-
Journal of Magnetism and Magnetic Materials 264 (2003) 1429
Memory effects and slow dynamics in ultra thin magnetic
films
S.P. Bromleya, J.P. Whiteheada, K. DeBellb,*, A.B. MacIsaacc
aDepartment of Physics and Physical Oceanography, Memorial
University of Newfoundland, St. Johns, Newfoundland, Canada A1B
3X7bDepartment of Mathematical Sciences, University of New
Brunswick at Saint John, Saint John, New Brunswick, Canada E2L
4L5
cDepartment of Applied Mathematics, University of Western
Ontario, London, Ont., Canada N6A 5B9
Received 17 September 2002; received in revised form 17 January
2003
Abstract
Monte Carlo studies of a uniaxial spin system on a square
lattice with both ferromagnetic exchange interactions and
dipolar interactions reveal three distinct dynamical regimes at
low temperatures. In the first regime the magnetization
decays through nucleation and growth of spin islands, in the
second regime, despite the magnetization being effectively
zero, there is a persistent memory effect most clearly revealed
by the asymmetry between the numbers of up and down
islands. The third regime is characterized by slow
crystallization of the equilibrium smectic phase from the
quenched
polycrystalline state.
r 2003 Elsevier Science B.V. All rights reserved.
PACS: 75.70; 75.60.L
Keywords: Memory effects; Magnetic; Ultra-thin film; Time
dependence; Dynamics
1. Introduction
Ultrathin magnetic films are of considerableinterest because of
their potential technologicalapplications in data storage media and
magneto-electronics [1]. While the rapid technologicaldevelopment
of these materials is almost concur-rent with the investigation of
the underlyingphysics [2], many aspects of these system remainto be
elucidated. In particular, a more completeunderstanding of the
stability and dynamics of thedomain structures, that determine many
of the
important properties of these materials, is highlydesirable.
A key feature of these materials is the inherentmagnetic
frustration which results from thecompetition between the
(ferromagnetic) exchangeinteraction, the dipolar interaction and
the surfaceanisotropy [3]. This frustration can result indomain
phases and, as a result, these materialscan exhibit a variety of
phase transitions. At lowtemperatures the system is typically
mono-domainwith spins oriented perpendicular to the plane ofthe
film, as the temperature is raised a smecticmulti-domain regime
occurs and this is usuallyevidenced by a loss of the net magnetic
moment.The static, equilibrium properties of these filmshave been
studied extensively both experimentallyand theoretically (for a
review see Ref. [4]).
ARTICLE IN PRESS
*Corresponding author. Department of Mathematics,
University of New Brunswick at Saint John, Saint John,
New Brunswick, Canada E2L 4L5. Tel.: +506-648-5615; fax:
+506-648-5650.
0304-8853/03/$ - see front matter r 2003 Elsevier Science B.V.
All rights reserved.
doi:10.1016/S0304-8853(03)00131-8
-
Dynamical properties, such as relaxation phe-nomena, have been
much less studied, howeverseveral significant experimental studies
have beenperformed. Berger and Hopster have examinedthe relaxation
of the magnetization fromsaturation in Fe/Ag(1 0 0) [5]. Their
experimentswere performed in the multi-domain regime.Berger and
Hopster observed that relativelysmall external fields E90 Oe were
required tosaturate the magnetization of the system;however, after
the field was switched off, severalseconds were required for the
magnetization tovanish at lower temperatures. It was arguedthat the
dynamics of these systems could becharacterized in terms of a
single activation energyat low temperatures. However, Berger and
Hop-ster found that the activation energy variedsignificantly
between samples. Cowburn and cow-orkers have studied dynamical
effects in anultrathin wedge of Fe between thin layers of Ag[6] and
found evidence for a thermally nucleateddomain structure. Venus et
al. [7] have alsodiscussed the importance of energy barrier
domi-nated dynamics in interpreting the results of lowfrequency AC
susceptibility measurements on Fe/2ML Ni/W(1 1 0).
Here we report the results of a Monte Carlostudy of the dynamics
of a system of Ising spins ona square lattice in which the spins
are alignedperpendicular to the plane of the lattice. Specifi-cally
we examine the case in which the system isprepared as a single
saturated (ferromagnetic)domain and is then allowed to relax to
theequilibrium phase.
At low temperatures, our results demonstratethat there are three
distinct dynamicalregimes during the relaxation of this system.
Atearly times the relaxation process is characterizedby nucleation
and growth of islands of spinswith the opposite orientation to the
initialmagnetization and the corresponding decay ofthe
magnetization.
In the intermediate regime the magnetizationhas dropped to a few
percent of its saturationvalue, however the system is not in
itsequilibrium state, since the number of islands ofup spins is
observed to differ significantly fromthe number of islands of down
spins. In addition
to, and closely correlated with, the asymmetrybetween the spin
up and spin down islands,a small remnant of the
magnetizationpersists.
