PHYSICAL REVIEW B VOLUME 41,NUMBER 6 15FEBRUARY 1990-II
Microscopic structure
of
the hydrogen-phosphorus
complex
incrystalline
silicon
P.
J.
H. DenteneerFaculty
of
ScienceE, lectronic Structureof
Materials, Catholic University Nij megen,Toernooiveld 1,6525EDNijmegen, The Netherlands
C. G.
Van de WallePhilips Laboratories, North American Philips Corporation, 345Scarborough Road, Briarcli+Manor, New York 10510
S.
T.
PantelidesIBMResearch Division, Thomas
J.
watson Research Center, Yorkto~n Heights, Ne~ York 70598 (Received 8 August 1989)The existing discrepancy between theoretical models and experimental results for
hydrogen-donor complexes in crystalline silicon is resolved using first-principles
pseudopotential-density-functional calculations for the hydrogen-phosphorus pair. In the configuration which is the global energy minimum, H islocated on the extension ofa
P-Si
bond on the Siside, with the Si-H pairrelaxing away from Pby 0.6 A, leaving the P atom threefold coordinated. The calculated
stretch-ing and ~agging vibrational frequencies associated with this configuration are in accord with ex-periment.
The structure and properties
of
hydrogen-impurity complexes in semiconductors have been studied intensive-ly in the last few years using both experimental and theoretical methods. ' ' For the hydrogen-boron com-plex in silicon, which is the prototypal hydrogen-acceptor complex that has been studied most elaborately, a con-sistent picture has emerged (see,e.
g., Ref. 5 and refer-ences therein). In the equilibrium configuration of the complex, the H atom resides inside aSi
—8
bond, forming a three-center bond. Also for the qualitatively different (H,Be)
pair inSi
and(H, Si)
pair in Ge, theory has pro-vided satisfactory explanationsof
the experimental results aswell as new insights (seeRef.
6and references therein). The H atom in the (H,Be)
pair is able to tunnel around the Beatom because its lowest-energy location isclose to theC
site (midway between two Siatoms bonded toBe),
where the relaxationof
the surroundingSi
atoms issmall. The H atom in the(H, Si)
complex in Ge islocated close to a tetrahedral interstitial(T4)
site.In contrast, the structure
of
hydrogen-donor complexes, e.g.,(H,
P)
inSi,
has so far not been determined con-clusively. Experiments have shown that all H-donor pairs inSi
have similar infrared absorption spectra, sug-gesting that H isnot bonded to the donor. The observa-tionof
a nondegenerate stretching mode around 1560 cm ' and adoubly degenerate wagging mode around810
cm ' suggests that the center has trigonal symmetry. Theoretical models have so far not reproduced these fre-quencies. In
Ref.
7,a model was proposed in which H is located on the extensionof
aP-Si
bond on the side ofSi.
Using empirical tight-binding calculations this
"AB
(anti-bonding) ofSi"
configuration was found to belower in en-ergy than the"AB
of
P"
configuration. The frequency for the H stretching mode was calculated to be 2145cm which is very different from the experimentally deter-mined valueof
1555cm'.
In a subsequent calculation by the same group, but using the more reliable first-principles pseudopotential-density-functional method, theconfiguration was qualitatively confirmed. ' However, in the latter calculation the stretching mode was found to be at400cm
Recently, a number of groups using various kinds
of
cluster calculations
"'
have proposed a configuration similar tothe one inRef.
10with the distinction, however, that theSi
atom closest to H relaxes from its lattice site towards H to become almost coplanar with its three nearest-neighbor Si atoms. Estreicher et al."
discuss the inherent difficulties in calculating vibrational frequencies to within a reasonable accuracy using quantum-me-chanical cluster calculations and do not attempt to calcu-late any frequency. DeLeo and Fowler4 and Amore Bona-pasta et aI.' calculate a H stretching frequencyof
2150
cm,
again in disagreement with experiment.Summarizing, it can be said that theoretical studies so far have not been able toput forward a microscopic model for the (H,
P)
complex that can be conclusively identified as the one that isexperimentally observed.In this paper, we present results of accurate first-principles calculations for the (H,
P)
pair. We determine the lowest-energy configuration and show that this configuration is responsible for stretching and wagging vi-brational frequencies that are in agreement with experi-ment. We have successfully used the pseudopotential-density-functional method before in studiesof
H in pureSi
andof
various complexes in Siand Ge. ' 'If
thecal-culations are properly converged with respect to all the numerical approximations involved, the method is very re-liable in determining defect configurations. In particular, total-energy differences between different defect config-urations can be calculated to within an accuracy
of
0.
