Content uploaded by Christine Revenant
Author content
All content in this area was uploaded by Christine Revenant on Mar 21, 2016
Content may be subject to copyright.
PHYSICAL REVIEW BVOLUME 52, NUMBER 315 JULY 1995-I
Anomalous low-temperature dopant diffusivity and defect structure
in Sb- and Sb/8-implanted annealed silicon samples
A. Armigliato
Consiglio Nazionale delle Ricerche, Instituto di Chimica eTecnologia dei Materiali eComponenti per 1Elettronica (LAMEL),
via P. Gobetti 101, 40129Bologna, Italy
F. Romanato, A. Origo, and A. Camera
Dipartimento di Fisica dell Universita, I1VFM, via Marzolo 8, 35131Padova, Italy
C. Brizard, J.R. Regnard, and J. L. Allain
Commissariat al'Energie Atomique, Departement de Recherche Fondamentale sur la Mati'ere Condensee,
SP2M/PI-17, rue des Martyrs, 38054 Grenoble Cedex, France
(Received 10 January 1995)
The diffusivity of antimony into silicon and its dependence on the Fermi-level position and on the
structure of lattice defects has been investigated by high-dose ion implantation (2 and 5X10' atoms
Sb cm ). The donor concentration has been strongly increased by asubsequent pulsed laser annealing
treatment. For comparison, laser annealing of samples coimplanted with the same doses of both Sb and
Bhas been performed in order to obtain astrong electrical compensation. The heat treatment for
diffusion experiments of both Sb- and Sb+B-implanted wafers have been performed at arather low tem-
perature (600 Cfor 1h). Contrary to the prediction of the extrapolated Fair's equation, asignificant
shift in the dopant concentration profiles, as well as the formation of Sb precipitates, Sb-vacancy, and/or
Sb-B pairing have been observed. To explain this diffusivity, adiffusion coefficient Dsb, independent of
the dopant concentration and seven orders of magnitude higher than that previously determined by Fair,
must be assumed. This means that the increased Sb diffusivity is not related primarily to the Fermi-level
position. The huge increase in D» is related to defects (e.g.,twins, dislocations, rodlike defects, precipi-
tates, and dopant complexes) which have been characterized by extended x-ray-absorption fine-structure,
Rutherford backscattering spectrometry and channeling, and TEM techniques analyses. Moreover, an
anomalous high tensile strain of the samples indicates alarge incorporation of vacancies. These defects
are also responsible for the dopant backwards diffusion and outdiffusion, which is another surprising
phenomenon which occurred during thermal annealing of most of the samples.
I. INTRODUCTION
In silicon integrated circuit technology, regions heavily
doped with both donor and acceptor atoms are frequently
encountered. In addition to the expected electrical com-
pensation, astructural compensation has been also re-
ported for III-V mixed doped silicon, which is due to the
formation of dopant complexes and pairs. 'An addi-
tional phenomenon, which takes place during high-
temperature annealing of antimony-boron coimplanted
silicon samples, is retarded Sb diffusion in the p+ doped
region with anear-uniform profile.
In one of our recent works on codiffusion in silicon at
900 and 1000'C of 8and Sb, both implanted at high dose
(2X10' cm ), secondary-ion-mass spectrometry (SIMS)
concentration profile measurements and transmission
electron microscopy (TEM) observations supported the
hypothesis of mobile donor-acceptor pairs and the in-
crease of antimony solubility through areduction in the
precipitated Sb fraction. Moreover, the diffusivity of
both dopants in codiffused samples was lower than that
of the individual dopants, the Sb+8 pairs being mobile,
though with alow diffusivity.
The purpose of this work is to investigate how Sb
difFusivity is affected by astrong variation of the Fermi-
level position during diffusion. Andersen et a/. found
extremely large diffusion coefficients of Sb during high-
temperature rapid thermal annealings in Si wafers con-
taining high concentrations of Pdonors. To achieve this
situation, in our experiments apulsed laser annealing
treatment has been performed in samples heavily im-
planted with Sb only, and in samples coimplanted with Sb
and B. In the first case, this leads to an electrically active
concentration of 2X 10 'cm, with the Fermi level close
to the conduction band, whereas in the latter samples the
strong electrical compensation brings the Fermi level
closer to the band-gap center. Moreover, heat treatments
for diffusion experiments of both Sb- and Sb+8-
implanted samples are performed at 600'C, a tempera-
ture markedly lower than the ones used by Margesin
et aI.,which should induce no shift in the correspond-
0163-1829/95/52(3)/1859(15)/$06. 00 52 1859 1995 The American Physical Society
A. ARMIGLIATO et al. 52
ing concentration profiles. Apreliminary extended x-
ray-absorption fine-structure (EXAFS) analysis of these
specimens showed, rather surprisingly, that both precipi-
tation and pair formation take place, thus indicating that
the diffusivity at 600'C should be larger than expected.