This asymmetry and the associated magnetiza-tion decay on a much
longer time scale thanthe initial decay of the magnetization
associatedwith the island nucleation and growth. In thissense, the
system retains a memory of its initialstate which gradually decays
in this intermediateregime.
At late times the spin up and spin down islandsarrange to form
regions consisting of the stripes ofalternating spins that
characterize the equilibriumsmectic phase. However the regions are
notoriented along a single axis but instead areanalogous to
crystallites in a polycrystallinesample. The dynamical processes
which result inthe equilibrium single smectic crystal phase
areextremely slow.
The possibility of a relatively long lived memoryeffect in
ultrathin magnetic films is intriguing.First, in experimental
studies to probe thedynamics it is important to be aware
thatrelaxation to a system with close to zero magne-tization does
not necessarily imply that the systemhas reached a state that is
close to the equilibriumstate. Moreover the existence of more than
onetime scale in the non-equilibrium regime hasimportant
consequences for the interpretation ofthe dynamical results. As we
show below, mea-surements of the relaxation time even for
themagnetization may be a useful diagnostic toolwhen probing the
nature of the multi-domainstate. Second, analogous memory effects
inalloys have potential technological applications(however the
nature of the long rangeinteraction is different in those systems).
While itis not clear if the memory effect reported here hasdirect
technological uses, understanding such aneffect is likely to be of
importance in theapplication of ultrathin magnetic films in
devices.The very slow dynamics observed in the thirdregime, in
addition to their intrinsic interest in thecontext of magnetic
films, may provide insight intothe more general problem of slow
dynamics in, forexample, solids with a crystalline
equilibriumphase.
ARTICLE IN PRESS
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 1429 15
-
2. The simulation model
The interaction Hamiltonian for the model isgiven by
H JX/ijS
sisj gXiaj
sisjR3ij
; 1
where the variable s 71 denotes the direction ofthe magnetic
moment perpendicular to the planeof the film. The first term
represents the effect ofthe exchange interaction, with exchange
constantJ: The exchange interaction is assumed to beferromagnetic
and J is therefore positive. Thesecond term arises as a consequence
of the dipolarinteraction.
Earlier simulation studies of the static propertiesof this
system revealed that it exhibits a multi-domain smectic phase at
low temperatures. At acritical temperature TO the smectic phase
isreplaced by a tetragonal phase and the orienta-tional order
associated with the smectic phase islost [8]. In subsequent
experimental studies thistransition was directly observed by direct
imagingof the magnetic domains [9]. A transition from
anorientationally ordered phase to a phase withoutorientational
order is also observed in simulationsof the corresponding
triangular lattice system[10,11]. However the exact nature of the
phasesand transitions is dependent on the lattice.
A value of J 8:9g is used throughout the workreported here
corresponding to a ground statestripe width of h 8 and a transition
from thesmectic striped phase to the tetragonal phase atTOE4:8g
[8]. A square L L lattice is used withperiodic boundary conditions.
L is chosen to be amultiple of h in order that the stripes will
becommensurate with the periodicity imposed by theboundary
conditions. The dipolar interaction istreated using the Ewald
summation technique [4].The simulations are performed using
standardGlauber dynamics and the initial spin configura-tion is
fully polarized fsigt0 f1g: As discussedbelow, the initial decay of
the magnetization isstrongly dependent on the size of the system
and asystem size of 256 256 appears to be theminimum system size
for reliable dynamical results[12]. All of the calculations
described in this workare performed on lattices of size N 256
256:
3. Relaxation kinetics
In these simulations the system is allowed torelax from its
initial saturated ferromagnetic statestarting at time t 0 toward
its equilibrium phase.Fig. 1 shows the spin configurations of a 256
256lattice at various times for a typical simulation runat T 4:25g:
This is below the critical temperatureTOE4:80g: For this value of T
the equilibriumphase therefore corresponds to the smectic
phaseconsisting of stripes of alternating spin orientedalong a
common crystallographic axis, with no netmagnetization.
The magnetization for several simulation runs atT 4:25g is
plotted as a function of time t inFig. 2, together with the average
(over the simula-tion runs) of the magnetization. The time t is
givenin units of Monte Carlo steps (mcs).
Since the initial saturated ferromagnetic state(all spins up) is
a metastable state of the system,the initial stage of the
relaxation process is verysimilar to other thermally activated
nucleationprocesses. Specifically our results show that at
veryearly times, many islands, each consisting of asmall number of
reversed spins (spin down), formthroughout the system. However,
there exists anenergy barrier preventing evolution towards
theequilibrium state. This energy barrier, limiting thegrowth of
the islands, is overcome only oncethe size of the islands exceeds
some critical value.We refer to the time during which the
systemremains on the metastable side of the energybarrier as the
wait time.