05-0.
1eV and typical H vibrational frequencies can be calculated with an accuracyof
about 100cm '(Ref.
5).
For detailsof
calculations in which such accuracy is achieved we refer to Refs. 5 and9.
In the present study, we closely examine various configurations with trigonal symmetry (see below), aswell as the regions close totheC
DENTENEER, VAN de %ALLE,AND PANTELIDES
and C' sites
(i.
e, H midway between twoSi
atoms bonded toP
and H midway betweenP
and a next-nearest-neighborSi,
respectively). TheC
and C'sites are atleast 1.5 eV higher in energy than the lowest-energy config-uration; we will not consider them further. The configura-tions with trigonal symmetry, which can be classified ac-cording to the order in which the H,Si,
and Patoms are found along a(111)
axis(H-Si-P,
Si-H-P, and Si-P-H, re-spectively), are optimized by relaxing up tonine atoms ac-cording to the Hellmann-Feynman forces on these atoms. These forces can be calculated with the same levelof
ac-curacy as total energies from the self-consistent solutionsof
the Schrodinger equation for the valence electrons. '3 In order to optimize the configurations we move the atomsin the direction
of
the calculated forces until the forces be-come negligible, thereby minimizing the total energy.We find that each
of
the three trigonal-symmetry configurations, including appropriate relaxations ofall the atoms, constitutes a local minimumof
the total-energy surface. Furthermore, the three minima are very close in energy: they all lie in an energy rangeof
only0.
5 eV (see Fig.1).
These small energy differences open the way for the occurrence ofmetastable states ofthe complex.Now we describe the two local minima and one global minimum configurations mentioned above.
Of
these three, the configuration highest in energy is the one inwhich H resides between a
Si
andP
atom forming a bond. We call this configuration"BC
(LLR
ofP)"
since it in-volves avery large lattice relaxation(LLR)
ofthe Patom(BC
stands for bond-center site). The Patom relaxes out-ward (away fromH)
by1.
22 A, whereas theSi
atom re-laxes outward by only0.
10 A. The H-Si distance in this configuration is1.
50 A, similar to the H-Si distances found in molecules,e.
g.,SiH4, and at a hydrogenated va-cancy. The H atom breaks theSi-P
bond and saturates the Si dangling bond; this allows for the large relaxation of Pthrough the plane ofits three neighboring Siatoms to a position where it is threefold coordinated. The charge density for this configuration is shown in Fig.2(a)
and displays alone pair onthe Patom pointing in the direction ofthe nearest Td site on the lineSi-H-P.
TheH-Si
bond that is formed has a calculated stretch frequencyof 1900
cm
',
much larger than the observed frequency. In the other local minimum configuration, which we call ABof
P,the H atom islocated very close to the Tqsite closest to the Patom. The energy
of
ABof P
is only0.
10eV lowerthan that
of
BC
(LLR of P)
(see Fig.1).
In this configuration, noneof
the atoms relax appreciably from their ideal lattice position, resulting in a H-P distanceof
almost an undistorted
Si-Si
bond length(2.
35A).
The calculated H stretch frequency for this configuration is 570 cm',
much smaller than the observed frequency. Finally, the global energy minimum configuration is the one called ABof
Si(LLR of
Si).
It
has an energy0.
35eVO
.
IOeVj
O.I6eV O.leeV BC(llr ofP) —ABof P — AB of Si ABof Si(llr ofSi)FIG.
l.
Relative energies ofdifferent configurations withtri-gonal symmetry for (H, Si,P) complexes in silicon. AB stands
for antibonding site, BCfor bond-center site, and LLRfor large lattice relaxation. A more detailed description of the four
configurations is given in the text (see alsoFig.
