated based upon the magnetic scattering that would be
expected from samples illuminated under normal incidence.
Second, the magnetization vectors near the domain walls were
canted as a function of their distance from the domain wall,
the orientation of the domain wall, and the depth within the
sample. Third, x-rays were propagated through the sample in
steps thin enough such that the Born approximation could be
locally applied.
To generate a 2D pattern of domains, we start with a
matrix of uniformly distributed random numbers, which are
convolved with an empirically defined kernel K as adapted
from Loh et al. [31],
K = F{sina (φ)/[(q/qmax − b)c + d]}, (1)
THREE-DIMENSIONAL CHARACTERIZATION OF Co/Pd . . .
where F is the discrete Fourier transform operator, φ is
the azimuthal angle, a is the parameter that controls the
degree of alignment of the domains, b is the modulus
of the normalized scattering vector corresponding to the
average magnetic domain period, c controls the distribution
of domain sizes, and d is a factor added to prevent division
by zero where q/qmax = b. For any given simulation, q/qmax
is the normalized discrete Fourier conjugate of the spatial
coordinate, and it is proportional to the scattering vector. The
exponent a was set to 0 for worm domains and to 12 for
stripe domains, and the parameters c and d were set at 2.5 and
2 × 10−6 , respectively—these values were chosen for the best
qualitative fit between the simulations and the experimental
results. The parameter c may be adjusted downward for a
more ordered domain pattern, or raised to allow the simulation
of a more disordered pattern. Following the application of
the convolution kernel, regions of the convolved matrix with
positive values were assigned a value of 1, and those with
negative values −1, producing a binary domain pattern similar
to Fig. 2(a).
Using the binary domain pattern as a base, the Euclidean
distance transform is calculated with respect to each of the
domain walls using the algorithm of Maurer et al. [32]
as implemented in MATLAB. The magnetization vector was
then canted as a function of this distance transform and
the depth within the sample, with its projection along the
surface set perpendicular to the domain wall. An example
of such simulated stripe domains is shown in Fig. 2, where
the different components of the magnetization vector field
are displayed, both in the horizontal plane (a) and as a
vertical cross section (b). In this model, we consider only the
principal, perpendicular magnetized domain regions plus the
in-plane closure domains, assuming that the closure domains
are perfectly triangular in shape with a constant average
magnetization. There is the possibility that a continuously
or discretely tilted closure domain magnetization reduces the
magnetization projection on the k vector of the incident x-rays
and hence the XMCD signal [1]. We treat this in the model
via a constant magnetization in the closure domain along the
direction shown in Fig. 2 together with a reduced magnitude
of the local magnetization. The best global agreement of the
model to the experimental data in the different geometries
and for both samples is obtained for a closure domain
magnetization with half the magnitude of that in the principal
domains, and we use this value for all simulations shown.
Clearly, our model contains significant simplifications, and
work to upgrade this model to one based upon micromagnetic
simulations is underway.
To simulate the scattering patterns, one must first simulate
the interaction between the x-rays and the sample: This
interaction is governed by the equation [33]
fn = (e · e )fcn + i(e × e ) · Mn fm1
n
+ (Mn · e )(Mn · e)fm2
n
, The observation of certain incidence angles and q values
(2)
where e is the incoming electric field vector direction, e is
the possibly rotated outgoing electric field vector direction,
n n
Mn is the sample magnetization vector, and fcn , fm1 , and fm2
are the charge, circular dichroic, and linear dichroic resonant
scattering factors, respectively. For L2,3 edge resonances of
094430-3
PHYSICAL REVIEW B 95, 094430 (2017)
3d transition metals, it is the first two terms that are of
interest, where regions magnetized parallel to the incident
Poynting vector have differing absorption compared to those
areas magnetized antiparallel, thus giving rise to magnetic
contrast. We estimated the absorption lengths from the Center
for X-Ray Optics (CXRO) data [34] and XMCD measurements
[35]. Working slightly off resonance at 775 eV, the absorption
lengths were estimated at 33 and 66 nm for the magneti-
cally enhanced or diminished cases (parallel or antiparallel
depending upon polarization). Precisely on resonance, the
absorption lengths are substantially shorter, resulting in a
weaker scattering signal, especially for thicker samples and
higher incidence angles.
Due to the thickness of the samples compared to the
absorption lengths and the multiple scattering that occurs
as a result, it is necessary to include the effects of multiple
scattering in the simulations. This was achieved by propagating
a Gaussian beam through a succession of slices perpendicular
to the incident beam using near-field Fresnel propagation,
each of the slices being sufficiently thin such that multiple
scattering may be ignored within each slice. These slices are
calculated from the 3D magnetization volume using linear
interpolation, and only the component of the magnetization
parallel to the incident beam is included in the calculation, in
accordance with Eq. (2). Finally, the effect of partially coherent
illumination is accounted for by approximating the coherent
mode representation [36] by averaging 10 independent rendi-
tions of the algorithm, each rendition representing a coherent
scattering pattern due to illumination by a 2.5-μm-diameter
circular Gaussian beam. The process of propagating the beam
through the sample is shown graphically in Fig. 4. Due to
the use of partially coherent illumination, the difference in
observed scattering patterns between the different polarization
states was negligible, and in the experiments we chose linearly
polarized radiation for the higher flux available. Linearly
polarized light is the incoherent sum of the two circular
polarizations, and a full discussion of its use for transmission
geometry characterization of magnetic domains may be found
in Refs. [37–39].