In practice, we use the time taken for themagnetization to decay
by 1% as providing somequantitative measure of the wait time. As
therelaxation process is stochastic, the wait times willvary for
different simulation runs at the sametemperature. The histogram of
wait times isapproximately a Poisson distribution for systemsof
size 256 256; however the distribution devi-ates significantly from
this if smaller system sizesare used [12]. Earlier work also
demonstrated thesensitivity of results for this system to system
size,however, the conclusions of this earlier work werestill based
on simulations of relatively smallsystems [13]. This combined with
the unusualboundary conditions employed in that work
ARTICLE IN PRESS
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 142916
-
makes the usefulness of conclusions drawn in thatearlier work
somewhat uncertain. Subsequentwork which also concentrated on the
effect ofthe ratio J=g (in our notation) used larger latticesbut
truncated the dipolar interaction [14]. Tomitigate system size
effects, all of the simulations
described here use a minimum system size of N 256 256:
The effect of the nucleation process on themagnetization is
manifest in the initial slopeof the magnetization curve shown in
Fig. 2(limt-0dM=dtE0).
As the number of islands that exceed the criticalsize increases,
the islands begin to grow and themagnetization decays rapidly. This
rapid decay ofthe magnetization following the initial
nucleationregime is clearly seen in Fig. 2. In Fig. 3 we showthe
average magnetization temperatures. The plotsshow the decay rate of
the magnetization increas-ing with increasing temperature.
However, while the magnetization in Fig. 2appears to decay to
effectively zero after aboutt 2000 mcs; it is quite obvious from
the spinconfigurations shown in Fig. 1, that the system isstill far
from its equilibrium phase. In addition acareful examination of the
magnetization datashows that m has not decayed to zero, but
insteada small remnant magnetization persists; thisremnant
magnetization decays over a much longertime scale.
Inspecting the spin configuration in Fig. 1, fort 2000mcs we see
that the spin configurationconsists of a large number of
disconnected spin
ARTICLE IN PRESS
1.0
0.8
0.6
0.4
0.2
0.0
mag
netis
atio
n
2000150010005000t (mcs)
T = 4.25g
Fig. 2. A plot of the magnetization as a function of Monte
Carlo time steps (mcs) for several runs. The parameters are
J 8:9g; T 4:25g and L 256: The solid line shows theaverage
magnetization.
t = 0 mcs t = 250 mcs t = 750 m
t = 2000 mcs t = 20000 mcs t = 30000
Fig. 1. Snapshots of the spin configuration at successive times
for a typical Monte Carlo run, with J 8:9; T 4:25 and L 256:
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 1429 17
-
down islands within a single connected island ofup spins. This
asymmetry between the number ofspin up and spin down islands serves
as a veryuseful quantitative measure of the memory thesystem has of
its initial configuration. We there-fore find it useful to define
the island asymmetry asthe difference between the number of spin up
andspin down islands, which we denote by a: Theisland asymmetry may
be elegantly and efficientlycalculated by generating the winding
number for aparticular spin configuration. The details havebeen
given previously [15].
The average island asymmetry is plotted forseveral temperatures
in Fig. 4. All the curves showthe same qualitative behaviour of a
rapid rise inthe asymmetry followed by a slow decay.
The time dependence of the island asymmetryprovides some useful
insights into the relaxationprocess. The rapid initial growth of
the islandasymmetry seen in Fig. 4, is directly related to
thenucleation of islands, since to begin with the islandasymmetry
is approximately equal to the numberof isolated spin reversed
islands. This is readilyunderstood, since at the earliest times the
spin upsites, except for short lived individual spin flips,are all
connected to form a single island which,because of the periodic
boundary conditions, doesnot contribute to the parameter a; the
island
asymmetry is therefore given by the number ofspin down
islands.
As the islands grow the process of islandnucleation abates and
the spin down islands startto coalesce to form larger more
elongated islands.At this point the island asymmetry reaches
amaximum and begins to decrease. As the spindown islands begin to
coalesce, not only do thenumber of spin down islands begin to fall,
but alsothe spin up sites are beginning to becomedisconnected
forming isolated pools surroundedby spin down sites, thus the
island asymmetrybegins to fall. A comparison of the time scales
inFigs. 2 and 4 shows that the time scale associatedwith the decay
of the island asymmetry is at leastan order of magnitude greater
than that associatedwith the decay of the magnetization. This
slowdynamic associated with the decay of the islandasymmetry
defines the intermediate regime.
While the slow decay of the island asymmetryserves to define the
intermediate regime, a carefulexamination of the magnetization data
in thisregime shows that it does not fluctuate about itsequilibrium
value but instead a small remnant, ofthe order of 12% of the
saturation magnetization,persists long after the bulk of the
magnetizationhas disappeared.
