HAL Id: jpa-00208662
https://hal.archives-ouvertes.fr/jpa-00208662
Submitted on 1 Jan 1977
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.
A simple approach to covalent surfaces
F. Flores, C. Tejedor
To cite this version:
F. Flores, C. Tejedor. A simple approach to covalent surfaces. Journal de Physique, 1977, 38 (8),
pp.949-960. �10.1051/jphys:01977003808094900�. �jpa-00208662�
A SIMPLE APPROACH TO COVALENT SURFACES
F. FLORES and C. TEJEDOR
Instituto de Fisica del Estado Sólido
(CSIC
andUAM)
Univ.
Autónoma, Cantoblanco,
Madrid34, Spain (Reçu
le23 fevrier 1977, accepté
le 19 avril1977)
Résumé. 2014 La surface d’un système covalent a été étudiée avec un modèle unidimensionnel dans le cadre de
l’approximation
d’une bande interdite étroite. Notre analyse montre : i) que ladistribution et le
potentiel
dans larégion superficielle
se comportent de façoncomparable
à ceuxd’un métal, et ii) que l’état
superficiel
correspondant à la liaison rompue est déterminé par unerègle
de somme basée sur la neutralité de la charge. Nous discutons le cas d’une surface 111 et ilapparaît que les conclusions de l’analyse précédente sont valables pour cette surface. En utilisant
ces résultats, le
potentiel
de surface du Si-111 a été obtenu etcomparé
de façon satisfaisante à d’autres calculsthéoriques.
Abstract. 2014 A 1-dimensional covalent surface is first studied
selfconsistently
within the narrowgap
approximation.
Our analysis shows : i) that the potential and charge distribution in the surfaceregion behave rather like those for the case of a metal, and ii) that the
dangling-bond
surface stateis determined by a charge neutrality sum-rule. A covalent 111-surface is then discussed. Our arguments strongly suggest that the same conclusions apply to this surface. By using these results, the 111-Si
surface
potential
has been obtained andcompared satisfactorily
with other independent theoretical calculations.Classification
Physics Abstracts
8.322
1. Introduction. - Current studies of the electronic
structure of covalent semiconductor surfaces
usually
involve a selfconsistent numerical solution of the
Schrodinger equation
for agiven
assumed surfacegeometry,
with the bare ionpotentials
of the semi- infinitecrystal plus
a suitable localpotential
to takesome account of
many-body
effects[1].
While a great deal of information has been obtained about surfaces in these calculations
[2, 3], perhaps
itcould be of interest to look for some
general
relationsbetween different surface
properties
whichmight
be contained but unnoticed in the selfconsistent numerical
computations
and whichmight
alsoapply
to other situations. This has been the
point
of viewmotivating
this work.In this paper we first show the two
following
resultsfor a 1-dimensional covalent surface :
i)
Thepotential
andcharge
distribution near the surfaceregion
behave rather like those for the case of ametal
(the
necessaryqualification
will begiven presently).
ii)
Thedangling-bond
surface state is determinedby
the condition of surfacecharge neutrality,
withindependence
of the surfacepotential
details.Later, we suggest that the same results can be
generalized
to a( 111 )
covalent surface.These surfaces will be discussed in the next two
sections,
while an illustration of these results will begiven
in section4,
where the Si-111 surfacepotential
is obtained in
good
agreement with selfconsistent calculations.2. A 1-dimensional model. - Let us start with a
1-dimensional
model,
where our results can beexactly proved.
First ofall,
it is of interest to considera metallic surface and derive its
density
of surfacestates, as well as a type of Friedel sum rule. In
figure
1 awe have drawn this model with a
hypothetical
surfacepotential.
In the
region
z >0,
well inside thecrystal,
thewavefunction for a
given energy E
can be written aswith E =
2 À,2.