In order to discriminate between effects induced by the
carrier concentration 'and those induced by the struc-
ture of the samples, an extensive structural characteriza-
tion has been performed. The use of SIMS, TEM, EX-
AFS, Rutherford backscattering spectrometry (RBS), nu-
clear reaction analysis (NRA), and ion-channeling tech-
niques allowed us to characterize dopant concentration
profiles, lattice defects, precipitates, and complexes in-
volving Sb and 8as well as the dopant lattice site loca-
tion and the strain of the dopant layers.
II. KXPKRIMKNT
A. Specimen preparation
The effect of the coimplantation of Sb+8 on Sb
diffusivity has been investigated in( 100) Si monocrystal-
line wafers, and has been systematically compared with
corresponding Sb-only implanted case. Two implanta-
tion doses have been studied: 2X10' and 5X10'
ionscm .In each coimplanted sample, the Sb and 8
doses are nominally the same in order to maximize the
coimplantation effect. The Sb+ and "8+ions have been
implanted at 160 and 20 keV, respectively, in order to ob-
tain the same impurity depth range. In fact, the project-
ed range R~ and the straggling hR are (79.5+23.3) nm
for Sb and (74.9+28.3) nm for B. In order to obtain a
nearly constant concentration profile of the dopants, and
in order to recrystallize the implanted layers, laser an-
nealing has been performed with apulsed excimer laser
(XeCl, A, =308 nm). The pulse duration was 40-ns
FTHM, and the energy density was set at 1.8Jcm .A
computer simulation of the thermal transient under the
present conditions indicates that the melt thickness
exceeds 300 nm. Part of the laser-annealed wafers were
then furnace annealed in anitrogen atmosphere contain-
ing 10% oxygen at a temperature of 600 Cfor 1h.
B. Ion-beam techniques
RBS, NRS, and ion-channeling experiments were per-
formed at Laboratori Nazionali di Legnaro by using
beams accelerated by a2-MV Van der Craaff accelerator.
The nuclear reaction yield from "8 atoms was ob-
tained by using a630-keV proton beam, allowing us to
exploit the maximum of the broad resonance in the
"B(p,a) Be nuclear reaction cross section. At this pro-
ton energy, the reaction probability is approximately con-
stant from the surface to the maximum depth where 8is
located in the samples (300—
400 nm). The yield from Sb
and Si atoms was obtained simultaneously by recording
the backscattered protons following the technique de-
scribed in Ref. 9. As both NRA and proton RBS suffer
from poor depth resolution, the results of such an
analysis are relative to the whole implanted and doped
layer. In order to improve the depth resolution, RBS ex-
periments were also performed using a2.0-MeV He+
beam to obtain the yield from Sb and Si atoms. The
backscattering angle was set in such away as to allow a
depth resolution between than 15 nm. The integrated
beam charge was obtained by using the whole scattering
chamber as aFaraday cup, and the detection solid angle
was calibrated by means of standard samples whose com-
position is known with an accuracy better than 2% (Ref.
10) and 5% (Ref. 11)for RBS and NRA, respectively.
For the channeling experiments the samples were
mounted on athree-rotation-axis goniometer' with two
translation axes allowing us to move the beam impact
point on the sample surface in order to avoid beam-
induced damage efFects. Channeling dips were obtained
for the [001], [101],and [111]lattice directions by tilting
the incident beam direction toward acarefully chosen
random direction. Depth-resolved channeling dips for Sb
and Si were obtained by recording the whole He RBS
spectra for each angular position, and afterwards the ap-
propriate energy windows were chosen aposteriori either
by observing the specific dechanneling features or on the
basis of the TEM observations. In the case of NRA-RBS
experiments (H+ beam), besides the aparticle peak from
the nuclear reaction on "8atoms the Sb and Si proton
RBS yields from energy windows corresponding to the
maximum Sb depth were recorded as afunction of the tilt
angle. Channeling experiments were also used to mea-
sure the tetragonal distortion of the doped layers follow-
ing the procedure described in Ref. 12.
Secondary-ion-mass spectroscopy (SIMS) "B and Sb
concentration profiles in the samples were measured by
means of aCAMECA IMS-4f spectrometer at the Phys-
ics Department of the Padova University. A5.5-keV
02+ primary beam rastered over a250X250-pm area
was used for sputtering, while the positive secondary ion
signals were collected from the 60-pm-diameter central
area. The erosion time to depth conversion was obtained
by means of aTencor Alpha-Step 200 stylus profilometer
which was used to measure the height of the sputtering
crater. The secondary-ion yields were converted into
atomic concentrations by using the RBS and NRA Sb
and 8dose data. In fact, in a11 samples the maximum
concentrations are above 1at %, and.the conversion ob-
tained by conventional low-dose ion-implanted calibra-
tion standards is affected by matrix effects. In order to
characterize the extension and the depth shift of the con-
centration profiles, we conventionally assume, as arefer-
ence, the depth where the concentration drops below 10'
at cm, i.e.,more than two orders of magnitude below
the maximum concentrations.
