HAL Id: jpa-00245600
https://hal.archives-ouvertes.fr/jpa-00245600
Submitted on 1 Jan 1987
HAL
is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire
HAL, estdestinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Influence of various hydrodynamic regimes in a melt on a solidification interface
J.J. Favier, A. Rouzaud, J. Coméra
To cite this version:
J.J. Favier, A. Rouzaud, J. Coméra. Influence of various hydrodynamic regimes in a melt on a
solidification interface. Revue de Physique Appliquée, Société française de physique / EDP, 1987, 22
(8), pp.713-718. �10.1051/rphysap:01987002208071300�. �jpa-00245600�
713
Influence of various hydrodynamic regimes in
amelt
on asolidification interface
J. J.
Favier,
A. Rouzaud and J. ComéraCEA-IRDI-DMECN-DMG-SEM, Laboratoire d’Etude de la Solidification,
Centre d’Etudes Nucléaires de Grenoble, 85X, 38041 Grenoble Cedex, France
(Reçu
le 10 octobre 1986, révisé le 24février
1987, accepté le 28 avril1987)
Résumé. 2014 L’influence de divers
régimes hydrodynamiques
dans un bainliquide
sur une interface L/S estanalysée expérimentalement
et de façon continuedepuis
unsimple
écoulement laminairejusqu’à
la turbulencedéveloppée.
Lareprésentation
des modes convectifs dansl’espace
de Fourier met en évidence l’alternance dezones
chaotiques
et de zones oscillanteslorsque
lechamp thermique
estprogressivement
élevé. Deplus
unmécanisme bien connu
d’apparition
de la turbulence(mécanisme
deFeigenbaum)
semble retrouvéexpérimentalement.
L’examen de l’incidence de telsrégimes
sur la solidification termine cette étude.Abstract. - The influence of various convective levels in a melt on a S/L interface is
experimentally investigated
in a continuous way from pure laminar flows to turbulent ones.Simple description
in the Fourier space makes apparent a chaotic andoscillating
zones alternance when thermal field isprogressively
increased.A classical theoretical transition towards the chaos
(Feigenbaum mechanism)
seems to be identified. Influenceon
crystal growth
isfinally
examined.Revue Phys.
Appl.
22(1987)
713-718 AOÛT 1987,Classification
Physics Abstracts
47.25Q - 68.45 - 81.10F - 81.30F
1. Introduction.
Growth from melt is until
today
the mainpractical
art to obtain suitable
single crystals.
Thehydrodyn-
amic state of the
liquid phase during
thegrowth
is aprimordial
parameter withregard
to the electronicproperties
of so grownmaterials,
because of itshigh
influence on the
dopant
distribution[1].
Since someten years an
important
effort has been undertaken in order toquantify
such an effect[2, 3].
Basic ideasare now well established but
unfortunately
the non-linear conservation
equations, supplied
with appro-priate complex boundary
conditions on the S/Linterface,
does not allow an accurate numerical simulation in thegeneral
case ofhigh
convectiveregimes.
The lack ofadequate
mathematical tool toanalyse
all of thepossible experimental configura-
tions from the pure diffusive flow to the turbulent one, gave rise to two different theoretical ap-
proaches :
- either turbulent flows
modeling by
classicalfluctuations methods
[4, 5],
- or
performing weakly
non-linearexpansions starting
from known linear solutions of low convec-tive systems
[6].
Concurrently,
availableexperimental
works oncrystal growth
areclassified,
oftenarbitrarily,
ac-cording
to these twolimiting
cases, into « low » or«
high »
convectiveexperiments.
Our
experimental
aim isquite
different since wepreferred
toqualitatively analyse
the influence of convection on a S/L interface in a metallic orsemiconductor
sample,
over a wide continuousrange of convective states in the
liquid phase.
Today,
transition towards chaotic state are notreally
understood and three mechanisms at least are prop- ounded
by
Ruelle-Takens[7],
Pomeau-Man-neville-[8]
andFeigenbaum [9]. So,
thesingle
ambi-tion of the
present
paper consists ondescribing
thestrong
influence of chaoticliquid
motions on a S/Linterface. ’
2.
Experimental
set-up.To correlate
hydrodynamics signals and S/L interface
response requires
some restrictive experimental
con-
ditions. Indeed a non
disturbing
method must beused to measure the evolution of a
given
interfacial parameter. Thesimplest
way toperform
such anexperiment
consists onsimultaneously measuring
Article published online by EDP Sciences and available at http://dx.doi.org/10.1051/rphysap:01987002208071300
714
Fig.
1. - Artistic view of Mephisto.the thermal field at a fixed
place
in theliquid
close tothe S/L
front,
and the real interfacetemperature.