To
study
thedensity
of surface states weproceed
intwo steps. First we
imagine
ahypothetical
infinitebarrier at z =
0,
and another one at z = L, with Las
large
as necessary. In thishypothetical
case, thewavefunctions are
given by
and the
eigenvalues
are obtained from :Article published online by EDP Sciences and available at http://dx.doi.org/10.1051/jphys:01977003808094900
FIG. 1. - 1-dimensional model of (a) a metal and (b) a two-band
semiconductor with a non-abrupt potential across the interface.
Notice that there is no bulk state for n = 0, since this is the conduction band
edge.
What we are inte-rested in is the
density
of states, which can be obtainedby comparing
our system with one oflength
2 Lwith Born-von Karman
boundary conditions, i.e.,
with no surface effects. The allowedeigenvalues
wouldbe
given by
2 LÀ= 2 nn, with n=O, ± 1,± 2, ..., + N.
If we denote the
corresponding density
of statesby A’2L,
then thedensity
of states for a 1-dimensional metal oflength
L and infinite barrier on both sides is :Now the defect in the
density
ofstates -1/ 2 6(E)
is to be shared between both surfaces. Thus the
density
of surface states
(S.S.)
on one surface is :Notice that this result
implies
a defect of2
of thecharge
of an electron at the surface of a 1-dimensionalmetal with an infinite barrier at z = 0. This can be
easily
checkedby looking
at the total surfacecharge
of this model
p(z),
since :where po and
kF
are the bulkcharge
and the Fermi momentumrespectively,
andNow we can
imagine
that thegiven
system with anarbitrary
surfacepotential
can be obtainedby lowering
the
initially
infinite barrier.Knowing
thedensity
of states for the case of infinite barrier at z =
0,
we can
easily
evaluate thedensity
of states for thesystem described
by
the wavefunctions(1).
In this case, the
eigenvalues
mustsatisfy
the condi-tion
no
being
thephase-shift produced by
the surfacepotential.
Comparing (6)
and(3)
we find the extradensity
of states introduced when the infinite barrier is lowered :
Then
Now the
important point
to be noticed is that, sincecharge neutrality
mustalways
beobeyed,
for any selfconsistent surfacepotential
we must havea sort of Friedel sum rule.
Let us now consider the semiconductor case.
The model is shown in
figure
1 b. In the bulk, a weakpseudopotential Veigz
+ c.c.(g
=2 7r/a)
creates anassumedly
narrow gap. Thispotential
variessmoothly
across the
interface,
up to the vacuum level outside thecrystal.
We take a metallicplane
at z = a, where theeffect of the surface barrier is
negligible.
In orderto
study
the wavefunctions inside thecrystal (z
>a),
we measure
energies
s from themidgap
energyt(g/2)2
anddistinguish
thefollowing
ranges :o s ) I I
V1.
Inside the gap the evanescent wavefunction is. 1 V I I s I 2-(g/2)2.
Just inside the lower band.Measuring crystal
momentum A from the B.Z. boun-dary g/2,
we havegA
= 2E2 - V2.,
and ’1 V I , I s I i(g/2)2. Further
down inside the lower band we haveThe wavefunction
(11)
becomes the free electron wavefunction :Outside the
crystal (z a), going
downsmoothly
from the vacuum
level,
as zapproaches
thematching plane,
we have aregion
in which thepotential begins
to oscillate so as to become the weak bulk
pseudo- potential.
Under these circumstances, in theregion
close to z = a, within a
length
of order a, sincethe weak
pseudopotential
satisfies the conditionI v I a
g, the effect of the smallripples
isnegligible compared
with the main smoothpotential profile.
We can then
write,
to agood approximation,
thatfor z :5 a the wavefunction is
where
and il’
is thephase
shiftproduced by
thepotential
which starts from the vacuum level and then
joins smoothly
the dashed line infigure
lb.Knowing il’
asa function of energy we use this
i)
to find the S.S.eigenvalue,
andii)
to find thephase shift il
of(11).
Thus, matching (13)
and(10) :
where
Ev
andEc
are the lower and upper bandedges,
and Vg2
has been used. This is the secularequation
for the S.S.
eigenvalue. Also, matching (13)
and(11) :
This
gives q
and hence fixes the wavefunction(11).