This long lived tail is clearly seen in Fig. 5 inwhich the
magnetization is plotted on an expandedscale along with the island
asymmetry for several
ARTICLE IN PRESS
1.0
0.8
0.6
0.4
0.2
0.0
mag
netis
atio
n
10008006004002000t (mcs)
T=4.0g
T=4.25g
T=5.0g
T=6.0g
Fig. 3. A plot of the average magnetization as a function of
Monte Carlo time steps (mcs) for several temperatures for J 8:9g
and L 256:
100
80
60
40
20
0
isla
nd a
sym
mte
ry
20103151050
t (mcs)
T = 4.0g
T = 4.25g
T = 3.75g
T = 3.5g
Fig. 4. A plot of the island asymmetry as a function of
Monte
Carlo time steps (mcs) for several temperatures for J 8:9gand L
256:
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 142918
-
ARTICLE IN PRESS
40
30
20
10
0
(i
slan
d sy
mm
etry
)
6010350403020100t (mcs)
1510-3
10
5
0
m (m
ag
ne
tisatio
n)
T = 3.75g
60
50
40
30
20
10
0
(i
slan
d sy
mm
etry
)
6010350403020100t (mcs)
2510-3
20
15
10
5
0
m (m
agnetisation)
T = 4.0g
30
25
20
15
10
5
0
(i
slan
d sy
mm
etry
)
6010350403020100t (mcs)
1210-3
10
8
6
4
2
0
m (m
agnetisation)
T = 3.50g
800
600
400
200
0M (
tota
l mag
netis
atio
n)
30252015105
(island symmetry)
T = 3.75
500
400
300
200
100
0M (
tota
l mag
netis
atio
n)
222018161412108 (island symmetry)
T = 3.50
800
600
400
200
0
M (
tota
l mag
netis
atio
n)
3020100 (island symmetry)
T = 4.00
Fig. 5. A plot of the island asymmetry against time for several
temperatures with J 8:9g and L 256: Also shown is the tail of
themagnetization data (dots) on an expanded scale. The insets show
the island asymmetry as a function of the magnetization.
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 1429 19
-
temperatures. Indeed this graph not only showsthe long lived
tail but also that the magnetizationin the intermediate regime is
strongly correlated tothe island asymmetry. The correlation between
theisland asymmetry and the magnetization is moreclearly seen in
inset in each of the graphs in Fig. 5where a is plotted as a
function of the totalmagnetization M m L2; in the
intermediateregime. The data shows a linear relationshipbetween the
island asymmetry and the magnetiza-tion. This suggests that the
long lived tail in themagnetization is a feature of the low
temperaturerelaxation process and reflects the fact that thesystem
has not yet quite relaxed to its equilibriumstate.
Describing the linear relationship between theisland asymmetry a
and the total magnetization Min the intermediate regime by the
coefficients Oand o; we write
M O oa: 2
Values for O and o are given in Table 1 for severaltemperatures.
The results show that o is virtuallyindependent of temperature.
(The estimates of Oshow no systematic variation. This is probably
dueto a high degree of sensitivity to noise in the datafor this
parameter.)
The slope o defined by Eq. (2), sheds some lighton the nature of
the relaxation process in theintermediate regime. If we assume that
therelaxation process in the intermediate regimeconsists of blocks
of spins flipping over asislands coalesce, then the slope o
provides somemeasure of the number of spins involved in theprocess,
and possibly gives some insight into thecollective excitation
driving the decay of the islandasymmetry.
At times greater than the relaxation time for athe system has
essentially lost its memory of theinitial state; nonetheless,
inspection of spin con-figurations reveals there is a substantial
timeperiod before the system reaches its equilibriumstate. Fig. 6
shows typical spin configurations froma single simulation run at
several hundredthousand (Monte Carlo) time steps. Inspection ofthe
configurations, such as those in Fig. 6, at latetimes in the
simulation shows that the smecticphase has formed locally; however
the regionscontaining these sections of smectic phase domainsdo not
have a common orientation.
It is useful to think of the regions of smecticphase as being
analogous to crystallites in apolycrystalline sample. Boundaries
between
ARTICLE IN PRESS
t = 300,000 mcs t = 600,000 mcs t = 900,000 mcs
Fig. 6. Snapshots of the spin configuration at t 300; 000; 600;
000 and 900; 000 mcs showing the boundaries separating the regions
ofhorizontal and vertical stripes, with J 8:9g; T 4:25g and L 256:
The spin configurations shown are constructed by tiling
fourlattices together to form a 512 512 lattice.
Table 1
Estimates of the parameters O and o in Eq. (2)
T=g O o
3.50 41.477.2 20.170.63.75 24.374.5 21.970.44.00 6.973.0
22.670.34.25 19.276.8 21.170.4
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 142920
-
crystallites are determined by the presence oftopological
defects. The extremely slow kineticsof these defects moving through
the system andbeing annihilated determines evolution towardsthe
equilibrium smectic crystalline phase.