In Fig. 3, we show simulated diffraction images for a
stripe domain sample that is 150 nm thick with an average
110 nm domain width, with and without 35 nm width and
depth triangular closure domains at the incidence angles of
0◦ , 36◦ , 56◦ , and 70◦ . The choice of angles is such that the
x-rays traverse on average exactly one, two, or four principal
domains at 36◦ , 56◦ , and 70◦ , respectively. It is noted that
there is little difference between the cases with and without
closure domains with regard to the diffraction images at 0◦
and 36◦ . In the 56◦ image, however, close inspection of the
third-order peak (circled) shows that in the closure domain
case it is singular and enhanced, whereas in the case without
closure domains it is split and much weaker.
where the first-order diffraction intensity suffers a marked
reduction and/or a split in q space is a key result of this work
[observable, for example, in Fig. 3(c) at q = 8 μm−1 ]. These
angles are denoted as critical angles in this paper, and they
occur for the following reason: At any illumination angle, the
projected x-ray intensity downstream from the sample may
be expressed in the absence of scattering within the sample
SAMUEL FLEWETT et al.
FIG. 3. Simulations of diffraction patterns at different incident
angles through a stripe domain sample of 110 nm domain width and
150 nm film thickness, with the third-order diffraction peak marked
in each case by the circle. Part (a) is normal incidence diffraction,
(b) is diffraction at 36◦ , (c) at 56◦ , and (d) at 70◦ . In each image, the
left-hand side is simulated with 35 nm triangular closure domains, and
the right-hand side without. In (d), the split first-order Bragg peak for
the case with closure domains means that it is almost exactly on the
second critical angle, whereas in the case without closure domains,
the angle of incidence is slightly below the critical angle as evidenced
by the intact first-order Bragg peak. In this and all subsequent figures,
the axes are qx and qy , which are denominated as the number of full
period repeats per μm omitting the factor of 2π often added.
according to the Beer-Lambert law. Additionally, according
to Eq. (2), the dot product between the magnetization and
the Poynting vector determines whether there is enhanced or
diminished absorption. When the incident beam is at an angle
where it traverses on average two, four, or six domains, half of
its trajectory within the sample is through regions where the
dot product between the Poynting and magnetization vectors
is positive, and the other half through regions where it is
negative. The position of the nth critical angle as illustrated in
Fig. 4(b), therefore, depends upon the precise geometry of the
magnetization pattern according to
θcrit = tan−1 (2nwmag /dmag ), (3)
where wmag and dmag are the magnetic domain widths and
thicknesses, respectively. If the stripe domain pattern were to
be perfectly ordered with uniform magnetic domain size, the
magnetic diffraction intensity at the critical angles would be
identically zero. However, with the disordered stripe patterns
of real samples, there is a natural distribution of domain widths.
For the narrower domains, the critical angle occurs at lower
incidence angles compared with the wider domains, where
the critical angle is shifted to higher incidence angles. In
experiments where the domains exhibit a distribution of sizes,
094430-4
PHYSICAL REVIEW B 95, 094430 (2017)
FIG. 4. (a) Geometry of the x-ray sample interaction at the first
critical angle. The sample is illuminated at an angle θ from the
normal n, such that the beam passes through on average two out-
of-plane domains while traversing the film. (b) Zoom showing the
angles of incidence corresponding to the first three critical angles
for a sample with 110 nm average domain size and 150 nm film
thickness, where the beam traverses two, four, or six out-of-plane
domains, respectively.
there is a corresponding distribution of critical angles, and what
is observed is that the diffraction at q vectors from domains
illuminated at their critical angle undergoes a strong reduction
in intensity, while diffraction from domains of sizes above and
below this value experiences a smaller intensity reduction.
The consequence of this effect is that a splitting occurs in the
broadened Bragg peaks when the film is illuminated at the
critical angle corresponding to the average domain size.
For the case of 70◦ illumination in Fig. 3(d), readers are
advised of the difference in critical angle for the two cases with
and without closure domains, as illustrated by the diffraction
lobe being equally split in one case but not in the other. The
reason for the shift in critical angle is due to subtle changes
in the magnetic sample thickness, as seen by the incident
x-rays—highlighting the utility of the critical angle as a
tool for sample characterization. In the experimental results
presented in the following section, it was observed in almost
all cases that the first critical angle occurs at a higher incidence
angle than expected, suggesting a magnetic thickness less than
the sample thickness. At this stage, we lack a model to fully
explain this observation, and further investigation is underway.
