HAL Id: jpa-00246582
https://hal.archives-ouvertes.fr/jpa-00246582
Submitted on 1 Jan 1992
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, est destinée 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.
Hole generation of prepeaks in diffraction patterns of glasses
Jean Dixmier
To cite this version:
Jean Dixmier. Hole generation of prepeaks in diffraction patterns of glasses. Journal de Physique I,
EDP Sciences, 1992, 2 (6), pp.1011-1027. �10.1051/jp1:1992188�. �jpa-00246582�
Classification
physics
Abstracts61.40 61.40D
Hole generation of prepeaks in diffraction patterns of glasses
Jean Dixmier
Laboratohe de
Physique
des Solides de Bellevue, C-N-R-S-,place
Aristide Briand, 92195 Meudon Cedex, France(Received 6 January 1992,
accepted
infinal form
10 March 1992)Abstract. It is shown that most of the covalent
glass
atomic structures can be derived from the Dense RandomPacking
of HardSpheres ID-R-P-H-S-
modelby introducing successively
a series of inflated subnetworksconsisting
of holes. From thispoint
of view the so called FirstSharp
Diffraction Peak IF-S-D-P-) of their diffraction pattem appears as a hole
generated prepeak
of the lstneighbor
component of the D-R-P-H-S- Structure Factor. The various MediumRange
Order (M.R.O.) features which can modulate the
prepeak
characteristics are reviewed localtopology,
network relaxation, chemicalordering, homogeneity.
A tentativerelationship
is made betweenprepeak
behavior andphysical
property variations with aspecial emphasis
on theoptical absorption
spectra of covalentglasses.
Introduction.
Not
surprisingly,
the firstambiguous
weakpeak
which appearsoccasionally
on the low q range of variousMetglass (M.G.)
structure factorsS(q) (q
=
4 w sin b/A
),
as well as thestrong composition dependent
« FirstSharp
Diffraction Peak »(F.S.D.P.)
of some covalentamorphous materials,
have attracted much attention since thebeginning
ofglass
studies.Indeed,
all disordered materials areconcemed,
from metallic to covalentglasses,
fromsingle
element materials tomulti-component amorphous alloys.
MediumRange
Atomic Order and Chemical Order are the usualkey
words which are associated with the F-S-D-P-designation.
The fact that thestrength
of one, andonly
one diffraction halomight
varyconsiderably
or evendisappear depending
on thesample composition, preparation
conditionsor more
astonishingly,
the type of radiation chosen for agiven
diffractionexperiment ii-e-
X- ray or neutronsource),
has oftengiven
amysterious
character to its hiddenphysical origin (See Fig, I).
Besides a renewed interest in the
understanding
ofprepeak
structuralorigins,
a few studies have also been devotedrecently
to therelationship
betweenphysical property
variations and their influence on theprepeak profiles.
At thepresent
time newkey
words should be added to theprevious
ones such as minimumminimorum,
for theconfiguration
energy, relaxationphenomena,
texture and moregenerally speaking
defects inamorphous
materials.In this paper we would like to
present
an overview of thesubject
withoutattempting
at an exhaustivebibliography.
We would like to demonstrate how ageneralized prepeak concept
4
Qi
3 »~
~~2
~
~*
i
* +
O 2 4 6 8 12
o i~~
Fig. 1. Revealing a
prepeak by
aweighting
contrast between different atomicscattering lengths
: X-ray and neutron diffraction structure factors of Co80P20. In the neutron
experiment
the P-P paircontribution to the total
S(q)
is enhanced from I.4IX-Ray)
to 10.2 9b. Hence thebump
at q = 2h~
' isgenerated by
an average P-P distance of 3.5h
due to chemicalordering.
can
help
to model the structure ofglasses
and toclassify
the differentpossible physical origins
of its
profile
variation.Finally,
the connection betweenprepeaks
andphysical properties
is undertaken withspecial emphasis
on theoptical
characteristics ofamorphous
semiconductors.1.
Prepeak
andpacking
fraction. Theorigins
of the hole model idea.To our
knowledge
the first paper that suggests that the F.S.D.P.might
arise from the appearance of a « zone of low atom occupancy » isVeprek
et al. 1981[1]. Later,
in1985,
anoverview of
prepeak problems
ispresented by
Moss and Price[2].
All materialsexhibiting prepeaks
arerewiewed,
from covalent to metallicglasses (M.G.).
In theiropinion
the clue to the appearance of the F-S-D-P- would arise from thepacking
of basic structural or molecularunits associated with Medium
Range
Atomic Order(M.R.O.).
Before
discussing
thephysical origin
ofprepeaks
let us first establish a definition of what shall be considered aprepeak
or a F-S-D-P-In the first studies
conceming glasses, liquid
andamorphous materials,
before thecalculation of Radial Distribution Functions
(R.D.F.),
the formulagenerally applied
to the first diffraction haloposition
was :X~
= 1.23 x A/2 x sin
b~
where is theX-ray wavelength.
This
procedure
wassupposed
toyield
the firstneighbor pair
distance between close atomsX~. However,
as Guinier reports(3),
this formula must be used with caution.Indeed,
while the firstpeak
of metallicglass
pattemsyields
the true firstneighbor
of a closepacked network,
the sameprocedure applied
to theamorphous
Si pattem leads to apair
distance 1.65 timeslarger
thanexpected. Applying
the same formula to the secondpeak position
leads to a value closer to the atomic diameter but there is still adiscrepancy
of about 20 fb. When various structure factors of metallicglasses
became available itappeared
thatthey
allexhibited the same
profile
with the well-known «shoulder» on the second halo. Thecorresponding
Radial Distribution Function(R.D.F.)
indicated that thistypical
feature arose from the hidden 5-fold symmetry of what should be described as the Dense RandomPacking
of Hard
Spheres.