We now
study
thedensity
of S.S.proceeding
in twosteps, much in the same way as we used to
study
themetallic surface. In the first step we
imagine
twoinfinite barriers at z = 0 and z = L, with L =
N(2 n/g).,
and N very
large.
The wavefunctions of the lower band are :and the
eigenvalues
are’ obtained fromNow there are no bulk states for n = 0, N,
i.e.,
the twoband-edges. By using
an argument similar to the metal-like case, it isstraightforward
to obtainthe surface
density
of states for the 1-dimensionalcrystal
with an infinite barrierKnowing
thedensity
of states for an infinite barrier at z = 0, we caneasily
evaluate it for the system describedby
the wavefunctions(11).
Here, theeigen-
values must
satisfy
the conditionand
finally
weconclude,
inanalogy
with eq.(8),
that the surface
density
of states for the valence band isgiven by :
In the semiconductor case it is
interesting
to noticehow il changes
across the band. For E --> 0, q -+0,
while for E =Ev,
a = 1and q
= nn(see
eq.(15)).
. This
gives
the whole surfacecharge
in the valence bandby using
eq.(20) :
where we have taken into account
spin degeneracy.
The value of n
depends
on - the surfacepotential.
As the barrier is lowered
and/or displaced
to theleft,
a S.S. goes down in energy into the lower band and the associated
charge changes by.
two electroniccharges.
Thus n takes on
successively
the values0,
1. Thisimplies
a constantoccupation of
the S.S. for reasonaoiechanges
in the surfacepotential, i.e.,
not solarge
thatn
changes by
one unit. For a surfacepotential
of thekind shown in
figure 1 b,
and with a S.S. near themidgap
it iseasily
seen that n = 0. Thisimplies 1/2 occupation
of the S.S.After
discussing
thedensity
of states, it is alsointeresting
to look at thecharge density
disturbance in real space near the surfaceregion.
We recallthat,
as
regards density
of states in energy, thechange
dueto the appearance of S.S. in the gaps is
exactly
can-celled
by accompanying
distortions in thedensity
ofstates in the continuum of the bulk
bands,
so that thechange
in the total number of states is zero[1, 4].
We now concentrate
specifically
on the consequences ofhaving
S.S. in semiconductor surfaces and inves-tigate
a similar but differentquestion, namely
thechanges
in the localdensity
of states associatedwith
(a)
the gaps, and(b)
the twoneighbouring
bulkbands.
The
problem
can be studiedstarting
from the wave-function
(11), assuming that I’
is known(see
eq.(13)
and
(15)).
From this wavefunction it is astraight-
forward matter to obtain the local
density
of statesN(E, z)
in the bands. The details and theintegrated density
aregiven
ofstates
inappendix N(E, z)
I.dEThe result is that the above two local densities turn out to cancel each other at every
depth
z. Thus, in our2-band
analysis,
if a gap is in a range ofoccupied energies,
then all distortions in the local densities(a)
and
(b)
cancel out, so that thespecific
features of the semiconductor case in the localdensity disappear entirely.
An
interesting
case appears when the S.S. is in themidgap.
Here, the distortions in the localdensity
ofstates in the valence and conduction bands are
equal,
due to the symmetry of the narrow gap.
Then,
thecharge
cancellation takesplace
also between thecharge defect
in the valence band and at occupied
S.S.With all these results we now
proceed
tostudy
self-consistently
the surface of a 1-dimensional semi- conductor model.Assuming
a narrow gap, weimagine
the
system
built up in successive steps as follows :we start with a
jellium
model and then switch on acrystal pseudopotential.
This has two effects on thecharge
near thesurface, namely, (i)
the consequence of a modification in the bulkcharge density
whichappears both for metals and
semiconductors,
and(ii)
the
specific
features of the semiconductor gap. While for a smallpseudopotential
the first effect issmall,
the second one can be much moreimportant
sincethe
charge rearrangement
between the S.S. and the valence band includes a very strong surfacedipole.