The device
[10, 12],
shown infigure 1,
allows such measurements. Thisapparatus
is based on Seebeck effect[10, 11, 13],
a cross effect which links thermal and electrical forces in aconducting
medium. Con-sidering
the thermoelectricloop
made upby
theSolid/Liquid/Solid metallic alloy
described infigure 2,
Seebeckvoltage
is definedby :
Assuming
that 17s, 17L, the solid andliquid
absoluteSeebeck coefficient remain constant over
one
obtains, following
the formalism defined infigure
2 :where 0394T means the
temperature
difference between the twoLiquid/Solid junctions.
At rest, when nopulling
isperformed, temperature
of both interfacesis
obviously
the same unless localunsteady
convec-Fig. 2. - Thermoelectric loop used in Mephisto.
715 tive motions induce a L/S front
displacement.
In thislast case, or when a
pulling
rate isextemally imposed by moving
the movablefurnace,
kineticand/or solutal effects may introduce an
undercooling
à
linearly
linked to the Seebeckvoltage provided
that the Seebeck coefficient
(ns - ’TI L)
can be re-garded
as a constant.Though
theexperiment princi- ple
iselementary,
itspractical carrying
outrequires
to take many
precautions.
Indeed metalsbeing
poorthermocouples,
thevoltages
to be measured are very low.Moreover,
to assume Seebeckvoltage only depending
on S/L metaljunctions requires
to avoidthermoelectric effects at Cu wire/solid
sample junc- tions,
which supposes tokeep
them at a samearbitrary temperature To (see Fig. 2). Today
adifferential
temperature regulation
at thesample
ends
ranging
about 5 x10- 3
K isobtained, inducing
a maximum electric noise about 30 x
10- 9
V.By using
an accuratenanovoltmeter,
anundercooling
higher than
2 x10- 2
K is thus tracked down in the metallic or semiconductoralloy. Finally,
it must bepointed
out that this method is notperturbative
since no current flows
along
thesample.
In order to
locally
measure the thermalfield,
which can be varied with thehelp
of thecorrespond- ing furnace,
two thinthermocouples
of 0.25 mm indiameter are inserted in the
liquid phase,
near thetwo interfaces. A
typical experimental temperature profile
which can be achieved in MEPHISTO is shown on thetop
offigure
1. Thegood isothermicity
is ensured
by
sodium heatpipes. Experiments
areperformed
in an horizontalconfiguration.
The deviceshown in
figure
2 allows to achieve solidification andmelting cycles
over a 150 mmlength.
However inorder to
separate
thermal effects due to convective motions from the onesimposed by solidification, experiments
have beenperformed
withoutpulling.
Therefore no Seebeck
voltage
isexpected
at theequilibrium.
The metallicalloy presently
used was aSn 1 at. % Bi
sample,
80 cmlong,
whose ther-mophysical properties
and absolute Seebeck coeffi- cients areperfectly
known.Special
calibrations of the thermoelectric power have been carried outseparately [10].
In our
experimental hydrodynamic configuration,
the
longitudinal
thermalgradient,
normal to the S/Linterface,
isobviously
the maindriving
force forconvection since it is
orthogonal
to thegravity
vector.
So, increasing
the furnaceregulation
tem-perature increases the thermal
gradient
in theliquid,
therefore raises convective motions in the
liquid alloy
from a laminar state up to a chaotic one. In this way,following
the classicalterminology
used in the turbulenceanalysis (see
forexample [9]
or[17]) imposed temperature regulation
will sometimes be called below stochasticparameter.
At the
beginning
of theexperimental
program, both furnaces areregulated
at 600°C,
which is thelowest
working temperature
for sodium heatpipes.
Once the
equilibrium
isreached,
a low thermalperturbation
isimposed
at one of the L/Sinterfaces, by
a smallchange
in theregulation temperature
of thefacing furnace,
up to the onset of anoscillatory regime
in theliquid
bulk. This is easy to obtain for low Prandtl numberfluids,
asexplained by
Azouni
[14].
Once such aregime
isobserved,
theconcerned furnace temperature is stabilized and thermal fluctuations in the
liquid
and at the interface(Seebeck signal)
aredigitized
and recorded. Afterthat, temperature
of the same furnace isprogressive- ly
increased in somestages
up to the maximum available temperature of 960 °C. An automatic dataacquisition
isperformed
at each stage.3.
Expérimental
results.The main data channels are scanned every 0.3 s.
This
scanning
rate isperfectly compatible, according
to Shannon
theorem [15],
with theanalysed
pre- filteredsignals
bandwith(0-1 Hz).
Recordedsignals
are next
processed by
adesktop
computer which calculates their Discrete Fourier Transformby using
a classical Fast Fourier Transform
algorithm [16].