At this
point,
it is worthwhile to compare(Fig. 2A)
the M-G-S(Q)
with those ofamorphous
tetracoordinated elements and ofbinary amorphous glasses mixing
2 elements with different valencies. In the latter case, it can be a mixture of elements of the fourthcolumn of the
periodic
table with either elements of the fifth column(coordination
number3)
or of the sixth column
(coordination
number2), obeying
the 8-N rule. We have chosenGeSe2
since itsX-ray
diffraction pattem is almostequivalent
to asingle
elementpattem,
due to the almostequal
size and diffusion factor values of Ge and Se atoms. The abscissae have beennormalized in order to eliminate the atomic size differences.
The
important
feature to note here is that thefight
side of all three curves,beyond
thepeaks
markedQi,
have almost the sameprofile, particularly
thetypical
shoulder on the second halo. Thisparticularity
is wellexplained
if one remembers that the covalentangle
inside an
elementary
tetrahedron is 109°28'which is close to the value of the pentagonangle, 108°,
encountered in thepseudo
icosahedral structure of the D.R.P.H.S. model[4].
The structure factor
S(Q)
ofglasses being govemed by
very fewpair distances,
the commonpart
of the 3 curves ismainly
made up of the firstneighbor
distance ri_ and the 1.65ri second distance which are common to all materials.
Following
thisobservation,
the ideaemerged early
that thepeaks
of covalentglass
pattems situated on the left side ofQi
should be derived from theD-R-P-H-S-, S(Q).
Therefore,
in thefollowing
discussion we shall callQo Prepeaks,
all thepeaks
which arelocated on the low
Q
side ofQi referring
to the lstneighbor peak
contribution toS(Q).
Bothexperimental
results and model calculations havesuggested
such a relation.On the
experimental
sideMangin
et al.[5]
have studiedamorphous Feji
_~~Si~alloys,
with xvarying
from 0.20 to I. The firstpeak
of the pure SiS(Q), emerged progressively
as a left shoulder ofQi
on a.FesiS(Q), increasing
with Si content.The continuous transformation from the D-R-P-H-S- to the disordered tetracoordinated
network,
the so called Random CovalentPacking (R.C.P.)
isaccompanied by
a drasticreduction in the local atomic
density,
which measures the number of atoms perA~.
Its value falls from 0.075 to 0.045.The
packing
fraction isequal
to ~ =l/6(wpo «~),
«
being
the atomic diameter. It is therefore reduced from 0.64 to 0.30.Using
theDebye
formula for the calculation ofS(q)
:S(q)
=
I + I/N
£
sinq.R,/q.R,
,
with
only
twodistances,
andtaking
into account the reduction from 12 to 4 firstneighbors, Mangin
et al. show that the emergence of theprepeak Qo
is linked to the reduction of thepacking fraction,
which in retum favors the 2nd distanceweight
on theS(Q) Profile.
Withoutusing
the wordexplicitly,
the hole model isalready implicit.
On the
modelling
side a similarapproach
has been usedby
Chaudhari et al. in 1975[6].
Using
the coordinates of the models of Steinhardt et al.(derived by
relaxation of the Polkmodel) [7]
and Connell Temkin[8]
for the disordered tetracoordinated network on the onehand,
and the Bennett model for the D-R-P-H-S- on the other hand[9] they
showed that onecan transform the first model into the second and vice versa.
In the first
transformation,
the four atoms of the first shell are removed in order to reach as new firstneighbors
the second shell of 12 balls, whileincreasing
thesphere
diameter.Conversely,
the second transformation is based on the decoration of some tetrahedra of the D-R-P-H-S-by placing
one atom at their center,consequently reducing
thesphere
diameter.The
Qo Peak
becomesQi
in the first transformation whileQi
becomesQo
in the second.S(i) A
~(%) B
a~ a a
3
al1.5
)ljl)11)1
~_~ ~_
=C;=~;;S
>' 3~
q~/~~T)~li
2
~'~'' 'l
I=I
AFT.5ij
,(=1.65A,f2
al ~l §=2 Apl _~~$
~
~~ b a~ b
~
l
Ql I A~=2
@
Rpl.65A,=3 o
~
ai
a~ V~~[~+~ #-',
O c ,
,
'
0 2
Ii /%~
o j 2 3
ij Iii
Fig.
2.IA)
Frommetglasses
to covalentglasses
the structure filiation a a.Ni-P b : a.Si ;c :
a.GeSe2.
All curves have been rescaled to Qi I-e- the first neighbor contribution to the total S(q). (B) Generation ofPrepeaks:
the hole model transformation of the D-R-P-H-S- into bondoriented structures. Calculated S(q) with the
Debye
formula:S(q)=Ai(sinqRj/qRj)+
A~(sin qR~/qR~)
+Ai(sin qRi/qR~).
(R values are the ratios ofexperimental
values to the lstneighbor
distance one). IA values are the relative
strengths
of theexperimental
G jr)). Insets equivalents in 2 dimensions (not related to the indicated R and Avalues)
: a) 2d closepacked
network ; b) the first hole subnetwork transforms a 6 fold coordinated model into a 3 fold oriented bond network ; c) the inflated 2nd hole subnetwork transforms theprevious
3 foldconnectivity
into a mixture of altemate 3 and 2 ones.Another
approach using
a decorationprocedure
has beenperformed by
Sadoc et al, onvarious
polytopes (polyhedra
in 4dspace).
Forexample,
if a vertex isput
at the centre of one among five of the 600 tetrahedral cells of the[3-5] polytope,
one transforms its 12-fold local coordination into a four-fold coordinatedpolytope (so
called « 240»),
which containsonly
even
rings
as in the C,T, model[10].