This is illustrated in
figure 2,
where this surfacedipole
has been drawn as a function of the S.S.
position
withparameters adapted
to the Si case. The S.S. is2 occupied
and thefigure
shows howstrongly
thisdipole depends
on the surfacephase-shift n’.
However, we are now
going
to argue that this effectdisappears
in a selfconsistent surface. First wepostulate
that the surfacepotential
can bereasonably
calculated in a first
approximation by neglecting
thespecific
features of the semiconductor case. The idea is toobtain q’
from the surfacepotential,
hence theS.S.
itself,
and then see whether the firstapproxima-
tion needs further corrections. We are interested in the upper energy range of the valence
band,
close toEv.
With a narrow gap this means
energies
close to theFermi level
EF. But,
for the reasonsjust stated,
westart
treating
the surfacepotential
as in the metal case.Here we have shown that
ilo(EF) = c/4,
the valueofyy’
then
being - n/4 (see
eq.(13)
and(1)).
With thisvalue the S.S. as
given by
eq.(14)
isplaced
at themidgap,
and we conclude that no netchange
in thelocal
density
of states appears as a consequence of the S.S. itself. Hence there is no need to correct thestarting point
of our firstapproximation.
FIG. 2. - Induced surface dipole as a function of the S.S. level for
a 1-dimensional model. The S.S.
is i
occupied andThis establishes the main result of this section : the selfconsistent surface
potential
of a 1-dimensional two-band semiconductor is rather like the surfacepotential
of a 1-dimensionaljellium
case, theonly
modificationscoming
from the induced bulkcharge,
this effect
being
the same in both metal and semi- conductor surfaces.Moreover,
withindependence
of the details on the surface
potential,
the S.S. isplaced
at themidgap with 2 occupation.
3. The
(111)
covalent face. - We now discuss how theprevious analysis
can be extended to a(I 11)
covalentface.
We assume a local one-electron
potential
in the formwhere G is a 2-dimensional
reciprocal
lattice vector. Forgiven
k vectorparallel
to the surface - henceforth K- the
Schrodinger equation
to solve is :and we want to
study
all energy ranges, those of the bulk band included.Very
detailed selfconsistent solutions have been obtained[2, 3]
in numerical form. Since thecomputational
task involved is ratherheavy,
theproblem
is sometimes
simplified by studying only high symmetry points
of the 2-dimensionalB.Z.,
such asT and X
infigure
3[2].
Surfacecharge
andpotential
are then constructed on the basis of the information obtained in thestudy
of these twopoints.
The results of(2)
show that one can obtain in this way a reasonabledescription
ofthe surface
potential. However,
if theproblem
isto be simplified
in this manner, it may be moreappropriate
to
study
otherhigh symmetry points.
Forexample, T would
seem to be the least convenient one, since itactually
has the
highest
energy of thedangling
bond band and is therefore empty. A more convenient choice wouldseem the J
point
which in some calculations turns out to have the lowest energy of thedangling
bond band andwhich,
moreover, has further surface states in the lower gaps[5, 6]
which can be relevant in theinterpretation
ofexperimental
data. We thus chooseJ
as morerepresentative
of the surfaceproperties
and will later substantiate this in the context of a morecomplete analysis.
Figure
3 shows the(111) projection
of the Jones zone(J.Z.)
of the diamond lattice and the B.Z. for this surface. As shownby
Elices et al.[7],
in order tostudy
S.S. it is sufficient to include allplane
wavescorresponding
to those
points
on or inside the J.Z.boundary,
such that their surfaceprojections
map onto thepoint
of the B.Z.under
study.
For the Jpoint
this meansjust
the wavevector whichproject
onto the threeequivalent points
infigure
3. Thepotential
is thenconsequently approximated by :
and we look for solutions of the form
FIG. 3. - 2-dimensional B.Z. (inner hexagon) and projected J.Z.