For closer inspection of the effect of different closure
domain shape and size on the observed scattering pattern,
a series of lineout plots (generated by integrating along the
qx axis) at the same angles is shown in Fig. 5. In Figs. 5(a)
and 5(c), the cross-sectional area of the closure domains is
kept constant at 1250 nm2 , and the width and depth are varied
keeping their product constant. In Figs. 5(b) and 5(d), the width
and depth of the triangular closure domains were varied from 5
to 35 nm in an even manner, maintaining their cross-sectional
widths and depths equal in each case. In all cases, the relative
in-plane magnetization of the closure domains was set at half
the value of the principal domains. Comparing the lineouts for
THREE-DIMENSIONAL CHARACTERIZATION OF Co/Pd . . . PHYSICAL REVIEW B 95, 094430 (2017)

FIG. 5. Lineout plots for the positive q half of diffraction simulated from 110 nm stripe domains in a 150-nm-thick film. Parts (a) and (b)
are simulated at 36◦ , and (c) and (d) at 56◦ incident angle. In plots (a) and (c), the closure domain aspect ratio is changed keeping the product
width × height constant at 1250 nm2 , whereas for (b) and (d) the width was maintained equal to the depth, and the overall size was varied. In
each plot, the diffraction orders are labeled, and in (c) and (d) the first order is split due to the critical angle occurring at precisely this incidence
angle.

both cases at 36◦ with those at 56◦ , note that at the critical angle
(56◦ ), the third-order peak intensity is much more strongly
dependent upon the closure domain parameters compared with
the situation at 36◦ . This identification of the direct influence
of the closure domains on the scattering pattern at the critical
angle is one of the chief observations of this work, and it
demonstrates the utility of studying stripe domain diffraction
at such angles of incidence. Comparing Figs. 5(c) and 5(d),
it is important to observe that the behavior at first order is
largely the same regardless of closure domain size, however
the behavior of the diffraction at third order changes widely as
the size and shape are varied.
In the case of worm domains, the concept of a single critical
angle no longer applies, because this is contingent upon a
preferred orientation of the domain pattern. For domains in
a disordered state at higher angles of incidence, there is a
range of domains oriented such that the angle of incidence is
such that the x-rays traverse exactly two, four, or six domains,
provoking a cancellation of the magnetic contrast for a range
of q that depends upon incidence angle. In Fig. 6, we show
the simulated diffraction from a worm domain sample with
the same thickness and average domain size as previous stripe
domain samples, also with incident radiation at 0◦ , 36◦ , 56◦ ,
and 70◦ . Here one notes that the circular domain pattern
at perpendicular incidence transforms into an ellipse, with
minima in the intensities arising at higher incidence angles due
to certain domains in the condition such that a cancellation of
the magnetic contrast occurs. Comparing the right-hand side
(RHS) of the diffraction images (no closure domains) with
the left-hand side (LHS) of the diffraction images (35 nm
triangular closure domains), we note that there are some subtle
changes in the position in q space of the zones of magnetic
cancellation, and, as with the stripe domain case, the principal
changes are observed at higher diffraction angles. Higher-order
diffraction is weaker in the worm domain case compared
with the stripe domain case due to the absence of Bragg
amplification, and it could not be observed experimentally.
However, the fine measurement of the angular positions of
the observed diffraction minima caused by critical angles was
possible, and it is discussed in the following section.
III. EXPERIMENTAL DETAILS
The Co/Pd multilayers were prepared on a 3 mm × 3 mm
silicon nitride membrane of 50 nm thickness by a magnetron
sputtering system QPrep500 from MANTIS (in the case of
the stripe domain samples prepared at the TU Berlin) and
an INTERCOVAMEX system (for the worm domain samples
prepared at the Universidad de Santiago de Chile). The base
pressure in the chamber was less than 5 × 10−8 mbar, and the

094430-5
SAMUEL FLEWETT et al.
FIG. 6. Simulations of diffraction patterns at different incident
angles through a worm domain sample of 110 nm domain width
and 150 nm film thickness. Part (a) is normal incidence diffraction,
(b) is diffraction at 36◦ , (c) at 56◦ , and (d) at 70◦ . In each image, the
left-hand side is simulated with 35 nm triangular closure domains,
and the right-hand side without.
working pressure of argon was 2.2 × 10−3 mbar for the stripe
domain samples and 4.0 × 10−3 mbar for the worm domain
samples. For each case of stripe and worm domains, three mul-
tilayer samples were prepared, namely Co(0.5 nm)/Pd(0.8 nm)
with 50 repeats, Co(0.8 nm)/Pd(0.8 nm) with 50 repeats, and
Co(0.8 nm)/Pd(0.8 nm) with 100 repeats. For the first two
multilayers, a Ta(3 nm)/Pd(3 nm) seed layer was used, whereas
for the 100 repeat multilayer only a Pd(5 nm) seed layer
was used. On the top of all the samples, a 1.2 nm Pd layer
was deposited to prevent oxidation [40]. A selection of MFM
images of the samples is displayed in Fig. 1, and the real
sample thicknesses were determined by hard x-ray reflectivity
measurements to be within 5% of their nominal values. The
samples were subsequently subject to an ac demagnetization
procedure with around 50 sweeps of an external magnetic field
in an exponentially decaying fashion from 1.2 T down to zero.