3).
FIG. 2. Total valence charge density in the
(110)
plane for (a) the BC (LLR of P) and (b) the AB of Si (LLR of Si)configurations for a (H, Si,P)complex in Si. The black dots
in-dicate atomic positions and the straight lines connect bonded
atoms. The broken Si
—
Pbond and theSi-H
bond are indicat-ed by dashed lines. The contour spacing is 1.87e/0,
where0
is the unit cell volume of bulk Si (which contains 8 electrons inbulk Si). The lowest-density contour shown (in the channels
be-tween the two atomic chains) is 2.32
e/0
and the highest-density contour shown (around the H atomic position) is 34.1e/0
in (a) and 28.5e/0
in (b). The maximum density in aMICROSCOPIC STRUCTURE OF THE HYDROGEN-PHOSPHORUS.
. .
3887lower than AB
of
P,and H is located close to the Tq siteof
aSi
atom bonded toP.
ThisSi
atom relaxes outward by0.
59 A (leaving the P atom threefold coordinated; see Fig.3).
The P atom relaxes by the small amount of0.
14A.(in the direction
of
theSi
relaxation, contrary to the re-sultsof
cluster calculations). The H-Sidistance is1.
66A, which is somewhat larger than a typical value for a H—
Si bond distance (see above), indicating a slight weakeningof
the bond. The H stretch frequency is therefore expect-ed to be lower than for a typicalSi-H
bond. Indeed, we calculate a frequencyof
146Qcm',
which in viewof
the error bar on calculated frequencies discussed above, is in agreement with the experimental numberof
1555 cm Also the calculated frequencyof
the H wagging modeof
740cm ' is in agreement with the experimental result
of
809cm
'.
The agreementof
both calculated frequencies with experiment, taken together with the factthat the ABof
Si(LLR
ofSi)
configuration has the lowest energy of all configurations studied justifies the identificationof
the experimentally observed complex with this ABof
Si (LLR
of Si)
configuration.In Fig.
2(b),
we show the valence charge densityof
the(H,P)
pair in the ABof
Si
(LLR of Si)
configuration. TheP-Si
bond is effectively broken and a lone-pair-like density, which isa remnantof
the previousP-Si
bond, is extending in the directionof
the former bond. All the valence electronsof
P are accounted for in this way. TheSi
atom has gone from an sp bonding configuration to an sp bonding configuration with respect to its threeSi
neighbors. The surplus electron
of Si
(which does not have to go in aP-Si
bond) pairs with the H electron to form aSi-H
bond. Indeed, the charge density betweenSi
and H isvery similar to the one found in the caseof
H saturating aSi
dangling bond. Bonding is indicated by the fact that the charge density around the H atom is clearly modified from the spherical form it has when Si and H are far apart (see, e.g.,the charge density for the AB ofSi
configuration inRef.
10).
For the sake
of
completeness and to make the connec-tion with the resultsof
other work, we mention that ifwe do not allow for relaxationof
theSi
neighborsof
theSi
atom between H and
P,
thisSi
atom relaxes outward by only0.
19A. This results in a ABof Si
(without large lat-tice relaxation ofSi)
configuration which is still lower in energy by0.
16 eV than the ABof P
configuration (see Fig.1),
but higher in energy by0.
19eV than the ABof Si(LLR of Si)
configuration. For this ABof Si
configu-ration, which is similar to the one found inRef.
10,the H-Si distance is 2.1 A, much larger than atypical H—
Sibond distance, and the corresponding H stretch frequency iscalculated to be
600
cm'.
The H wagging mode for this configuration has acalculated frequencyof 600
cm as mell, indicating the absenceof
H bonding. The configuration that we find to be lowest in energy is almost the same as the one found in Refs. 4,11,
and 12. In those calculations, theSi
atom relaxes by an amount betweenH
0--FIG. 3. Schematic representation of the ABof Si (LLR of Si)configuration, which isthe lowest-energy configuration for a (H,Si,P) complex in Si (see also Fig.
1).
One Si atom hasre-laxed from its lattice position (indicated by a vertical bar) by
0.59Atowards H and is only 0.19Aaway from being coplanar ~ith itsthree Sineighbors.