(outer hexagon) for the (111) face of the diamond lattice with
special points shown. Other authors use the following notations
for special points : r XJ - rMK (3) -> rJK (2).
where (o =
exp - 1 .2 3 n ),
andonly
the three wavevectors icj, K 2 and x 3 are involved. The functionsf(ll(z) satisfy
theequations
where
Thus the
matching problem
for the Jpoint
factorizes into threeindependent
1-dimensionalproblems.
For
large positive
z - inside thecrystal -
thepotentials V(i)(z)
becomewhere
V3, V8, V11
are thepseudopotentials
associated with thecomponents (111), (220), (311),
and h is the modulus of(111).
In the J.Z.approximation
so that
FIG. 4. - Bulk energy bands in Si for k varying perpendicularly to a (111) face - J point. Dotted lines : real energy loops for complex k.
The
pseudopotential components originate
gaps at the wavevector 4h/3, h, h/3
in the extended zone scheme.The one at 4
h/3
is the fundamental gap. As notedby
Heine and Jones[8]
its width to second order is determined notonly by V8
but also to alarge
extentby V3,
A similar effect appears with the gap at 2h/3,
which to secondorder is created
by V 3 ex p i 3 hZ . 3
The band structureresulting
from(27)
and(29)
is shown infigure 4,
reducedto the first
B.Z.,
whose border is ath/3.
Thecomplex
band structure is also shown.We remark that the
independent
1-dimensional eq.(26)
here obtainedcorrespond
to the threeequations
of
[7]
derived in thestudy
of the fundamental gap at the Jpoint.
It was shown there that a two-band modelgives actually
rather accurate resultscompared
with those based on a morecomplete description
of the bulk band structure. The sameapplies
when each gap - besides the fundamental one - is studiedseparately, i.e.,
an
adapted
two-band model which focuses on each gapprovides
a rather accurate basis tostudy
different gap ranges. We shall use here thisapproximation,
so that all 1-dimensionalproblems
will involve a two-band model in the mannerexplained,
and thecomplete
3-dimensionalstudy
of the Jpoint
will be built up from the threeindependent
1-dimensionalproblems.
Now, we concentrate on the
study
of the fundamental gap, since the effects on the surfacepotential
of theother three gaps for each 1-dimensional
problem disappear completely.
This was shown in theprevious
sectionfor a 1-dimensional gap in a range of
occupied energies.
In each one of the three 1-dimensionalproblems,
weconcentrate in the fundamental gap within a two-band
approximation
with an effectivepseudopotential Y8 [7].
The three surface
potentials
become :which
correspond
to threeindependent
1-dimensionalproblems,
with thepseudopotentials displaced by 0,
±
a/3.
Now, we can
analyse selfconsistently
the surface of the 3-dimensionalcrystal
as in theprevious section,
theonly
differencebeing
the simultaneous consi- deration of three 1-dimensional cases. Tobegin with,
we need to know the
phase-shift
’10’ which was deter- mined in the 1-dimensional caseby
a sort of Friedelsum rule. In our
present
case, we can use a genera- lization of this sum rule as shownby Langreth [9]
and
Appelbaum
and Blount[10].
Their results statethat,
in ametal,
the averagevalue ( n 0 >
for all wave-functions at the Fermi level is determined
by
thecondition of surface
charge neutrality,
its valuebeing n/4.
This is a rather evidentgeneralization
of the1-dimensional result.
We use this value
for n0 >
instudying
the contri-bution of
(30a),
whereas for the other1-dimensional
n 2 n
problems
we must take no= 4
±3
due to thedisplacement
of thepseudopotentials by
±a/3.
Weshall discuss later the
validity
of thisapproximation.
Proceeding
in this way we find that(30a) yields
a S.S.eigenvalue exactly
at themidgap,
whereas no S.S.arises from
(30b, c).
But in the latter two cases we find results which aresymmetric
with respect to themidgap, i.e.,
the local surfacedensity
of states in thevalence
band/conduction
band of one case isequal
to that in the
C.B./V.B.
of the other case(Fig. 5).