This demagnetization was performed in-plane for the stripe
domains, and out-of-plane for the worm domains. Closure
domains are expected to grow with increasing film thickness
[27], from barely present for the thinnest films to well defined
and substantial in size for thicker films.
Diffraction patterns were measured during two synchrotron
beamtimes, with the experimental setup shown schematically
in Fig. 7, and the energy was tuned to near the L3 edge of Co at
the point where the magnetic scattering signal was maximized.
The stripe domain samples were measured at the CXS
endstation of beamline UE49-SGM of BESSY II in Berlin,
Germany [41], and the worm domain samples at the U11A-
PGM beamline of the Brazilian Synchrotron Light Laboratory
094430-6
PHYSICAL REVIEW B 95, 094430 (2017)
FIG. 7. Schematic diagram of the experiment, with the sample
rotated about the axis of the stripe domains.
(LNLS) in Campinas, Brazil. The instrument in Berlin is a new
high-brightness endstation, optimized for soft x-ray diffraction
and coherent x-ray scattering experiments. In Brazil, to com-
pensate for the lower flux of the second-generation source, we
used longer exposure times. The CXS chamber at the BESSY
II endstation is equipped with a translation table, which can
move by 1 cm in all three Cartesian directions, and a movable
CCD camera, which can cover 20◦ in the radial and 40◦ in the
azimuthal directions, respectively. At LNLS, the endstation
offers a similar range of camera movement. The samples in
both experiments were mounted on a rotation stage, and in
the case of stripe domains the rotation axis was set parallel to
the axis of stripe orientation as represented in Fig. 7. In this
configuration, we were able to measure diffraction patterns
at incidence angles of up to 76◦ from normal incidence. The
diffraction data from both experiments were recorded on a
Princeton Instruments PI-MTE CCD camera, with 13.5 μm
pixel size (2048 pixels) and a sample-camera distance of 27 cm
at BESSY II and 11 cm at LNLS. Measurements were made
in angular steps ranging from 5◦ near normal incidence down
to 0.3◦ near critical angles where the qualitative nature of
the diffraction pattern displayed a stronger dependence on the
angle. The incident photon flux was varied by adjusting the
beamline slits to allow the minimum exposure time possible,
and therefore it was at a maximum for the highest incidence
angle measurements. This act of opening the beamline slits to
allow greater photon flux also reduced the spatial and temporal
coherence of the radiation, thus precluding the use of phase
retrieval techniques for wavefield reconstruction and removing
any benefit of using circularly polarized light as opposed to
linearly polarized light [37].
IV. EXPERIMENTAL RESULTS—STRIPE DOMAIN CASE
Given that the primary aim of this work was to identify
the contributions of in-plane magnetized domains in a pre-
dominantly out-of-plane system, it was necessary to measure
to as high an incidence angle as possible. According to the
simulations in Fig. 5, there is little difference at 36◦ incidence
angle between the cases with and without closure domains,
however at the critical angle at 56◦ there is a marked difference
between the two. To illustrate this effect with experimental
THREE-DIMENSIONAL CHARACTERIZATION OF Co/Pd . . . PHYSICAL REVIEW B 95, 094430 (2017)

FIG. 8. Experimental diffraction images at the critical angles in (a) at 54.3◦ for the Co(0.8 nm)/Pd(0.8 nm) with 100 repeats sample
(slightly below the critical angle), and (b) at 66◦ (LHS) and 67.5◦ (RHS) for the Co(0.5 nm)/Pd(0.8 nm) with 50 repeats sample. In each plot,
the third-order diffraction peak is marked with a circle. Due to Ewald sphere curvature for the case of the thin sample (b), in order to show
critical angle behavior on both sides, it was necessary to splice two images at slightly different incident angles.