A recent contribution to the renewed
general
interest forprepeak studies,
anexplicit
hole model for the tetravalentglass
structures has beenrecently proposed by Bl£try [11]. Bl£try
uses a
computer algorithm,
which creates a closepacked
mixture ofspherical
holes and atoms of the same atomic size and concentration.Using
the Bathia-Thomton formalism to calculate the totalS(q)
ofbinary alloys
therepartition
of holes isoptimized
toyield
a maximum chemicalpartial
order between these atoms and holes. The calculatedS(q)
showsclearly
theQo Prepeak.
Nevertheless the shoulder of theQ2 Peak disappears.
Similar to the holechemically
orderedapproach
ofBldtry
for tetravalent monoatomicnetworks,
S. R. Elliott attributes theS(Q)
F-S-D-P- of covalentalloy glasses
tochemically
ordered holes around cation centered structural units[12].
Much earlier Warren hasalready
describeda.SiO~
as amore or less ordered
packing
ofSi04
tetrahedra[13]
and he succeeded inreproducing
theoretically
the first main halo of thediagram.
Following
thequalitative approach
ofMangin
et al, it istempting
toapply
it to a mixture of 2 elements of different valencies. Since we have to include those contributions of the M-R-O-which should be
responsible
for a newprepeak formation,
let us compare thepair
correlationfunction
G(r)
derived from theexperimental S(Q)
ofa.Co~ooP~o, a.Si,
anda.Gese~
(Fig. 3).
Theadvantage
ofcomparing G(r)
instead of R.D.F. is thatG(r)
is limited to the fluctuations of the atomic order around the mean atomicdensity.
Hence thecomparison
is not affectedby
thepacking
fraction reduction.At this
point
we would like to find the M-R-O-pair
distances whichmight
be related to thenew hole
model,
which isimposed by
the mixture of 4-2 coordination of Ge and Se. We have3
72 t c
~
~ l 3
ri/ri
Fig.
3. Pair correlation Function G jr) for : a) the metglass N180P20 (continuous line) b) amorphous Si (stars) ; c)amorphous
Gese~ (after 32)(interrupted
line). The first hatched zone at a 1.65 times theatomic diameter
Rj
is generatingQo
of a.Si while the 2nd hatchedzone at about 1.652 times Rj is at the
origin
ofQi
of a.GeSe2.already
seen that the decrease of the first shell occupancy from M-G- to a.Si resulted in an increase of the relative 2ndpeak height (first
hatched zone ofFig. 3). Similarly,
a newenhanced domain characterizes the
G(r)
ofa.GeSe2
ascompared
with that of a.Si(second
hatchedzone).
In order to introduce the remote distances of the second hatched zone in the
simplified
theoreticalS(Q)
we have added one more term to theDebye
formula ascompared
with what has been doneby Mangin
et al.Nevertheless,
thisapproach
remainsqualitative
sinceG(r) actually
consists of a continuum ofpair
distances rather than discretepeaks.
This remark hasalready
been made in the paper ofVeprek
et al.II-
So the Ai,
A~
and A~pair frequency
values, that we have used in thecalculation,
are somewhatarbitrary
even ifthey
aresupposed
to represent the relative
strength
of the 3peaks
onG(r).
It is shown infigure
28 that the introduction of a newcomponent
at 2.6 rigenerates,
asexpected,
a newprepeak
that we shall denoteQ(
since it should be considered as aprepeak
bis of the M-G-S(Q).
How can we
describe,
based on the D-R-P-H-S-model,
the 2nd hole modelground
rules ?Recalling
theprevious procedure,
I-e-reducing
from 12 to 4 the closeneighbors,
andtaking
into account the almost flat covalent
angle
Ge-Se-Ge(144°
inaverage)
weprocede
as follows(Fig. 4)
: after the lst holerepartition
has been made we repeat it on the 12 2ndneighbors
of the tetracoordinated network which hasjust
been created. Indoing
so we create a new holesubnetwork with a
larger
cellparameter
and a 4,2 mixture ofbonding connectivity.
/
Fig.
4. The schematic hole inflation full circles : 4 fold coordinated atoms ; stars : 2 fold coordinated atoms open circles : holes. The reduction of 12 to 4neighbors
in the lst shell isrepeated
in the 2nd shell.A two dimensional
equivalent
model is shown in the insets offigure
28. From the closepacked (necessarily ordered)
2d network one can form a subnetwork with a coordination number of threeby connecting
the atoms left over from theregular
first hole distribution. Theprocedure
is thenrepeated
on the atomic subnetworkhaving
the same parameter as the lst hole subnetwork. One can see that we generate a newlarger
hole subnetwork which is made up of analtemating
3-2 bondedconnectivity.
In summary, one can conclude
that,
at leastqualitatively,
thegeneral
features of covalentglass
models can be derived from theD-R-P-H-S-, through
a kind of self similar hole distributionprocedure,
which generates at each step acorresponding prepeak.
Therefore theprepeak
formation does notrequire
apriori
additionalphysical origin hypotheses
such as 2d entities or «superstructural
» units which shouldbe,
ifanything,
aby-product
of the hole model.This is not the case for another
intriguing prepeak
case, TheS(Q)
ofamorphous
phosphorous,
arsenic and antimonium are shown infigure
5. Anamazing point
to notice isthat,
while a verystrong prepeak Q(
is visible on the a-PS(q)
its relativeintensity
isdrastically
da
P
ao a~
ao
°'
As
Ql ao
Q<
1
-1
Fig.°(~)
c)
(after 16) : thevanishing of
Qo going a-P to a.Sb.reduced to the a-As case and that it
disappears permanently
from the a.Sb structure factor. It isworthwhile, nevertheless,
to add thatbeyond Q
i the Sb
S(q)
has aprofile
which hasnothing
to do with the
« shoulder
»
shape
of theprevious diagrams.