This result is of paramount
importance
inestablishing
the
occupation
of theS.S.,
because now we havedetailed local cancellation of the actual surface
charge density
in the V.B.arising
from(30b)
and(30c) (Fig. 5).
Therefore we are left withjust
the 1-dimen- sionalproblem
of(30a)
and with no extracharge
orpotential
due to(30b, c).
After theprevious
sectionthis means
occupation 2
for theS.S.,
and no netchange
in the local
density
of states as a consequence of the S.S. itself. Thisfinally
extends to the(111)
covalentfaces the results of the 1-dimensional semiconductor model as stated
in
theprevious
section.FIG. 5. - The density of surface states associated with each one of the three different 1-dimensional gaps at J. In a) the S.S. is at the
midgap, while in b) and c) the density of states in the valence (V)
and conduction (C) bands are related by :
bNb
=6Nf
andbN6
=6N?.
Notice that in the three cases the total density of S.S.around the gap is zero.
We finish this section
by discussing
thevalidity
ofthe
approximations leading
to our results. These are :i)
the useof n0 > for J,
andii)
the use ofJ
as arepresentative point
of the whole B.Z.To understand the
meaning
of the firstapproxi-
mation let us consider the 3-dimensional surface of a
jellium
model. In this surface theonly
non-zeropotential
isVo(z),
and thephase-shift
110 for the different 1-dimensionalproblems changes
conti-nuously
from a maximum at K = 0(T point),
to a mini-mum at K =
kF.
The average valuen/4
can beexpected
to occur at some intermediate
point
near J wherek 1. = 3/4
h. This can be illustrated with asimple model,
an infinite
potential
barrierplaced
at a distance b fromthe jellium edge,
so as to neutralize the surfacecharge.
In this case, it is an easy matter to find the
phase-
shift
at J
--This
expression gives
for the Sidensity
very close to
n/4.
Let us now discuss the
validity
ofusing J
as arepresentative point
of the 2-dimensional B.Z. Thiscan, be clarified
by looking
at thecomplex
bandstructure in the
(111) direction,
as we move off the Jpoint (7).
In the same Jpoint
we find three identicalloops,
one of themgiving
a S.S. As we leave thispoint,
the threeloops split and, depending
on theselected
direction,
1(or 2)
of theseloops
move up inenergies,
while the other 2(or 1)
move down. At thesame
time,
the S.S.changes
itsposition
in relation toits supporting loop,
in such a waythat,
forinstance,
at
T
theS.S.
is lower than itscorresponding midgap,
while at X the
opposite
situation occurs[2, 3, 7].
Under these
conditions,
thevalidity
ofusing J
as arepresentative point
meansthat,
on the average, theweight
of theloops
with the S.S.placed upwards
itsmidgap
is counterbalancedby
theweight
of thosewith the S.S.
placed
downwards. An idea of thiscompensation
can be obtainedby estimating
thecentre of the three
midgaps averaged
over the differentpoints
of the whole B.Z. When this isdone,
we find that this centre is at 0.5 eV above’the midgap
atJ.
We think that this result is a
good
indication of thecompensation existing
between different2-dimensional points. Moreover,
we remark thatusing J
as a repre- sentativepoint
we have founda 2 occupation
of theS.S., demonstrating
ageneral
theoremproved by
Kleinman
[11].
It seems to us that this resultexemplifies
the afore mentioned
compensation
between different 2-dimensionalpoints.
4. The
(111)-Si
surface. - In this section we check theprevious results, applying
them to obtain the surfacepotential
and surface states at( 111 )-Si
sur-face. The idea is to
study
this surfaceusing
a methodemployed
elsewhere[12]
to obtain the interfacepotential
at metalsurfaces, including
the discreteness of the lattice.Following
thepoint
of viewgiven
in thisapproach,
we start from an
equation
which relates the actual surfacepotential V(G, z)
to the bare ionpotential Yo(G, z) through
the dielectric function~(G, G’ ;
z,z’)
of the surface system :
In
principle
the difference between metal and semi- conductor surfaces lies in different functionsE-1.