C. KXAFS
The EXAFS experiments were carried out at the
wiggler station 9.3of the Daresbury synchrotron. Near-
grazing-incidence geometry fluorescence EXAFS was
performed at the Sb Kedge (30491 eV). The incident an-
gle was approximately twice the Si critical angle (70 mil-
lidegrees at this energy). Hence the x-ray penetration
52 ANOMALOUS LOW-TEMPERATURE DOPANT DIFFUSIVITY AND. ..1861
depth was approximately 500 nm; that is, larger than the
Sb- and 8-implanted layers. Therefore, the full implanted
layer was probed during these EXAFS experiments. The
high x-ray energy used explains why very few similar ex-
periments have been carried out. To our knowledge, only
Van Netten, Stapel, and Niesen' performed similar
EXAFS experiments on a70-ppm Sb-implanted silicon
crystal.
The fluorescence detector used was aCanberra 13 Ge
diode detector with an excellent signal-to-noise ratio due
to its energy discrimination. As the samples are mono-
crystalline, this device is also useful to eliminate the
Bragg peaks present in the EXAFS spectra. The signals
from the different detectors were summed all together
after removal of the Bragg peaks. Several scans were per-
formed for each sample in order to obtain better statis-
tics.
For the EXAFS analysis, the first reference sample
used was ametallic Sb foil in order to obtain Sb-Sb back-
scattering and phase information. This reference is ap-
propriate because at high annealing temperatures many
Sb-implanted atoms are in precipitates in the metallic Sb
rhombohedric structure. Each Sb atom has three first-
nearest neighbors (NN's) at 2.90 A, and three second
NN's at 3.36 A.'The second reference used is a2X 10'
Sb and 8at. cm sample, for which it is supposed that
each Sb atom is surrounded by four Si neighbors. Indeed,
after laser annealing, the Sb atoms are in substitutional
sites up to aconcentration of approximately 2X10 '
atoms cm .Bechstedt and Harrison' theoretically es-
timated that the Sb-Si distance in the substitution ap-
0
proximation is 2.52 A.
The EXAFS spectra have been analyzed by using the
spherical wave program EXCURV. 'As the Sb atom is
heavy, the backscattering amplitude is still important at
high-k values; hence the EXAFS signal has been record-
ed and analyzed within a700-eV range. Conversely, the
Batom is light; hence the EXAFS backscattering ampli-
tude is weak and limited to small-k values. Actually, the
8atom looks like avacancy regarding the EXAFS
analysis. This effect has been shown by simulating the
EXAFS spectrum of aSb atom surrounded by either
three Si atoms and one 8atom or three Si atoms and one
vacancy by using EXCURV theoretical backscattering am-
plitudes and phases. No difference has been found be-
tween the two EXAFS spectra. Therefore, no direct in-
formation is obtained for the Sb-8 bond. It will be shown
that the inhuence of 8can be seen through the number of
Sb atoms in precipitates.
D. TEM
TEM observations were performed at Consiglio Na-
zionale delle Ricerche, Istituto di Chimica eTecnologia,
dei Materiali eComponenti per 1'Elettronica (LAMEL),
by using aCM30 TEM/STEM operating at 300 keV.
Cross-sectional TEM (XTEM) was obtained by preparing
the specimens according to the conventional procedures
of gluing, sawing, and mechanical polishing down to 20
micrometers and ion-beam milling to perforation. '
III. RESULTS
A. As-implanted samples
TABLE I. Summary of the investigated samples and of the
annealing treatments. The samples have been named according
to the following scheme: the first number identifies the nominal
implantation dose (X10' at cm )of the elements specified by
the subsequent chemical symbols; the last letter identifies the
treatment, A, as-implanted; L, laser annealed; T, laser +600'C
1-h annealed. In the case of coimplantation of the elements the
nominal dose is the same. The last columns report the dopant
content measured by RBS (Sb) and NRA (B).