Whengoing
downthrough
the Vth column it is indeed well known that the covalent character of the elements is reduced infavor of a metallic
tendency.
The atomic structure
modelling
of thesesingle
elementglasses
has been apuzzling
task fora
long
time. Theproblem
of theprepeak generation
iscomplicated by
the existence of a 2dlayered
structure in thecrystalline
state. Puckeredlayers
of covalent bonded 3 atomtriangular pyramids
arepacked
with alayer spacing depending
on theintensity
of the Van der Waals bonded atomsbelonging
to two differentlayers.
Many
studies have been devoted to thissubject [14-18].
Smith et al.[17] pointed
out the verystrange
fact thatusing
amicro-crystalline
model of orthorhombic or rhombohedralarsenic does not
yield
the calculatedS(q)
with aQi Prepeak. Nevertheless,
thisghostlike prepeak might
be createdby slightly increasing
theinterlayer spacing
of themicrocrystalline
model.
Then it becomes clear that the
prepeak origin is,
in this case, more a textureproblem
than atopological
one. Thebreaking
of thepuckered layer packing
leaves some kind of cavernsseparating
twomicrolayers
whose lateralexpansion
is at a nanometric scale. For this reason the sth column case can be classified in thepacking
fractiondependent prepeak
class.2.
Prepeak profile
modifications : theirrelationship
to the medium range order fluctuations.Now that we have learned how to
generate
aprepeak
onS(q) by reducing
thepacking
fraction of atomsthrough
a uniform distribution of holes in matter a newquestion
arises : for agiven packing
fraction what are theparameters
that canmodify
theposition
of q andintensity profile
of thecorresponding prepeak
? Thisquestion
isobviously
related to themetastability
of
glasses
and to theexpected
existence of a minimum minimorum for their freeenthalpy.
It is alsolikely,
apriori,
that thephysical properties
ofglasses
will bequite dependent
on theM-R-O- and
consequently
on theprepeak
behavior. This is the aim of this new section.2.I PREPEAK
Qo
AND CHEMICAL ORDER IN METALLIC GLASSES REVEALING THE PREPEAKBY A WEIGHTING EFFECT OF THE SCATTERING LENGTHS BETWEEN THE ALLOY CONSTITUENTS.
In this case,
Qo
is notgenerated by
the hole model. M-G- areclose-packed
structures.Even if the differences of sizes between the
alloy
constituentsmight weakly
contribute to theprepeak formation,
its mainorigin
arises from theaffinity
between hetero-atomsyielding
chemical order electronic fluctuations.
Historically
the term «prepeak
» has been used Flrst in the M-G- case while the F-S-D-P-appeared
inamorphous
covalent elements andalloys.
Theprepeak
of M-G. isalways
a weak shoulder ofQi
even in the case of anoptimized
contrastexperiment
where the constituents have very differentscattering lengths.
The M-G-
prepeak
was revealed for the lst time in a.Co~oP~o, by
Sadoc et al,[19]- Using coupled X-ray
and neutron diffractionexperiments,
thepartial
R-D-F- werederived, showing clearly
a chemicalaffinity
between hetero-atoms-Many
«partials
» have since been derived for various M-G-composed
either of metals and metalloids or of 2 different metallic transitionelements such as
Cu, Zr,
Ti etc.Preparation dependent inhomogeneities
have been studied aswell,
but the scale of the volume involved islarger
than thatrequired
forprepeak modifications,
I-e. 20,50/k
instead of 5-10/k- Hence,
thecomposition
anddensity
fluctuations within materials such asCu~oZr4o prepared by
cathodicsputtering (as compared
to melt spunsamples)
are revealedby
thepresence of a new «
prepeak
» in the SmallAngle Scattering
domainIS-A-S.) [20].
As far as
magnetic amorphous alloys
areconcemed,
M-R-O-compositional
in-homogeneities
areresponsible
for both nuclear andmagnetic
order diffractionrings
at lowangles.
Boucher et al- haveextensively
studied thisproblem
in rare,earth basedalloys, by
neutron S-A-S-
experiments [21]-
Ina-Tb~~Cu7~
forexample,
aring
appears at q = 0-2i~
which is
interpreted
ascoming
from small aggregates of Tbseparated by
50/k-
Let us remark at this
point
the verycomplex,
butinteresting
transition frompoint
defects tolarge
scaleinhomogeneities
in the direct space which are detected in thereciprocal
spacewithin the « no man's land domain » between S-A-S- and
Large Angle Scattering IL-A-S-)-
It is not asimple
task tocorrectly
describe the passage fromtopological
order to thelong-
range ordered
compositional
waves of aspinodal-like decomposition.
In thereciprocal
spacethe observed «
prepeaks
» at smallangles
aregenerated by
constructive interferences between beams issued from individualparticles
instead of atoms.It is worth
making
acomparison
withequivalent phenomena
in thecrystalline
state. It is well known that well definedscattering rings
appear sometimes in the S-A-S- domain whenlong
range electronic fluctuations establish a 3dperiodic superlattice,
as in thespherical
Guinier-Preston(G-P-)
zone ofAl-Ag alloys [22].
2.2 PREPEAK AND TOPOLOGY i THE EVEN FOLD RING
(E.F.R.)
AND THE ODD-EVEN FOLDRING
IO-E-F-R-
MODELS FOR DISORDERED TETRACOORDiNATED MATERIALS. Since the firstattempts
to construct disordered models for a.Si or a-Geusing
sticks and balls it hasbecome clear that there were 2
possible
solutions to connect theelementary
tetrahedra withoutintroducing
too much strain energy into the models.The
generated rings
can include either more or less atoms than the 6 membered « chair » or« boat
» like
puckered rings
of thecrystalline
structures. Asalready mentioned,
Polk built the first modelinvolving 5, 6,
7 memberedrings
that we shall denote here as the O-E-F-R-model,
while Connell and Temkin
proposed
ahomogeneous
six memberedring
model denoted hereas the E-F-R- model.