Typical digitized
data and theircorresponding
DFT-FFT are
presented
infigure
3 where anexciting oscillating hydrodynamic regime,
as measuredby
the thermal
probe (channel2),
is observed for agiven temperature
of 800 °C. The related output Seebecksignal (received
onchannel 1)
exhibits asimilar behaviour.
Mathematically
this suggests aquite good
correlation ofsignals
between theinput
and the
output
of thequasi
linear system made upby
the S/L
junction
and theneighbouring liquid.
Such acorrelation is also verified for other temperatures
ranging
between 600 °C and 900 °C. In order todirectly
visualize all of theexperiments,
anotherdescription
has been chosenby plotting
a top view of both Fourier spectra for the variousexperiment working
temperatures. The relativeheight
of thevarious
peaks
is achievedby drawing
spectra level lines in the(frequency-temperature) plane.
Due totheir discrete nature
imposed by
a finite measure-ment
duration,
near about 200 s, suchplotted
mapsare difficult to
quantitatively analyse
but present theprime advantage
togive
acomprehensive
overallpicture.
Three
experiments
series have beenperformed
under the same
experimental
process.They
arepresented
infigures 4,
5 and 6.In
figure 4,
the thermal oscillationsspectrum
shows at lowtemperatures
asimple
initialoscillating regime
made up of a basic mode and its first harmonic.Increasing
thermalfield,
itrapidly
van-ishes and a first turbulent
regime
appears withlarger bandwith,
between 680 °C and 780 °C.Surprisingly,
a new
quasi
sinusoidalregime
next reappears(basic
716
Fig. 3. - Typical
experimental results and theircorresponding
Fourier transform. Channel 1 : Seebecksignal [V].
Gain : 1 V H 1.76 °C. Channel 2 :
thermal" signal [V].
Gain :’1 V H 83.4°C.mode + first
harmonic)
whichdefinitively
ends at860
°C,
where acomplete
chaotic state is reached.This last one is characterized
by
aramped
noisespectrum, as observed
by
Hurle[17].
The
corresponding
Seebeck map is morecomplex,
in
spite
of astriking
likeness with theprevious
one.The main
frequencies
of theexciting input signal
arefound in the
output
one, but small local solidificationphenomena
also induce their own natural fre-quencies
with their own harmonics. Indeed thermalfluctuations induce fluctuations of the interface
position ; thus, phenomena
as soluteboundary layer
evolution and interfacial kinetics introduce their
own
dynamics.
All thosepeaks
areconverging
towards the two chaotics domains. The
only
one,close to 1
Hz,
which is unaffectedby
the chaos seemsto result in an external cause : noise from electronic
measuring
circuits orparasite
effects at the ends of thesample.
This is alsosuggested by
its constantpresence
during
all of theexperiments.
Fig. 4. - Left : thermal oscillations map of the first
experimental
series. Right : Seebeck oscillations map of the firstexperimental
series.717 The main four domains observed in
figure
4 arefound
again
infigure
5 whichcorresponds
to thesecond
experimental
series. However in this last caseoscillating regimes
arepartially damped
down andextremely
lowfrequencies prevail
in suchregions ( 0.1 Hz).
On the other hand theoutstanding
existence of a first chaotic zone in the same
tempera-
ture range as before is to be
emphasized.
The third series visualized in
figure
6 alsogives interesting
informations. Theonly
difference with the twoprevious
ones lies in the fact that nooscillating regime
could appear in the 600 C exper-iment, possibly
due to an infinitesimal difference in theapparently
identicalexperimental
conditions.Thereby
thesimplest
monochromaticregime
isreached
only
for 700°C,
andpersists
up to 800 °C.At this last temperature a first
period doubling (fui - f 1/2 )
isclearly shown,
and a second one( fi/2 - f 1/4 )
seems to be present in the thermal oscillations map but cannot be tracked down on thecorresponding
Seebeck oscillations map where a lowfrequency
noise conceals allpossible
information.These effects are also shown on the local representa- tion
(Fig. 3).
Such a transitionsuggests
aperiod
doubling
mechanism identifiedby Feigenbaum [9]
seems to be present. This author showed the
logistic equation :
describes the
following
process :when the stochastic
parameter s
is increased from si = 3/4(first transition)
upto sc
= 0.86237(pure
chaotic
state).
Feigenbaum theory
morepredicts
arapid
increasesin
period doublings
for a small increase in the stochasticparameter s
once the first critical value’ si
has been reached.Equivalent phenomenon
is alsoobserved here since
complete
chaotic stateprevails
at the
following
stage of 820 °C.4. Conclusions.