The general characteristics of the descriptionabove are common
to all simulations performedwith TtTO: At higher temperatures,
T\TO; therelaxation process does not exhibit the memoryeffect that
characterizes the low temperature,TtTO; behaviour and the system
decays to thetetragonal fluid phase with both the magnetizationand
the island asymmetry decaying to zero atapproximately the same
rate.
In summary of this section, at temperaturesTtTO the relaxation
of this system has manyanalogies with the relaxation of a
crystalline solidinitially prepared in the wrong crystalline
phase,however at temperatures T\TO the behaviour ofthe system is
analogous to a fluid able to readilyreturn to its equilibrium
phase.
4. Phenomenological model
Given the complexity of the nucleation processthat govern the
dynamics of the spins at theearliest times and low temperatures
(ToTO it ishardly surprising that the average magnetizationcurves,
such as those shown in Fig. 3, cannot bedescribed in terms of a
simple exponential decay.This is in contrast to the assumptions
made inearlier studies [13,14].
To construct a phenomenological description ofthe time
dependence of the magnetization we notethat the initial decay of
the magnetization isgoverned by the nucleation and growth of
thespin down islands. If we assume that the islandgrowth is
proportional to the total perimeter of thespin reversed islands,
then the rate of change in themagnetization may be determined from
a straight-forward scaling argument as
dm
dtE
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffin1 m
p; 3
where n denotes the number of spin reversedislands. If we
further assume that in the nucleationphase the number of spin
reversed islands, n; is
given by
nE1 mF; 4
then the initial rate of change through themagnetization is
related to the magnetizationthrough the relation
dm
dtE r1 mY 5
with Y 1 F=2:Following the island nucleation the magnetiza-
tion then decays to about 1% of its initial valueafter which it
is proportional to the islandasymmetry. Neglecting the remnant
magnetizationfor the time being and simply assuming thatthe
magnetization decays exponentially to zero,requires that
limt-N
dm
dtE
m
t; 6
where t denotes the relaxation time associatedwith the decay of
the magnetization following thenucleation phase.
Eqs. (5) and (6) can be combined into a simpleequation that
interpolates the two regimes.
dm
dtE
m
t1 mY1 rt 1m: 7
To determine the parameters in the phenomen-ological model in
Eq. (7) we fit m and mdetermined from the Monte Carlo data to
theabove functional form by means of a regressionanalysis. The
resultant curves and the data for fourtemperatures are shown in
Fig. 7. Estimates of theparameters for a range of temperatures are
givenin Table 2.
With the parameters obtained from the regres-sion analysis of
the m vs. m data, it is possible tosolve the differential equation,
given in Eq. (7),numerically for a given initial value of m:
Theresultant function mt and its associated deriva-tive mt are
plotted with the corresponding MCdata in Figs. 8 and 9 for four
temperatures. Thesolution to the differential equation is
insensitiveto the initial value provided it is less than but
closeto 1.
Assuming that the relaxation time t and the fliprate r are
governed by an activation energy andhence vary exponentially with
the inverse
ARTICLE IN PRESS
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 1429 21
-
temperature as
tEt0expDt=T; 8
1
rE
1
r0expDr=T; 9
we construct an Arrenhuis plot for both the relaxationtime t and
the flip rate r; as shown in Fig. 10.
An examination of the data in Fig. 10 suggeststhat the
relaxation process exhibits crossover atTETO: Therefore we fit the
data above and belowthe transition temperature, TO; separately.
Thelines of best fit are shown in Fig. 10 and yield thefollowing
parameter estimates in the two tempera-ture ranges.
Dt 17:5g70:5g for ToTO29:5g70:2g for T > TO:
(10
ARTICLE IN PRESS
Table 2
Estimates of the phenomenological model parameters obtained
at various values of temperature
T=g Y F 2Y F
3.50 0.55770.016 0.32670.004 0.78870.0324.00 0.60570.009
0.28670.005 0.92470.0194.25 0.63370.008 0.27570.003 0.99170.0174.75
0.68270.012 0.26770.002 1.09770.0245.00 0.65970.014 0.24570.005
1.07370.028
-1.610-3
-1.4
-1.2
-1.0
-0.8
-0.6
-0.4
-0.2
0.0
dm/d
t
1.00.80.60.40.2m
T = 4.0g
-2.010-3
-1.5
-1.0
-0.5
0.0
dm/d
t
1.00.80.60.40.2m
T = 4.25g
-610-3
-5
-4
-3
-2
-1
0dm
/dt
1.00.80.60.40.20.0m
T=5.0g
-60010-6
-500
-400
-300
-200
-100
0
dm/d
t
1.00.80.60.40.20.0m
T=3.5g
Fig. 7. A plot of m vs. m for several temperatures.