data, we compare the measured diffraction from two samples
as they are defined in Sec. III, the first being 100 repeats of
Co(0.8 nm)/Pd(0.8 nm), and the second being 50 repeats of
Co(0.5 nm)/Pd(0.8 nm). The thicker sample has approximately
110 nm domains, and a critical angle near 55◦ , whereas the
thinner one has 55 nm domains and a critical angle near 67◦ . In
Fig. 8 and in the corresponding lineout plots at angles on and
near the critical angle shown in Fig. 9, we contrast diffraction
from the two samples on a logarithmic intensity scale. In all
cases, a split first-order peak is observed as expected at the
critical angle, however the most important difference between
the two figures is the different behavior observed with respect
to the third-order diffraction peak. In the case of the thick
sample on the RHS, the diffraction peak exhibits an enhanced
intensity, is not split, and remains relatively constant in q as the
incidence angle is varied over the critical angle. For the thin
sample, the third-order peak on the RHS is of weaker intensity,
and is observed to “jump” to higher q as the incidence angle is
increased across the critical angle, in a manner similar to that
observed for the first-order peak. In Fig. 8(a), on the LHS, the
third-order peak is only just visible as a weak scattering peak
at a q value of 24 μm−1 , this contrast with the RHS being
due to a longer light path within the sample and therefore
greater absorption of the scattered light on the LHS compared
with the RHS. For the case of the LHS of Fig. 8(b), the
third-order peak is obstructed by part of the experimental
apparatus. Readers should also note the positional asymmetry
present in the diffraction patterns due to curvature of the
Ewald sphere (similar to the effect of oblique illumination
of a diffraction grating), an effect of ever greater importance at
higher incidence angles and q. At higher incident angles, we
observed the presence of wide vertical stripes with spacing in
reciprocal space inversely proportional to the thickness of the
sample. These stripes are also observed in diffraction patterns
measured at off-resonant energies (760 eV) where magnetic
scattering is absent, and can thus be considered a form of
background signal for the purposes of this work.
To estimate the size of the closure domains, we compare
the experimental lineout plots at incidence angles close to
the critical angle in Figs. 9(a) and 9(b) with simulations in
Figs. 9(c) and 9(d), where 35 and 7.5 nm triangular closure
domains of equal width and depth were used, respectively.
These closure domain sizes were found to provide the best fit
between the model and the experiment, up to the limitations of
the current model. The quality of fit for the thick sample was
notably better than for the thin sample, which indicates that

094430-7
SAMUEL FLEWETT et al. PHYSICAL REVIEW B 95, 094430 (2017)

FIG. 9. Lineout plots of the diffraction (on a logarithmic intensity scale) observed at incidence angles close to the critical angle where
the incident light traverses exactly two magnetic domains of opposite orientation. Parts (a) and (b) show the experimental plots for the
Co(0.8 nm)/Pd(0.8 nm) sample with 100 repeats and the Co(0.5 nm)/Pd(0.8 nm) sample with 50 repeats, respectively, and (c) and (d) are
the corresponding simulations. The simulations were generated with 35 nm closure domains for the Co(0.8 nm)/Pd(0.8 nm) sample with 100
repeats and the 7.5 nm closure domains for the Co(0.5 nm)/Pd(0.8 nm) sample with 50 repeats, using estimated magnetic sample thicknesses
of 150 and 45 nm, respectively. The arrows mark the “jumping” of the third-order diffraction peak as the incidence angle is moved across the
critical angle, and the order of each diffraction peak is marked in the figure.

the model of triangular closure domains has greater validity
for thicker rather than thinner samples.
Another observation was that the experimental critical
angles are shifted upward compared to their expected values
calculated from Eq. (3) using the average domain sizes
measured from the scattering data and the sample thicknesses.
This shift is from 59◦ to 67◦ in the case of the thin sample, and
from 54◦ to 57◦ for the thick sample. This effect, especially
strong for the thin sample, is suggestive of a magnetically
thinner sample than was fabricated (45 nm as observed versus
65 nm fabricated). The reasons for this observation are not
yet understood and are beyond the simplified model of the
magnetization used in this article. Work is underway to
upgrade the model of the magnetization, and higher precision
scattering data have also been collected, hopefully permitting
a quantitative fit of the upgraded model to the data.
At incidence angles above the first critical angle, the first-
order diffraction peak was observed to continue moving as
expected toward higher q. Subsequent critical angles reached
in the case of the thick sample at 67◦ and 76◦ correspond to
where four and six primary domains are traversed, however
for the thin sample such higher-order critical angles are not
observed due to hardware limitations. The incidence angles at
which the higher-order critical angles occur is consistent with
an increase of observed magnetic thickness with incidence
angle, consistent with an increased contribution of in-plane
magnetization at higher incidence angles. Data collected in
the most recent experiment were obtained at incidence angles
of up to 85◦ , which will allow for a full investigation of these
observations to be reported in our subsequent article.
V. EXPERIMENTAL RESULTS—WORM DOMAIN CASE
From the simulations in Fig. 6, we note that as was the case
for the stripe domains, the effect of the closure domains begins
to become visible at higher angles of incidence, in the form of
shifts in q space of the observed diffraction minima. Therefore,
we show in Fig. 10 the measured diffraction (on a logarithmic
intensity scale) observed at a 65◦ and 70◦ incidence angle,
respectively, on the left and right sides of each image for each
of the thick and thin samples. [For this measurement, the thick
sample was the same as previously, however the thin sample
was 50 repeats of Co(0.8 nm)/Pd(0.8 nm).] For the thicker
sample in this case, the degree of spatial disorder was greater

094430-8
THREE-DIMENSIONAL CHARACTERIZATION OF Co/Pd . . . PHYSICAL REVIEW B 95, 094430 (2017)

FIG. 10. Diffraction images for the worm domain samples. Each image is a montage of diffraction images at 65◦ and 70◦ incidence angles,
with the 65◦ image on the left and the 70◦ image on the right. In each case, the intensity color scale is logarithmic, and the q vectors of the
diffraction minima for the 70◦ case are marked in the figure. Part (a) shows the experimental diffraction pattern from the Co(0.8 nm)/Pd(0.8 nm)
sample with 50 repeats, (b) shows the diffraction simulated for 18 nm triangular closure domains and a magnetic sample thickness of
65 nm, whereas (c) shows the diffraction pattern simulated without closure domains and the full nominal sample thickness of 80 nm. Part
(d) shows the experimental diffraction for the Co(0.8 nm)/Pd(0.8 nm) sample with 100 repeats, while (e) shows the simulated diffraction pattern
including 35 nm triangular Closure Domains, and a magnetic sample thickness of 150 nm. Part (f) shows the diffraction simulated without
closure domains, and with the full 160 nm nominal sample thickness.