After our
previous
arguments we shall use forE-1
the sameapproximations
as in the metal case. We summarize here theprocedure
followed[12] :
i)
Weneglect
the effect ofcrystallinity
on theexchange
and correlation electronicpotential,
sothat we take the
Lang-Kohn [13]
solution in the surfaceregion.
An idea of the error introduced in this way can be obtainedby comparing
the value of thispotential
for the averagedensity
of Si with the average valueresulting
from adensity varying
accord-ing
to the bulkperiodicity.
The errors introduced arebetween 0.1 eV and 0.2 eV at most.
ii)
We use forE-1
an extension of the semiclassical method as discussed in[12].
The limitation of thisapproximation
- based onjellium
models - forsemiconductors lies in the
spatial
variation of thecharge density,
in the bulk as well as near the surface.We have corrected for this
by using
the response function evaluated not for the average bulkdensity
po, but for the G = 0 component of thedensity
at z = 0.This amounts to
replacing
poby
0.67 po,using
dataof Bertoni et al.
[14]
which we have done ingeneral
for all information needed about the bulk.
Using
thiseffective
charge,
we must calculate the response to asurface
charge
and a surfacedipole.
In the first case, we use aninterpolation
of the results asgiven by Lang-Kohn [13]. Furthermore,
we also need theresponse to a surface
dipole,
which in[ 12]
was obtainedin a linear response. Here instead of the linear response
we have calculated the
screening by using
the sameapproach [12]
withoutlinearizing
thepotential.
We have thus calculated the surface
potential
forthe ideal 111-Si surface.
Figure
6 shows the G = 0component
of thispotential,
andfigure
7 shows thispotential
as a function of zalong
two linesgoing through
the atoms in the first and second atomiclayers respectively.
These results can becompared
with those of Schluter et al.
[3], although
in the secondcase these authors
give
their results in the form ofequipotential
contours. Onextracting
the information from their results andcomparing
with ours, we findlarge discrepancies
in the bulkpotential
near the ions.This is
simply
due to the use of very different bareFIG. 6. - Surface potential for Si(111), averaged over the surface G = 0 as a function of z.
FIG. 7. - Surface potential for Si(111), as a function of z, along
two lines going through the atoms in the first and second atomic
layers.
pseudopotentials (1). However,
thepotential shape
is very similar in the
regions
away from the ionsand,
inparticular,
theprofile
of the surfacepotential
barrier turns out to be very
similar,
thusproviding
arather
satisfactory
check on ourapproximations.
It is also
interesting
to look at the results for the ionization energy(the
difference between the vacuumlevel and
Ev). Taking
for theposition
ofEv -
cor-responding
to T - relative to the zero of bulkenergies
the value 10.8 eV
(E. Louis, private communication),
we obtain an ionization energy of 4.5 eV which compares with the
experimental
value of 5.15 eV[15].
Moreover,
itmight
be in order to discussbriefly
theS.S. that our model
yields.
We concentrate on theJ
point.
The calculation was donefollowing
the samegeneral
lines as in[7],
andusing
the same detailedmany band bulk structure, but with the
following important
differences :i)
We did not use a steppotential,
but thepotential profile
obtained in theprevious calculation,
andii)
we studied the fundamental gap and the lower gaps
as well. Our results have been
published
elsewhere[1]
and shall not be
given
here.However,
we remark that the S.S. in the fundamental gap isplaced
very close to themidgap
in agreement with oursimplified analysis
and other calculations[2, 3].
At the sametime,
two other S.S. appear at lowerenergies,
over-lapping
with the valence band. The lowest one agrees with the state foundby
Louis[5]
andby
Falicov and Yndurain[6].
This statebelongs
to a band of S.S.which,
as foundby
theseauthors,
exists over domains of the 2-dimensional B.Z. close to theJ point.
In
figure
8 we have drawn thecharge density
asso-ciated to the S.S. of the fundamental gap at
J.
Thiscan be
compared favourably
with the samedensity given by
Schliiter et al.[3], although
here these authors have a relaxedgeometry.