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 142922
-
The corresponding values for t0; the inverse of theattempt
frequency, are
t0 3:6 mcs70:4 mcs for ToTO
0:46 mcs70:07 mcs for T > TO:
(11
Similarly the energies and inverse attempt fre-quencies
associated with the flip rate are foundto be
Dr 32:8g71:1g for ToTO14:0g72:4g for T > TO;
(12
1
r0
0:05 mcs70:01 mcs for ToTO1:8 mcs70:6 mcs for T > TO:
(13
We note that the difference in the estimatedactivation energies
cannot be ascribed to statisticaluncertainty in the data. A more
detailed w2
analysis comparing the above fitting procedurewith a simple
linear fitting function, indicates thatusing the separate fitting
functions provides abetter description of the data.
While the above arguments provide anadequate phenomenological
description of the
ARTICLE IN PRESS
1.0
0.8
0.6
0.4
0.2
0.0
m
1400120010008006004002000
t (mcs)
T = 4.25g
1.0
0.8
0.6
0.4
0.2
0.0
m
5004003002001000
t (mcs)
T = 5.0g
1.0
0.8
0.6
0.4
0.2
0.0
m
500040003000200010000
t (mcs)
T = 3.5g
1.0
0.8
0.6
0.4
0.2
0.0
m
2000150010005000
t (mcs)
T = 4.0g
Fig. 8. A plot of m as a function of time for several
temperatures. The solid line is that obtained by numerically
solving Eq. (7) using
parameters obtained from a regression analysis.
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 1429 23
-
magnetization, a similar analysis of the islandasymmetry is more
difficult. In part this is becausethe fluctuations in Monte Carlo
data for the islandasymmetry is greater proportionally than
thecorresponding fluctuations in the magnetization.Our studies also
show that the slow decay of theisland asymmetry at long times
cannot be de-scribed in terms of a simple exponential. Despitethis
we can nevertheless construct a simple modelfor the asymmetry based
on some reasonableassumptions about the relaxation process.
Asdiscussed earlier we can see from the data thatthe island
asymmetry initially rises rapidly asislands of opposite spin
orientation nucleate. As
the islands continue to grow the nucleation processeffectively
stops (or, at least, is negligable) and theisland asymmetry begins
to decrease as the islandsbegin to connect together and the
backgrounddisconnects into multiple islands. Quantitativelywe can
describe this by
at it Z t0
gt t0at0 dt0; 14
where it describes the rapid nucleation of theislands to some
saturation value, while the secondterm describes the decay of the
island asymmetrydue to the islands coalescing and is
characterizedby a relaxation function gt: This separation into
ARTICLE IN PRESS
-1.610-3
-1.4
-1.2
-1.0
-0.8
-0.6
-0.4
-0.2
0.0
dm/d
t
25002000150010005000
t (mcs)
T = 4.0g
-610-3
-5
-4
-3
-2
-1
0dm
/dt
6004002000 t (mcs)
T = 5.0g
-2.010-3
-1.5
-1.0
-0.5
0.0
dm/d
t
12008004000 t (mcs)
T = 4.25g
-600x10-6
-500
-400
-300
-200
-100
0
dm/d
t
500040003000200010000 t (mcs)
T = 3.5g
Fig. 9. A plot of m as a function of time for several
temperatures. The solid line shows the numerical solution of Eq.
(7).
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 142924
-
two distinct processes is only useful if the timescales
describing the nucleation and decay arequite distinct.
At low temperatures (TtTO) we can assumethat this separation of
the time scales does occur asthe island nucleation appears to occur
on the sametime scale as the decay of the magnetization. Ourearlier
discussion implies that the two processes,the island nucleation and
decay of the magnetiza-tion are strongly correlated at early times;
indeedEq. (4) implies that at early times itE1 mtF:Plotting a as a
function of m for severaltemperatures (Fig. 11) we see that at
early timesmt1 the Monte Carlo data does suggest a powerlaw
relationship. We therefore fit the data shownin Fig. 11 using
it a1 mtF1 bmt; 15
where the 1 bm term allows for the fact that aappears to
saturate and even decrease somewhatbefore the magnetization has
decayed to zero. Thisreflects the continued island growth even
thoughthe nucleation process has ceased. The best fit linesobtained
by regression for several values oftemperature, T=g; are shown in
Fig. 11 togetherwith the corresponding Monte Carlo data. Thevalues
for F obtained from the fit are given inTable 2, for several
temperatures, together withthe quantity 2Y F: We note that, except
at the
lowest temperature considered, the calculatedvalue of 2Y F is
close to unity, giving goodagreement with the value predicted by
the simplescaling theory that underlies the current analysis.
To determine the time dependence of at fromthe nucleation
function requires the explicit formfor the relaxation function gt
t0 defined byEq. (14). Thus far we have been unable to obtain
asuitable function for gt t0 that describes thedecay of a over the
temperature range of interest.With the absence of suitable form for
gt t0we therefore make the simplest possible choiceapproximating it
by a constant.
gtE1
ta: 16
While more complicated functional forms ofgt t0 yield a better
fit to the data, they addlittle to the analysis and do not reveal
any usefulsystematic behaviour with respect to the tempera-ture.