than for the simulations in Fig. 6, resulting in a diffraction
signal observed over a greater region in reciprocal space than
in Fig. 6, with intensity minima due to some domains being at
critical angles noted. For the thinner sample, less variation
in domain size is present, and the magnetic diffraction is
observed to exhibit an elliptical ring [Fig. 10(a)], similar to the
case of the earlier simulations. Simulations of the diffraction
(this time with a greater degree of domain size variation) are
shown in Figs. 10(b) and 10(c) for the thin sample and in
Figs. 10(e) and 10(f) for the thick sample. The simulations in
Figs. 10(b) and 10(e) are generated using the set of parameters
found to provide the best qualitative fit to the experimental
data: These are triangular closure domains of 18 nm width
for the thinner sample and 36 nm for the thicker sample, and a
reduced magnetic sample width of 65 and 145 nm, respectively.
For comparison, Figs. 10(c) and 10(f) were generated using
the sample thicknesses as fabricated (80 and 160 nm) and
neglecting the presence of closure domains. In Figs. 10(c) and
10(f), we note that the minima caused by the critical angles
are shifted in q compared with Figs. 10(b) and 10(e), but not
in a uniform manner. The first minima occur at higher q in
the experimental case of Fig. 10(d) compared with Fig. 10(f),
but higher-order minima are observed to occur at lower q in
the experimental case [Fig. 10(d)] and the theoretical case
where the domain walls are included [Fig. 10(e)]. Such shifts
in the minima were of the order of 10%, as noted by q-space
positions of the minima marked directly on the text. In each
of the theoretical diffraction images, the 70◦ simulation was
enhanced in intensity by a factor of 10 to allow the wider
spread in q space of the observed diffraction to be displayed
alongside the 65◦ incident angle image.
As with the stripe domain case, an asymmetry in the
diffraction intensity was observed between the different halves
of the diffraction patterns due to within-sample optical path
length differences, with the asymmetry being top/bottom
rather than left/right due to the sample rotation being about
the horizontal axis at the LNLS. The asymmetry was weaker
than observed for stripe domains, due to the fact that
such asymmetry increases with q, and that only first-order
diffraction was observed for the worm domain samples. The
asymmetry in the q positions of the diffraction minima and
maxima does, however, exhibit a different pattern compared
with the stripe domains, due to these minima and maxima
no longer representing different orders of a grating-type
diffraction pattern, but rather first-order diffraction and the
corresponding minima caused by the critical angles from

094430-9
SAMUEL FLEWETT et al.
domains of different sizes under simultaneous illumination.
For both the stripe and worm domains, such asymmetries were
present in the simulation results.
VI. CONCLUSION
From this work, we can make three general conclusions.
(i) Three-dimensional structures in perpendicular magnetic
anisotropy thin films can be seen using transmission geometry
scattering, with critical angles at which the primary domain
structure cancels the XMCD signal, allowing for the effects
of a higher-order domain structure to be readily studied.
(ii) As the angle of incidence is increased, it becomes necessary
to model the diffraction using models that account for the
possibility of multiple scattering. This is computationally
expensive, and it has implications for the development of
tomographic reconstruction algorithms. (iii) Magnetic sample
thickness estimations: In all cases, the observed magnetic
sample thickness of the primary domains was estimated from
the first critical angle to be less than the nominal as fabricated
(and confirmed by hard x-ray reflectivity) sample thickness.
[1] A. Hubert and R. Schäfer, Magnetic Domains (Springer, Berlin,
2000).
[2] I. K. Schuller, S. Kim, and C. Leighon, J. Magn. Magn. Mater.
200, 571 (1999).
[3] S. S. P. Parkin, M. Hayashi, and L. Thomas, Science 320, 190
(2008).
[4] D. A. Allwood, G. Xiong, C. C. Faulkner, D. Atkinson, D. Petit,
and R. P. Cowburn, Science 309, 1688 (2005).
[5] N. Nagaosa and Y. Tokura, Nat. Nanotech. 8, 899 (2013).