It was believed that while the O-E-F-R- model
might
wellrepresent
the local structure of a-Si ora-Ge,
the E-F-R- was betteradapted
to III-Vmaterials,
where the local electricalneutrality
couldonly
beproduced by using
an altemate evenring
solution. If the atomcoordinates of the 2 models are introduced in the
Debye
formula their theoreticalS(q)
are derived : what kind of differences should appear on theirrespective profiles
?They
are, in
fact,
concentrated in the relativepositions
of the various diffraction halos. We can see that anopposite
shift movesQi
andQo
towards eachother,
whengoing
from the E-O-F-R- to the E-F.R, modelS(q) (Fig- 6)-
~O al
220 311
ill
Q (
ill
O 1 2 3 4
°(i~~)
Fig.
6. TheoreticalS(q)
for the O-E-F-R- (Polk) model : continuous line and the E-F-R- (Connell- Temkin) model : dashed lines. The vertical arrows indicate thepositions
of thecrystalline
dspacings
in Si. The horizontal arrows show the shifts of theQo
andQi Peaks.
In table I we have
gathered
the ratiosQi/Qo
for variousexperimental S(q)
ascompared
with the theoretical values for the E-O-F-R- and E-F-R- models. It is rather
amazing
thatexperimental
casesreproduce clearly
the 2simple
ball and stick models[23].
Asexpected,
the E-F-R- model is encountered incompounds
such as GaAs-More
spectacular
is the case ofa-Sii _~Sn~ alloys
whose low Sn contentsS(Q)
are of the O-E-F-R-type
while the E-F-R-progressively
dominates whenapproaching
theequiatomic composition, revealing
verysimply
an obvious chemical order[24].
Coming
back tofigure 6,
it is worthwhile to focus on theQo
location ascompared
with the I11 line of the diamond closepacked interspacing,
The O-E-F-R- modelQo
isdefinitely
onthe low q side of the 111
line,
while on thecontrary
the E-F-R- modelQo
has drifted on itshigher
q side. The atomic radiusbeing
the same, such a differencesuggests
that theinterspacing
distance between fictive denseplanes
islarger
in the first model than in thesecond,
so that the localdensity (say
inside the first 3shells)
should be greater in the E,F-R-network. At the same
time,
due to theuniqueness
of the 6 foldvariety
ofrings,
alarge
dihedral
angle
statistical distributionyields
more strain energy into the E-F-R, model.Table I. The ratio
Q i/Qo for
tetracoordinated models andexperimental amorphous
elements andalloys.
Qi/Qo Qi/Qo
References(models) (experimental)
O-E-F-R-
(Polk) [7]
Relaxed 1.82
[6]
a.Si
IS-P-)
1.736[27]
a.Si : H
IS-P-)
1.797[27]
a.Si : H
(G-D-)
1.849[27]
a-Si~osnio
1.796[24]
E-F-R-
(Connell-Temkin) [8]
Relaxed 1.76
[28]
Unrelaxed 1.67
[28]
a.GaAs 1-71
[23]
a-Ga-P 1.70
[23]
a-Gasb 1.70
[23]
a-InP 1.67
[23]
a-
Si5oSn5o
66[24]
These overall features induce model
dependent
constraints on the medium range order level which areexpected
to be reflected on the band structure characteristics,Ching
et al.[25]
have studied this effect
theoretically, using
the well knowntechnique
oforthogonal
linear combinations of atomic orbitals(OLCAO).
The calculation shows alarge
band gap difference between the 2 models. The greater the localdistortion,
the smaller thebandgap (2.19
eV for O-E-F-R- and 1.24 eV forE-F-R-)- Despite
thelikely over-simplified approximations
of thecalculation,
it is demonstratedqualitatively
thatlarge
electronic property differences can begenerated by
two differenttopological
M-R-O-building rules,
evenleading
to an atomicassembly having
at amacroscopic
scale the samepacking
fraction. At theexperimental
level itcan be stressed that a very
simple
criterion I-e- theQi/Qo
ratio canyield significant
information
regarding
the M-R-O,2.3. PREPEAK AND NETWORK RELAXATiON : a.Si AND a.Si-Ge ALLOYS. In this section we
enter an
important
field of interest withrespect
to our purpose since it deals with a direct correlation between theQo Peak
behavior and somephysical
property variationsdepending
on the
glass preparation
conditions. We shalldevelop
in detail the case of tetracoordinated materials which are of both theoretical andpractical
interest.A considerable amount of studies has been devoted
during
thepast
20 years toa-Si,
a.Si : H thin films as well as to theiralloys
with C and Ge elements.Finding
the freeenthalpy
minimum minimorum of the system is crucial for some
applications
such as the determination of the maximumefficiency
ofphotocells.
Indeed the
photocarriers
lifetime isdrastically
reducedby
the localized states in the valence and conduction band tailsgenerated by
the atomic disorder.A
4
Q
+M 3
~
~
W~ 2 ~
~
I i
b
~'o.5
I-o I.5Energy ie.V.)
B
i
u
/
#
#
~~ j
~
'fo.,
Energy le.V.I
Fig.
7.Optical
spectra obtainedby
Photo-DeflectionSpectroscopy
(P.D.S.). IA)Amorphous
Si a) pure Si IS-P-) b) a.Si: H (G.D.) c) a.Si : H IS-P-) ; a and b
samples
are the same as those offigure
9. (B)Amorphous GeSe2
a)deposited
film b) annealed film ; a and bsamples
are the sameas those of
figure
9.Among
the various methods ofdeposition,
thedecomposition
of silanesby
theglow discharge technique (G-D.),
and the reactive cathodicsputtering IS-P-)
aretypical
methods ofpreparing
a-Si:Halloys.