The above form for the relaxation functionalso has the advantage
that it yields an exactsolution for at in terms of the nucleation
functionit
at it 1
ta
Z t0
exp t t0
ta
it0 dt0: 17
Using the value of Y in Table 2, and selecting asuitable value
for ta and substituting the expres-sion for it into Eq. (17) we can
determine theremaining coefficients a and b by linear
regression.The results from the model are shown in Fig. 12together
with the data.
The slow decay of the residual magnetization inthe intermediate
regime together with its lineardependence on a suggests a simple
generalizationof the phenomenological equation for m by theaddition
of a term linear in a: The full system ofequations for the
phenomenological model is then
dm
dt
m
t1 mY1 rt 1m ca a0; 18
at it 1
ta
Z t0
exp t t0
ta
it0 dt0; 19
it a1 mtF1 bmt; 20
where the coefficient c o=L2t is determined byEq. (2). The
results for the full set of equations
ARTICLE IN PRESS
6
5
4
3
0.280.260.240.220.200.180.160.140.12
log log r
1/T
log
Fig. 10. Arrenhuis plot for the relaxation time t and flip rate
r:
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 1429 25
-
together with the corresponding data are shown inFig. 13 for T
4:25: (For the purposes ofillustration we use o 20; L 256 and t
216for calculating the fitting line).
While alpha decays to zero, or at least fluctuatesabout zero in
our finite size system, at the longesttimes simulated in this
study, Fig. 6 clearly showsthat the system has still not reached
its equili-brium. Instead we see regions, which we refer to
ascrystallites, in which stripes with a well definedorientation
have formed, but these regions do notshare a common orientation.
Thus while a 0 is anecessary condition for equilibrium it is not
a
sufficient one. It would be interesting to extend thepresent
phenomenological analysis into the crys-tallite regime by coupling
the orientational orderparameter Ohv [8] into the analysis however,
at thepresent time, it would be impractical given the timescales
involved.
5. Experimental and technological implications
The results of the simulations reported heredemonstrate that the
relaxation process above TOis qualitatively different from that
below TO: In
ARTICLE IN PRESS
35
30
25
20
15
10
5
0
1.00.80.60.40.20.0
m
T =3.5g
200
150
100
50
0
1.00.80.60.40.20.0
m
T =5.0g
70
60
50
40
30
20
10
0
1.00.80.60.40.2
m
T =4.0g
80
60
40
20
0
1.00.80.60.40.2
m
T =4.25g
Fig. 11. A plot of a vs. m for several temperatures.
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 142926
-
addition to the difference in the activation energyfor
relaxation of the magnetization noted pre-viously, the relaxation
process at low temperature(ToTO) exhibits three distinct regimes:
the decayof the magnetization, the decay of the islandasymmetry,
with the associated loss of memory ofthe initial state, and finally
the growth of orienta-tional order with the gradual crystallization
of theequilibrium striped phase from the quenched state.By contrast
the relaxation process at high tem-peratures (T > TO) can be
described in terms of asingle regime in which the magnetization and
theisland asymmetry rapidly decay to zero, beyond
which the system appears to be in the equilibriumtetragonal
phase.
As noted earlier, direct imaging of the domainshas recently been
used to observe the smectic totetragonal transition [9]. However,
indirect probesof the magnetic properties which would also allowthe
smectic and tetragonal phases to be distin-guished will greatly
facilitate the characterizationof ultra thin magnetic films and our
present studiesindicate that dynamical phenomena such asrelaxation
are candidates for such indirect probes.Early experiments used
direct domain imaging tofollow spin reversal and nucleation
following the
ARTICLE IN PRESS
35
30
25
20
15
10
5
0
6010350403020100 t (mcs)
T = 3.5g
200
150
100
50
0
2000150010005000
t (mcs)
T =5.0g
80
60
40
20
0
500040003000200010000 t (mcs)
T = 4.25g
70
60
50
40
30
20
10
0
20x103151050
t (mcs)
T = 4.0g
Fig. 12. A plot of a as a function of time t for several
temperatures. The solid line is the function determined from
thephenomenological model described in the text.
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 1429 27
-
reversal of an external magnetic field [16].Although somewhat
different to the process ofrelaxation to a zero external field
state simulated inour study the results are qualitatively
consistent.In particular the system does not return to itsvirgin
domain phase (in this case a singledomain the size of the sample
imaged by Faradayrotation) but forms a more complex domainpattern
which locks in if the external field isswitched off. The
experiments performed by
Berger and Hopster [5] deal with relaxation froma saturated
state to a zero external field state in atemperature range where
(we believe) the equili-brium state would be multi-domain. The
variationof the magnetization with time observed by Bergerand
Hopster is not a simple exponential curve butrather appears to have
horizontal regions whichlast for macroscopic periods of time. This
may bean indication that different regions of the sampletake
different lengths of time to relax and is
ARTICLE IN PRESS
80
60
40
20
0
1.00.80.60.40.2 m
T = 4.25g
1.0
0.8
0.6
0.4
0.2
m
2000150010005000
t (mcs)
T = 4.25g
Fig. 13. m as a function of t and a as a function of m using the
full phenomenological model of Eq. (20).