[6] F. Büttner, C. Moutafis, M. Schneider, B. Krüger, C. M. Günther,
J. Geilhufe, C. v. Kroff Schmising, J. Mohanty, B. Pfau, S.
Schaffert, A. Bisig, M. Foerster, T. Schulzm, C. A. F. Vaz, J. H.
Franken, H. J. M. Swagten, M. Klaüi, and S. Eisebitt, Nat. Phys.
11, 225 (2015).
[7] W. J. an P. Upadhyaya, W. Zhang, G. Yu, M. B. Jungfleisch,
F. Y. Fradin, J. E. Pearson, Y. Tserkovnyak, K. L. Wang, O.
Heinonen, S. G. E. te Velthuis, and A. Hoffmann, Science 349,
283 (2015).
[8] G. Chen, J. Zhu, A. Quesada, J. Li, A. T. N’Diaye, Y. Huo,
T. P. Ma, Y. Chen, H. Y. Kwon, C. Won, Z. Q. Qiu, A.
K. Schmid, and Y. Z. Wu, Phys. Rev. Lett. 110, 177204
(2013).
[9] G. Chen, A. T. N’Diaye, S. P. Kang, H. Y. Kwon, C. Won, Y. Wu,
Z. Q. Qiu, and A. K. Schmid, Nat. Commun. 6, 6598 (2015).
[10] A. W. Rushforth, P. C. Main, B. L. Gallagher, C. H. Marrows,
B. J. Hickey, E. D. Dahlberg, and P. Eames, J. Appl. Phys. 89,
7534 (2001).
[11] W. Wulfhekel and J. Kirschner, Annu. Rev. Mater. Res. 37, 69
(2007).
[12] J. Sthöer, H. A. Padmore, S. Anders, T. Stammler, and M. R.
Scheinfein, Surf. Rev. Lett. 5, 1297 (1998).
[13] A. Bisig, M. Stärk, M.-A. Mawass, C. Moutafis, J. Rhensius,
J. Heidler, F. Büttner, M. Noske, M. Weigand, S. Eisebitt, T.
094430-10
PHYSICAL REVIEW B 95, 094430 (2017)
In the medium term, we hope to achieve improved agree-
ment between theory and experiment, both as a result of the
higher precision experimental data and through the application
of more sophisticated theoretical models. The development of
soft x-ray ptychography for such magnetic systems such as
the one studied in this article is currently only limited by a
lack of coherent photon flux, which in the next few years will
be helped by the opening of the new facilities MAX IV and
SIRIUS.
ACKNOWLEDGMENTS
S.F., L.N., and A.O. are grateful for funding from the
Chilean Research Organization CONICYT under FONDE-
CYT Iniciación Grant No. 11130563. We acknowledge the
Helmholtz Zentrum Berlin for use of the synchrotron fa-
cilities BESSY II. J.E. and J.C.D. acknowledge financial
support from FONDECYT 1150952, 1140195, and from
CONICYT Proyecto Basal FB0807, CEDENNA. S.O. grate-
fully acknowledges PAI-CONICYT Grant No. 79140036 and
CONICYT Proyecto Basal USA1555 for financial support. We
acknowledge LNNano/CNPEM for the facilities for hard x-ray
reflectivity and MFM.
Tyliszczak, B. Van Waeyenberge, H. Stoll, G. Schütz, and M.
Kläui, Nat. Commun. 4, 2328 (2013).
[14] H. A. Dürr, E. Dudzik, S. S. Dhesi, J. B. Goedkoop, G. van der
Laan, M. Belakhovsky, C. Mocuta, A. Marty, and Y. Samsonl,
Science 284, 2166 (1999).
[15] C. Blanco-Roldán, C. Quirós, A. Sorrentino, A. Hierro-
Rodrı́guez, L. Álvarez Prado, R. Valcárcel, M. Duch, N. Torras,
J. Esteve, J. Martı́n, M. Vélez, J. Alameda, E. Pereiro, and S.
Ferrer, Nat. Commun. 6, 8196 (2015).
[16] O. Boulle, J. Vogel, H. Yang, S. Pizzini, D. de Souza Chaves,
A. Locatelli, T. O. Mentes, A. Sala, L. D. Buda-Prejbeanu, O.
Klein, M. Belmeguenai, Y. Roussigne, A. Stashkevich, S. M.
Cherif, L. Aballe, M. Foerster, M. Chshiev, S. Auffret, I. M.
Miron, and G. Gaudin, Nat. Nanotech. 11, 449 (2016).
[17] S. Fin, R. Tomasello, D. Bisero, M. Marangolo, M. Sacchi, H.
Popescu, M. Eddrief, C. Hepburn, G. Finocchio, M. Carpentieri,
A. Rettori, M. G. Pini, and S. Tacchi, Phys. Rev. B 92, 224411
(2015).
[18] H. Popescu, F. Fortuna, R. Delaunay, C. Spezzani, V. Lopez-
Flores, N. Jaouen, and M. Sacchi, Appl. Phys. Lett. 107, 202404
(2015).
[19] R. Streubel, F. Kronast, P. Fischer, D. Parkinson, O. G. Schimdt,
and D. Makarov, Nat. Commun. 6, 7612 (2015).