Q.6and
0.
8A and a Si-H distance between1.
4 and1.
5 A.is found, which is smaller than our value
of
1.66
A. Con-sequently, those calculations render amuch larger stretch-ing mode frequencyof
about2150
cm'.
More recently, Chadi et al.' repeated the calculationsof
Ref. 10 and found similar results tothose presented here byus.Both configurations with large lattice relaxations dis-cussed above are reminiscent
of
recently proposed models for theEL2
andDX
defect centers in GaAs.'5's
In the caseof
EL2,
it is proposed that an As antisite can be in-duced by optical excitation to move by about1.
3 A from its lattice position to a metastable configuration. ' In the caseof
theDX
center, aSi
donor in GaAs may move1.
2 A from its lattice site.' In both cases, the configuration with alarge lattice relaxation isinherently associated with a simple point defect and can be provoked to materialize. In the subject ofour present study, it is the H atom with its one unpaired electron that is able to promote different bonding environments for the simple substitutional P donor involving large lattice relaxations ofeither aP
orSi atom. In this way, the Patom can yield to its natural ten-dency to be threefold coordinated. The configuration with a large lattice relaxation ofSiis found tobe lowest in en-ergy. We suggest that such complexes with large lattice relaxations be further investigated experimentally by means of ion-channeling techniques to confirm our findings.In conclusion, we have shown on the basis
of
first-principles calculationsof
total energy that the configu-ration with H at an antibonding positionof
a$i
neighborof
P, in which this$i
atom relaxes by0.
6 A, can be identified with the complex that is experimentally ob-served. In doing so, the discrepancy between resultsof
earlier theoretical studies and experiments isresolved. This work was supported in part by the
U.
S.
Officeof
Naval Research under Contract No. NQ0014-84-C-0396. One
of
the authors(P.
J.
H.D.
)
thanks theIBM
Research Division for hospitality during part ofthe execution time3888 DENTENEER, VAN de %ALLE,AND PANTELIDES
'S.
J.
Pearton,J.
W.Corbett, andT.
S.
Shi, Appl. Phys. A 43, 153(1987).2E.E.Hailer, in Proceedings
of
the Third International Confer ence on Shallow Impurities in Semiconductors, Linkoping,1988, edited by B.Monemar, IOP Conf. Ser. (Institute of Physics and The Physical Society, London, 1989),p.425. M. Stavola,
S.
J.
Pearton,J.
Lopata, and W. C.Dautremont-Smith, Phys. Rev. B 37,8313(1988).
46.
DeLeo and W. B.Fowler, in Hydrogen in Semiconductors, edited byJ.
I. Pankove and N. M. Johnson (unpublished); Bull.Am. Phys. Soc. 34, 834(1989).
P.
J.
H. Denteneer, C.G.Van de Walle, andS.
T.Pantelides, Phys. Rev. B39, 10809(1989).
6P.
J.
H. Denteneer, C.G.Van de Walle, andS. T.
Pantelides, Phys. Rev. Lett. 62, 1884(1989).
~N. M.Johnson, C.Herring, and D.
J.
Chadi, Phys. Rev. Lett.56, 769(1986).
K.Bergman, M. Stavola,
S.
J.
Pearton, andJ.
Lopata, Phys. Rev. B 37, 2770(1988).
9C. G.Van de Walle, P.
J.
H.Denteneer, Y.Bar-Yam, andS. T.
Pantelides, Phys. Rev. B39,10791(1989).
K.
J.
Chang and D.J.
Chadi, Phys. Rev. Lett. 60, 1422 (1988)."S.
K.Estreicher, L.Throckmorton, and D.S.
Marynick, Phys. Rev.B39,13241(1989).
'2A. Amore Bonapasta, A. Lapiccirella, N. Tomassini, and M. Capizzi, Phys. Rev.B39, 12630
(1989).
'3M.
T.
Yin and M.L.Cohen, Phys. Rev. B26,3259(1982). '4D.J.
Chadi et al.(private communication).'5J. Dabrowski and M. ScheSer, Phys. Rev. Lett. 60, 2183 (1988);D.