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 142928
-
consistent with the idea that several differentregimes of
relaxation behaviour occur. Howevera comparison of the simulations
reported herewith the work of Berger and Hopster is compli-cated by
the fact that the system they studied alsoexhibits a reorientation
transition. In order tosimulate this it would be necessary to
consider aHeisenberg model. A very interesting, thoughlargely
qualitative, simulation study of this kindhas recently been
reported by Iglesias and Labarta[17]. A quantitative analysis of
such a study todetermine to what extent a phenomenologicalmodel of
the type used here can be applied wouldaid in making comparisons
with experimentalsystems many of which exhibit a
reorientationtransition.
Technology based on either the magneticproperties alone or in
combination with electronicproperties offer the potential for
devices withhigher speed and capacity, and lower powerconsumption
than conventional microelectronicdevices [1]. The development of
such devices hasbeen greatly advanced by recent innovations
inmaterials engineering including using compositesof layers with
different magnetic properties. Thepresent work indicates that some
care must beapplied in interpreting experimental results and
intechnological uses of these materials as (at lowtemperatures) the
abrupt drop of the magnetiza-tion following the switching on and
subsequentswitching off of an external magnetic field does notof
itself indicate a return to equilibrium but ratherthe system may
retain a memory of the initial statefor a time considerably longer
than the initialdecay of the magnetization would indicate.
Whilethis memory effect may be a difficulty in someapplications the
possibility that there may beapplications of the effect in its own
right (perhapsthe ability to recover data even after very
longtimes) is an intriguing one.
Acknowledgements
This work is supported in part by the NaturalSciences and
Engineering Research Council ofCanada. One of us (KDB) thanks the
Centre forInterdisplinary Studies in Chemical Physics, Uni-versity
of Western Ontario, for a Senior VisitingFellowship during the
early stages of this work.
References
[1] S.A. Wolf, D.D. Awshalom, R.A. Buhrman, J.M. Daugh-
ton, S. von Moln!ar, M.L. Roukes, A.Y. Chtchelkanova,
D.M. Treger, Science 294 (2001) 1488.
[2] G.A. Prinz, Science 282 (1998) 1660.
[3] B. Heinrich, J.F. Cochran, Adv. Phys. 42 (1994) 523.
[4] K. DeBell, A.B. MacIsaac, J.P. Whitehead, Rev. Mod.
Phys. 72 (2000) 225.
[5] A. Berger, H. Hopster, Phys. Rev. Lett. 76 (1996) 519.
[6] R.P. Cowburn, J. Ferr!e, J.-P. Jamet, S.J. Gray, J.A.C.
Bland, Phys. Rev. B 55 (1997) 11593.
[7] D. Venus, C.S. Arnold, M. Dunlavy, Phys. Rev. B 60
(1999) 9607.
[8] I.N. Booth, A.B. MacIsaac, K. DeBell, J.P. Whitehead,
Phys. Rev. Lett 75 (1995) 950.
[9] A. Vaterlaus, C. Stamm, U. Mair, M.G. Pini, P. Politi,
D. Pescia, Phys. Rev. Lett. 84 (2000) 2247.
[10] M.M. Hurley, S.J. Singer, Phys. Rev. B 46 (1992) 5783.
[11] A.D. Stoycheva, S.J. Singer, Phys. Rev. Lett. 84 (1999)
4657.
[12] S. Bromley, J.P. Whitehead, High Performance Comput-
ing Systems and Applications, Kluwer, Boston, 2000,
p. 587.
[13] L.C. Sampaio, M.P. de Alburquerque, F.S. de Menezes,
Phys. Rev. B 54 (1996) 6465.
[14] T.G. Rappoport, F.S. de Menezes, L.C. Sampaio, M.P.
Albuquerque, F. Mello, Int. J. Modern Phys. C 9 (1998) 1.
[15] I.N. Booth, Honours Thesis, Memorial University of
Newfoundland, 1995.
[16] J. Pommier, P. Meyer, G. P!enissard, J. Ferr!e, P.
Bruno,
D. Renard, Phys. Rev. Lett. 65 (1990) 2054.
[17] O. Iglesias, A. Labarta, J. Magn. Magn. Mater. 221
(2000)
149.
ARTICLE IN PRESS
S.P. Bromley et al. / Journal of Magnetism and Magnetic
Materials 264 (2003) 1429 29
Memory effects and slow dynamics in ultra thin magnetic
filmsIntroductionThe simulation modelRelaxation
kineticsPhenomenological modelExperimental and technological
implicationsAcknowledgementsReferences