Due to the nativehydrogen
formed in the S-P- method thistechnique yields
a greatvariety
of Hbondings
such asSiH, SiH~,
SiH clusters, etc. butunfortunately
the energy of the extracted atoms from the target is such thatthey produce
apermanent bombardment of the
deposit inducing
irradiation defects.It is therefore useful to make a
comparison
with thesample
characteristicsprepared by
the G-D- method which is not soenergetic.
It has beenproposed
that besides the reduction of the a-Si electronic gap statesthrough
thedangling
bondsaturation,
H atoms alsoplay
a role innetwork
relaxation,
I,e. minimize the bandtailing
effect.The tail states are
conveniently
characterizedby optical absorption
spectra. This is thereason
why
we have undertakencoupled X-ray
and PhotodeflectionSpectroscopy (P.D.S.)
studies on
samples prepared
under the same conditions.The P-D-S-
technique
measures theoptical absorption
under theabsorption edge,
due to gap states and transition from extended states in the one band to localized states in the tail of the other band. Theoptical
spectra of bothhydrogenated
andnonhydrogenated samples
areshown in
figure
7A : a drastic reduction of the subband gapoptical absorption
is observed when a.Si ishydrogenated.
This reduction is nevertheless more substantial in G-D-
samples
than insamples
of S.P. Theoptical
orderparameter
isEo,
the inverseslope
value of the linearpart
of theabsorption edge (Urbach tail).
The more relaxed thematerial,
the smaller the value ofEo,
I-e- the steeper theedge.
On the structural
side,
a verysimple
relevant order parameter is the coherencelength L,
I-e- the size of the volume in whichX-rays
« see » an ordered network. L is derived from the
Full Width at Half Maximum
IF-W-H-M-)
ofQo-A(2b) by
the formula:L
= 0-9
A/(A(2 b)
x cosbo)-
We have shown in
(26)
that Lactually
decreaseslinearly
withEo (Fig- 8)- However,
thebonding
type of H(identified by
infra,redspectroscopy) plays
a determinant role in the network modification. The atomic fraction of Hbeing larger
than several percent, the saturation of thedangling
H bonds is achieved in all materials.w
w .
Y _
@~
o
deterioration of the structural and
optical
orderparameters
: hereSiH2 dihydrides
dominate the I,R.spectra.
In another
study (27)
is has been stated that the network relaxation also modifies theprepeak position (Fig- 9a).
Aspointed
out in theprevious
section theQo
location relative to the c,Si I I I dspacing
is an indirect measure of the bound distortions. Thus one canpredict
that all
parameters releasing
the strain energy should shiftQo
towards lower q values. It can be seen that theprepeak
of thehydrogenated sample
is in fact shifted from theright
to the left side of the theoretical 111 line. Butcontrary
to thetopological effect,
theQi position
is notmodified
by
the relaxationphenomenon.
Now,
ifalloys
a-SiGe : H areprepared by
these 2techniques, Qi
follows a linearVegard's law, independent
of thepreparation conditions,
whileQo
on thecontrary
isquite dependent
on them
(Fig- 10)-
A numerical simulation has been
performed by
Mosseri[28]
of the minimization of the strain energy in a-Si-Gealloys.
Thealloys
were made upby randomly substituting
Ge atoms for Si atomsii-e-
with alarger
atomicdiameter)
within the O-E-F-R- and E-F-R- models. If the theoreticalpredictions
agreequite
well with theexperimental
behavior ofQi (reflecting
the mean atomic size
variation,
a function of thecomposition),
theQo
evolution with x on thecontrary
exhibitslarge preparation dependent
fluctuations ascompared
with the theoreticalpredictions.
The
alloy
networktopology
remains of the O-E-F-R-type
but theslopes
as well as the absolute values of theQo positions obey
different linear laws.These results suggest the interest of
investigating
the M-R-O- differences in the direct space associated with their localized counterparts onQo
in thereciprocal
space. This has been doneby
anomalous diffusionscattering experiments
on the Ge Kedge
whichyield
the differential R.D.F, It is shown that H reduces the dihedralangle
distribution betweenadjacent
tetrahedra in thealloy network, reducing
the number ofpair
distances between the 2nd and 3rd shells ofatoms. It is also confirmed that the M-R-O- atomic order in G-D-
samples
is more relaxed than in S-P.[29],
Finally
thesharpness
ofQo
as well as the ratioQi/Qo
ofamorphous
tetracoordinated semiconductorsS(q)
should serve as verysimple
tests for the calibration of thedegree
of reduction of theoptical
and electronical active defects in these materials.j~
,
Qi ~
Q
°0
= ~
i I d
I I
~ ~
~ ~
~ ~
uJ uJ
~ ~
~ ~
+ &
b
+ +
0 lo 20 30 40
e deg_] 4 i~ 2D 2e ~
Fig-
9.Experimental X-Ray
pattems of :IA) amorphous
Si films, a) a-Si :H ; b) a.Si : the small vertical arrow marks the Ii ii) c.Si line. F-W-H-M- is the Full Width at Half Maximum which is used to calculate the coherencelength,
L offigure
8-(B) Amorphous GeSe2
films. a)deposited GeSe2i
b) annealed GeSe2.i
2.05~
-Si,
~
~~i
~3
o
Fig. 10. The effects of
hydrogen
and the modes ofpreparation
on Si~i_~~Ge~ filmsQj
andQo
variations with x.Squares sputtered
pure Si ; full squaressputtered hydrogenated
SiIS-P-)
full circles :hydrogenated
Sidecomposed
from silanes (G.D.) ; dotted lines : theoretical simulations after(27)
(P stands for Polk and CT forConnell-Temkin)-
One can see thatQi obeys
the same law for allsamples,
which is in agreement with the theoretical results, besides aslight
shift. On the contrary,Qo
isstrongly dependent
on the H content and thepreparation
conditions. Their behaviors are alsodifferent from the calculated values.