[20] C. Donnelly, V. Scagnoli, M. Guizar-Sicairos, M. Holler, F.
Wilhelm, F. Guillou, A. Rogalev, C. Detlefs, A. Menzel,
J. Raabe, and L. J. Heyderman, Phys. Rev. B 94, 064421
(2016).
[21] C. Donnelly, Hard X-ray Magnetic Tomography of Micrometre-
sized Magnetic Systems, ALS User Meeting Work-
shop: Nanotomography of vector fields, October 5 2016
(unpublished).
[22] D. Wolf, L. A. Rodriguez, A. Béché, E. Javon, L. Serrano, C.
Magen, C. Gatel, A. Lubk, H. Lichte, S. Bals, G. V. Tendeloo,
THREE-DIMENSIONAL CHARACTERIZATION OF Co/Pd . . .
A. Fernández-Pacheco, J. M. D. Teresa, and E. Snoeck, Chem.
Mater. 27, 6771 (2015).
[23] D. Navas, C. Redondo, G. A. Badini Confalonieri, F. Batallan,
A. Devishvili, O. Iglesias-Freire, A. Asenjo, C. A. Ross, and
B. P. Toperverg, Phys. Rev. B 90, 054425 (2014).
[24] I. Manke, N. Kardjilov, R. Schäfer, A. Hilger, M. Strobl, M.
Dawson, C. Grünzweig, G. Behr, M. Hentschel, C. David et al.,
Nat. Commun. 1, 125 (2010).
[25] C. Donnelly, M. Guizar-Sicairos, V. Scagnoli, M. Holler, T.
Huthwelker, A. Menzel, I. Vartiainen, E. Müller, E. Kirk, S.
Gliga, J. Raabe, and L. J. Heyderman, Phys. Rev. Lett. 114,
115501 (2015).
[26] D. A. Shapiro, Y. Yu, T. Tyliszczak, J. Cabana, R. Celestre,
W. Chao, K. Kaznatcheev, A. L. D. Kilcoyne, F. Maia, S.
Marchesini, Y. S. Meng, T. Warwick, L. L. Yang, and H. A.
Padmore, Nat. Photon. 8, 765 (2014).
[27] O. Hellwig, G. P. Denbeaux, J. B. Kortright, and E. E. Fullerton,
Physics B 336, 136 (2003).
[28] S.-K. Kim and J. B. Kortright, Phys. Rev. Lett. 86, 1347
(2001).
[29] U. Rüdiger, J. Yu, L. Thomas, S. S. P. Parkin, and A. D. Kent,
Phys. Rev. B 59, 11914 (1999).
[30] E. Dudzik, S. S. Dhesi, H. A. Dürr, S. P. Collins, M. D. Roper,
G. van der Laan, K. Chesnel, M. Belakhovsky, A. Marty, and Y.
Samson, Phys. Rev. B 62, 5779 (2000).
094430-11
PHYSICAL REVIEW B 95, 094430 (2017)
[31] N.-T. D. Loh, S. Eisebitt, S. Flewett, and V. Elser, Phys. Rev. E
82, 061128 (2010).
[32] C. R. Maurer, R. Qi, and V. Raghavan, IEEE Trans. Pattern Anal.
Mach. Intell. 25, 265 (2003).
[33] J. P. Hannon, G. T. Trammell, M. Blume, and D. Gibbs, Phys.
Rev. Lett. 61, 1245 (1988).
[34] B. Henke, E. Gullikson, and J. Davisv, At. Data Nucl. Data
Tables 54, 181 (1993).
[35] C. Boeglin, E. Beaurepaire, V. Halté, V. López-Flores, C.
Stamm, N. Pontius, H. Dürr, and J.-Y. Bigot, Nature (London)
465, 458 (2010).
[36] E. Wolf, J. Opt. Soc. Am. 72, 343 (1982).
[37] S. Flewett, S. Schaffert, J. Mohanty, E. Guehrs, J. Geilhufe,
C. M. Günther, B. Pfau, and S. Eisebitt, Phys. Rev. Lett. 108,
223902 (2012).
[38] A. Tripathi, J. Mohanty, S. H. Dietze, O. G. Shpyrko, E. Shipton,
E. E. Fullerton, S. S. Kim, and I. McNulty, Proc. Natl. Acad.
Sci. USA 108, 13393 (2011).
[39] J. J. Turner, X. Huang, O. Krupin, K. A. Seu, D. Parks, S. Kevan,
E. Lima, K. Kisslinger, I. McNulty, R. Gambino, S. Mangin, S.
Roy, and P. Fischer, Phys. Rev. Lett. 107, 033904 (2011).
[40] O. Hellwig, A. Berger, J. B. Kortright, and E. E. Fullerton,
J. Magn. Magn. Mater. 319, 13 (2007).
[41] D. Engel, D. Mishra, and S. Eisebitt, J. Large-Scale Res. Facil.
2, A56 (2016).