Some
experimental
resultsconcerning
the influence of convection on a S/Linterface,
for a wide range of convectivelevels,
have beenpresented
here.Fully
aware that transition towards the chaos still remains
Fig. 5. - Top : thermal oscillations map of the second Fig. 6. - Top : thermal oscillations map of the third
experimental series. Bottom : Seebeck oscillations map of experimental series. Bottom : Seebeck oscillations map of the second
experimental
series. the thirdexperimental
series.718
an intact
problem,
as well on a theoreticalpoint
ofview as on an
experimental
one, wedeliberately
limited our
analysis,
in a firststep,
toqualitative
considerations. No extensive
signal processing
methods have thus been undertaken to characterize the
Input/Output signals
mathematicalproperties,
ascoherence and transfer functions. However
simple
data
handling
andprocessing emphasize
someimportant practical
results :- The first one is a new confirmation of the strong
dependence
of fluid motions selection on thesystem
externalboundary
conditions.Indeed,
evenfor
apparently
identicalexperimental conditions,
inspite
of manyprecautions,
noperfectly
similarhydrodynamic
states are observed for each exper- imental series.- More
important
because itsrepercussion
oncrystal growth
is thesurprising
existence of somechaotic zones
separated by quasi
monochromaticones. This result has a
practical importance
since itshows
decreasing
thedestabilizing driving
force(i.e.
Rayleigh
number which is similar to stochastic - parameters)
is not a sufficient way toundoubtedly
obtain
steady growth
conditions. On the contrary aregular time-dependent regime
may thus occur anddramatically damage
electronicproperties
ofgrowing crystals.
Indeed the interface natural response shows itself in striations. A direct relation between measured fluctuations in theliquid
near the interface andcrystal
defects isalways
observed. Such induced striationsphenomenon
is illustrated infigure 7 which’"exhibits
themeiallographic analysis
of a Sn-Bisample
grown at lowvelocity
from a melt whereprevailed
sinusoidal fluctuations. It also must be noticed thatequivalent separated
chaoticregions
, have
recently
been observedby
Muller[18]
in cen-trifugation experiments. Finally,
theonly
way toFig.
7. -Metallographic analysis
of an Sn-Bi samplegrown from a melt where
prevailed
a sinusoidal fluctu- ation.avoid such troubles seems to consist on
using
newgeneration crystal growth
devices which allow to realtime
supervise
the main parameters(liquid
andinterface
temperature)
measuredby appropriate
nondisturbing
methods.Acknowledgments.
This work is part of the studies
performed
in theframe of the MEPHISTO
Project (collaboration
between CEA-CNES and
NASA).
The authors are indebted to P.
Contamin,
R. Ginet and G.
Marquet
for theirprecious daily
assistance in the
project.
This work has been done in the frame
J of
theagreement
GRAMME between CNES and CEA- IRF., References
[1]
FLEMINGS, M. C.,Solidification
processing(Mac
Graw
Hill)
1974.[2]
HURLE, D. T. J., JAKEMAN, E., Phys. Chem.Hyd- rodyn.
2(1981)
237.[3]
OSTRACH, S., J. Fluids Eng. 105(1976)
5.[4]
REYNOLDS, W. C., Annu. Rev. Fluid Mech. 8(1976)
5.
[5]
BRADSHAW, P., Engineering Calculation methodsfor
turbulent
flow (Academic Press)
1981.[6]
NORMAND, C., POMEAU, Y., VELARDE, M. G., Rev.Mod. Phys. 49
(1977)
581.[7] RUELLE,
D., TAKENS, F., Commun. Math. Phys. 20(1971)
167.[8]
POMEAU, Y., MANNEVILLE, P.,Commun.
Math.Phys. 74
(1980)
189.[9] FEIGENBAUM,
M., Phys. Lett. A 74(1979)
375.[10]
FAVIER, J. J., Thesis Grenoble(1977).
[11]
THOMSON, W., Proc. Roy. Soc.Edinburg,
Trans. 21, Part I(1857)
123.[12]
COMERA, J., CONTAMIN, P., FAVIER, J. J., MAR- QUET, G., Patent N° 86 06107(28/04/1986).
[13]
PRIGOGINE, J., Thermodynamicsof
irreversible pro-cesses
(J. Wiley
andSon)
1985.[14]
AZOUNI, M. A., Phys. Chem.Hydrodyn.
2(1981)
295.
[15]
MAX, J., Méthodes et techniques de traitement dusignal
etapplication
aux mesuresphysiques,
Tome 1, 4e Edit.
(Masson)
1985.[16]
BRIGHAM, E. O., The Fast FourierTransform (Pren- tice-Hall)
1974.[17]
HURLE, D. T. J.,Workshop Aix-la-Chapelle (1984)
55.