2.4 PREPEAK
Q(
AND GLASS HOMOGENEITY. THE PHASE SEPARATION IN IV VI GLASSES.We shall now focus on a more remote
part
of the M-R-O- at about three times the firstneighbor
distance. Henceforth we aredealing
with the secondprepeak Q(
of the D-R-P-H-S- At this near-nanometric scale the M-R-O- can begovemed
notonly by topological
features but alsoby
localcomposition
fluctuations as well asdeparture
from chemicalordering.
Thissituation is encountered when
alloying
elements A of the Ivth or Vth columns with those B of the Vlth column. Let us examine the IV VI case. Numerous studies have been devoted to thestudy
of the atomic structure ofSi,S, Si-Se, Ge,Se, GeTe,
etc. It has often been observed inAi
~B~alloys
such as Gese that aQ( prepeak
isprogressively emerging
on the left side ofQo
when thecomposition approaches
the stoechiometricAB2
value[30].
The Gese case has received considerable attention in the
past.
Its firstsharp
diffractionpeak (Q()
has been considered as « anomalous »by Phillips [3 II
andnecessarily
associatedwith molecular
clusters,
whose center-to-centerspacing
should be of the order of 5I
as
derived from the
Q( position. Ruling
out amicrocrystalline
model since the F-S-D-P- subsists above themelting point,
alayer-like
structure is neverthelessproposed
as a reminiscence of thehigh
Tcrystalline
from ofGese~-
It is the well known «outrigger
» raft model whichconnects
GeSe4/~
tetrahedra but breaks the chemical order at theedge
of theraft, making
Se- Se wrong bonds. This model wasopposed
to astrictly chemically
ordered oneconnecting
alsoGeSe412
tetrahedra but linked to form a 3d network,In a recent
study [32]
Fischer-Colbrie et al. tried to elucidate theproblem by studying
the structure evolution ofGese~ glasses
with thesample
thickness. As astarting assumption
thelayer
structure should be favored in very thin filmsla
fewI thickness),
while alikely isotropic spherical tendency
should beexpected
in bulksamples.
Despite
thehigh precision
of theexperimental
data(high
flux of theX-ray synchrotron
radiation source and
grazing incidence)
the authors did not detectsignificant
differences between the spectra. After a detailedinvestigation
of the M-R-O- and the failure ofmatching
the
S(q) by layered models, they finally
obtain agood
fit with a 3dspherical microcrystalline
model except for the
Qi Position
whichrequires
aslight
increase of thecrystal layer spacing
of theexperimental
value. We havecarefully compared
theQ(
values for their thinsample
and bulk material. Theposition
is shiftedslightly
from Ii~
to 0-94i~
for the thinsamples.
So therequired interspacing
increase shouldagain
belarger
for the thin filmS(q) matching.
More
recently,
an intermediate case has been obtained inrelatively
thickevaporated
films of I micrometer[33].
It can be seen infigure
98 that thedeposited
filmQ(
can beroughly
decomposed
into 2subpeaks
which tum out to be located at 0-85i~
and Ii~
~,
respectively.
Remembering
the above remark it islikely
that the smallerangle component corresponds
to thebeginning
of thedeposit
when it was verythin,
while thelarger angle
component should be attributed to an almost bulk state. But theimportant
fact to underline is that, afterannealing
at T= 330 °C the lst component is
drastically
reduced and appears as a weak shoulder of theQ( peak
at Ii~
~-
We can conclude from this that the
possible decomposition
ofQ(
into 2subpeaks
which behavedifferently
onannealing clearly
demonstrates that 2 different solutions for the M-R-O-can exist, Are
they representative
of the 2d and 3d models ? More work is needed to answer thisquestion.
The
investigation
ofphysical properties might bring
about somehelp
to make a choice.Unfortunately, divergent
conclusions are drawnthrough
ultra-sonic results madeby
Bellessa et al.[34] being interpreted only
in a 2dframework,
and X-P-S-experiments by Theye
et al.which rule out the
outrigger
raft model since no wrong Se-Se bonds are detectedby
thistechnique.
The
optical spectral
behavior onannealing
exhibits an increase in theoptical
gap of 0.3 eVaccompanied by
an increase of thepseudo
gap defect state contribution to the low energyoptical absorption (Fig- 78).
In this case, the evolution is very different from whathappened during
the relaxationphenomena
of a-Si and a-Si-Gealloys
which wereaccompanied by
adecrease in the gap defects without any
change
in their gap width. This is aparadoxal behavior,
sinceannealing
of a vaporquenched
materialusually
releaves strains and local defects.It is also
unexpected
that theQi Profile
exhibits acomplex
structure in theas-deposited
state. This is nevertheless consistent with the observed evolution of
annealing
towards asingle
stoichiometric
phase
which is known tocorrespond
to thelargest optical
gap of thealloy.
Conclusion.
We have seen that the atomic structure of covalent
glasses
can be derived from the disorderednetwork of a
compact
hardsphere packing,
The hidden 5 foldsymmetry, goveming
thegeneral
features of the M-G-pattems
is retainedduring
the reduction of thepacking fraction, through
aduality
transformation from apseudo-icosahedral
environment towards apseudo-
dodecahedral environment.
Prepeaks
appear atdecreasing
q valuepositions according
to the homothetic inflation of the holegeneration procedure.