The relaxation must enhance the localization ofcharge
near thesurface,
this willexplain
the small differences between the twocharges.
FIG. 8. - Charge density associated to the S.S. at J along a line going through the atom in the first atomic layer. The dots represent
plane layers.
In
conclusion,
we find that theprevious
resultsshow a
general
agreement with more elaboratecomputations, giving
a strong support to the conclu- sions of section 3.At this
point,
it istempting
to use thegeneral
method outlined in
[12]
to obtain the ionic relaxation at the Si surface. This has been done and we have found a very smalltendency
towards relaxation.Of course this is
only
aplausible
indication and must not be taken tooliterally
after the manyapproxi-
mations involved.
However,
it isinteresting
to notethat in a recent paper, Shih et al.
[16]
have estimated from LEEDexperiments
the inward relaxation of the last atomiclayer
asonly
0.12A.
(1) We have used the data of Bertoni et al. (based on a Heine-
Abarenkov pseudopotential) kindly communicated to us by the
authors.
Finally,
insupport
of theviewpoint expounded
in
paragraph 3,
we refer to a recentpublication [17],
where it has been shown how the
dangling-bond
S.S.at the
(111)
Si and Ge surfaces can be obtainedquite
well with a
suitably displaced abrupt
barrier. The idea is todisplace
theabrupt
barrier in order to achieve metalliccharge neutrality.
This fixes the averagephase
of the wavefunctions at the metallic Fermi level in its correct valuen/4,
thusgiving
the bestapproxi-
mation to the S.S. The results of this work
give
a newsupport
to thepoint
of viewexpressed
here.5. Conclusions. - In the first
part
of this paper a 1-dimensional covalent surface has been studied.Our
analysis
shows that :i)
thepotential
andcharge
distribution in the surface
region
behave rather like in the case of ametal,
andii)
thedangling-bond
S.S.is determined
by
the condition ofcharge neutrality.
This
gives a i occupation
of theS.S.,
and a S.S.placed
at themidgap.
A covalent
(111)
surface has been discussed later.Our arguments
strongly
suggest that the same conclu- sions canapply
to the surface of this 3-dimensional semiconductor. This allows us tostudy
the surfacepotential
in asimple
way as in the case of ametal,
the resultscomparing quite
well with the ones of elaboratecomputations.
Finally,
we stress that the type ofanalysis
shown inthis work can be used to
study
different surface condi- tions. Forexample,
in a recent work[18]
these ideas have beenapplied
toanalyse selfconsistently
the Si-metal
junctions
withquite good
results.Acknowledgments.
- We wish to thank Professor F. Garcia-Moliner for manyhelpful
discussions and the criticalreading
of themanuscript.
Appendix
I. - In thisappendix
we calculate the localdensity
of statesN(E, z)
and theintegrated density
of
states fN(E, z)
dz, near a 1-dimensional semiconductor surface.The wavefunction in the valence band is
given by :
while in the conduction band we have :
The
phase-shift ’1
is determinedby matching
the wavefunction to the onegiven by
eq.(13) :
where it has been assumed A’
g/2,
and we have definedOnce
#(z)
has been obtained andproperly normalized,
one has the localdensity
of states :Here
and
ps
giving
the surface contribution. In theseexpressions
the +
/
-sign standing
for thevalence/conduction
band. It isinteresting
to note that far from the gap, a -> oo or zero, in such a waythat p’
-+ - cos2 2 9 + , (z - zo) , like in the
metal case. In order to obtain thetypical
contribution for the semiconductorsurface,
this last metallic limit has to be subtracted. This leaves uswith the
expression :
The
integrated density
of states in the valence band isgiven by :
The contribution
coming from p’
can besimplified using
the symmetryin A,
sincea( - A)
=1 /a(A).
In this waywe can write :
with
Now, to obtain
(1.10)
wechange
theindependent
variable from A to a :and
Thus
The conduction band can be
analysed following
the same steps, in such a way that we can write :and
the