Remembering
that the low q domain of diffractionpattems
reflects the medium andlong
range
order, prepeak profiles
andpositions
are very sensitive to thetopology,
relaxationphenomena
and to thehomogeneity
of covalentglasses.
The various
origins
of local order fluctuations do reactdifferently
toS(q) affecting
eitherthe whole
diagram
oronly
theprepeak shape and/or position.
At the same
time, physical properties
exhibitparallel
fluctuations.Prepeak
behaviors reflect at leastqualitatively
these variations.The difficulties
begin
whenquantitative
evaluations are undertaken in theoretical studies.Beyond qualitative trends,
the true gap, the valence and conduction bandprofiles
and tail width need to becalculated,
a moreprecise participation
of theM-R-O-, involving
severalhundreds or even thousands of atoms. Are theoretical calculations able to fulfil these
implements
? Newapproaches
are in progresstaking
into account moreprecisely specific bonding
and dihedralangle
distortions[35].
It is now reasonable to foresee that fruitfuldevelopments
on the band structureknowledge
of covalentglasses
will arise from the use of M-R-O- detailsprovided by
fits of the exactprofiles
ofprepeaks-
Acknowledgements.
I would like to thank R. Mosseri for fruitful discussions and
help
in thegeneration
of thegraphics
of the 2D models.References
[ii VEPREK S., BEYELER H. U., Philos.
Mag.
B 44 (1981) 557.[2] Moss S- C., PRICE D. L.,
Physics
of disordered materials(Plenum Publishing Corporation,
1985).[3] GUINIER A., Thdorie et
Technique
de laRadiocristallographie
(Dunod, Paris) p. 452.[4] SADOC J- F., DIXMIER J. and GUINIER A., J. Non
Crysl.
Solids 12(1973)
46.[5] MANGIN Ph., MARCHAL G., RODMACQ B-, JANOT Chr., Philos. Mag. 36 (1977) 643.
[6] CHAUDHARI P., GRACzYK J. F., HENDERSON D-, SrEINHARDT P.
(1975)
727.[7] POLK D. E., J. Non
Cryst.
Solids 5 (1971) 365.[8] CONNELL G. A. N., TEMKIN R. J.,
Phys.
Rev. B 9 (1974) 5323.[9] BENNETT C. H-, J. Appl.
Phys.
43 (1972) 2727.[10] SADoc J. F. and MOSSERI R., Philos. Mag. B 45 (1982) 467.
[11] BLLTRY J-, Philos.
Mag.
B 62(1990)
469.[12] ELLIOT S. R.,
Phys.
Rev. Lett. 67 (1991) 711.[13] WARREN B. E.,
Phys.
Rev. 45 (1934) 657.[14]
BEYELER H. U- VEPREK S., Philos.Mag.
B 41(1980)
327.[15] GREAVES N-, ELLIOTT S- R. and DAVIS E. A-, Adv.
Phys.
28(1979)
49.[16] VON KREBS H. and STEFFEN R., Z.
Anorganische Allgemeine
Chemie. Band 327 (1964) 224.[17] SMITH P. M-, LEADBETTER A. J., APLING A. J-, Philos. Mag. 3
(1975)
57.[18] BELLISSENT R, and TOURAND G., J.
Phys.
France 37 (1976) 1423.[19] SADOC J. F. and DIXMIER J., Mater. Sci.
Eng.
23 (1976) 187.[20] NAUDON A- and FLANK A. M., Proc. 4th Int. Coat. on
rapidly
Quenched Metals (1981) p. 425.[21] BOUCHER B- and CHIEUX P., J.
Phys.
Cond. Matter 3(1991)
2207.[22] WALKER C- and GUINIER A-, Acta. Met. 1
(1953)
568.[23] DIXMIER J., GHEORGHIU A., THEYE M. L-, J.
Phys.
C : Solid StatePhys.
17 (1984) 2271.[24] VERGNAT M,, MARCHAL G-, PIBCUCH M- and GERL M., Solid State Commun. 50 (1984) 237.
[25] CHING W. Y., LIN C. C., GUTTMAN L.,
Phys.
Rev. B 16 (1977) 5488.[26] ESSAMET M., HEAP B., PROUST N. and DIXMIER J., J. Non
Cryst.
Solids 97-98 (1987) 191.[27]
DIXMIER J-, ELKAIM P-, CUNIOT M-, LABiDi H-, CHAHED L., GHEORGHIU A., THEYE M- L-,Proc, of the 8th- Eur. Photov. Sol.
Energy
Conf- Florence(1988)
976.[28] CUNIOT M., DIXMIER J-, MOSSERI R., J. Non
Cryst.
Solids 97,98 (1987) 171.[29] BOUCHET,FABRE B., ELKAIM P., CUNIOT M., DIXMIER J-, Proc, of the 10th Eur. Photov. Sol.
Energy
Conf. Lisbon (1991) p. 950.[30] FAWCETT R. W., WAGNER C. N. J. and CARGILL G. S., J. Non
Cryst.
Solids 8,10 (1972) 369.[3 II PHILLIPS J. C., J. Non Cryst. Solids 43 (1981) 37.
[32] FISCHER-COLBRIE A., FUoss P. H., J. Non
Cryst.
Sol. 126(1990)
1.j33] THEYE M. L., KOTKATA M. F., KANDILL K. M., GHEORGHIU A., SENEMAUD C., DIXMIER J. and PRADAL F., I-C-A-S- 14 Garmisch-Partenkirchen
(August
19-23, 1991).j34] DUQUESNE J. Y., BELLESSA G., Eur.
Phys.
Lett. 9 (1989) 453.[35] DAVIDSON B. N., LUcovsKY G. and BERNHOLE J., I-C-A-S- 14 Gamlisch-Partenkirchen
(August
19-23 1991).