Name
2Sb A
2BA
2sbB A
2sbL
2sbT
2SbBL
2SbBT
Annealing
process
as-implanted
as-implanted
as-implanted
laser
laser+ therm
laser
laser+ therm
Sb
dose
2.20+0.03
2.24+0.03
2.22+0.03
2.09+0.03
2.26+0.03
2.30+0.03
B
dose
2.07+0.1
1.92+0.06
1.99+0.04
1.94+0.04
5Sb A
5SbBA
5SbL
5SbT
5SbBL
5sbBT
as-implanted
as-implanted
laser
laser+ therm
laser
laser+ therm
5.40+0.05
5.37+0.05
5.36+0.05
4.95+0.05
5.34+0.05
4.93+0.05
4.86+0.06
4.83+0.06
3.91+0.06
RBS-channeling analysis of the low-dose Sb as-
implanted sample shows that the thickness of the Si sur-
face amorphous layer is 216 nm. This thickness rises to
235 nm after successive 8implantation. For the high-
dose-implanted samples, the amorphous thickness is
about 250 nm for Sb- and Sb+8-implanted samples. The
Sb as-implanted doses, measured by RBS, are reported in
Table I, where the acronyms used for the different sam-
ples are also explained. These doses are about 1l%%uo and
7% higher than the nominal doses for low- and high-dose
samples, respectively. The reduced discrepancy for
high-dose samples is attributed to sputtering. On the
contrary, 8postimplantation appears to have anegligible
sputtering effect, and the 8-implanted doses, measured by
NRA, turn out to be about 3/o lower than the nominal
doses (see Table I), i.e.,in the systematic error of the
measure. SIMS concentration profiles (not shown) are
approximately Gaussians with peaks at adepth of (70+5)
nm for both ions, the maximum concentration being
about 3X10 'and 7.5X10 'at. cm for low- and high-
dose samples, respectively, and for both ions.
The EXAFS Fourier transforms (FT's) of the implant-
ed specimens have only one peak corresponding to the
first NN's [Fig. 1(a)], confirming the amorphous structure
of the implanted layers. The first-NN results for the
amorphous samples are reported in Table II. Whatever
the implantation dose and 8coimplantation or not, the
Sb local environment is 3.2Si atoms at approximately
2.60 A. This Sb-Si distance is larger than the one corre-
1862 A. ARMIGLIATO et al.
TABLE II. Results of the EXAFS analysis. N, R, and 2o. ,
respectively, are the number, the distance, and the Debye-
Waller factor of the first NN's around Sb.
22
10 sII I IjIIIIfIIII)IIIIiIIII)IIIItIIIIiI I I
)pZ1 /-
Sample
name Si R(A) 2O~ (A )
Si Si 1V
Sb R(A)
Sb tpZO
2Sb A
2SbBA
2SbL
2Sb T
2SbBL
2SbBT
5Sb A
5SbBA
5SbL
5SbT
5sbBL
5sbBT
3.3
3.2
4.4
3.9
3.8
3.2
3.1
3
1.1
3.1
2.2
2.61
2.58
2.54
2.55
2.54
2.54
2.59
2.59
2.56
2.59
2.54
2.55
0.014
0.006
0.008
0.006
0.014
0.011
0.007
0.007
0.4
1.4
0.2
2.82
2.87
2.85
)019
.a)
(g
L
Q) 1021
C)
O
)020
Is
IIIIIIiIIII
~aAIksssta
IIIIiIIII)IIIIJIIIItII I I
sponding to substitutional Sb surrounded by four Si
neighbors. '
~019
b)
18 ~
50 100 150 200 250 300 350 400
depth (nm)
B. Low-dose samples
0.9
Os6
(3
/
a2Sb
2SbB
5Sb
5SbB
C0.3
J2
CC
O
C5
~s
P
a
0I
Os8
II
—
2Sb
s
1.5
Distance (A)
s
3.5
s
45 5
FIG. 1. Fourier transform of the EXAFS spectra of {a) the
as-implanted and {b)the laser-annealed samples.
The SIMS concentration profiles of the dopants in the
annealed low-dose samples are reported in Figs. 2(a) and
2(b) for the Sb- and Sb+B-implanted sampies, respective-
ly.As aconsequence of the laser annealing process, the Sb
and Bconcentration profiles exhibit the expected
features. In fact, as-implanted profiles have been redistri-
buted nearly uniformly in asurface layer where the con-
FIG. 2. SIMS concentration profiles of the low-dose samples:
(a) samples 2SbL and 2SbT; (b) samples 2SbBL and 2SbBT.
Solid lines and squares refer to Sb concentration profile, while
the hatched line and circle to B; open and full symbols refer to
samples before and after thermal annealing, respectively.
centration varies from 1to 2X10 'at cm . Moreover,
this uniform composition layer is followed by a tail ex-
tending to depths equal to or larger than those of the as-
implanted amorphous layer, confirming that the laser an-
nealing process leads to the melting of at least all the
amorphous layer, allowing subsequent epitaxial regrowth
from the crystalline substrate. The Sb and Bprofiles in
the coirnplanted sample are nearly the same as expected
for adiffusion in the liquid phase.
As aconsequence of our low temperature thermal an-
nealing process (600'C), no appreciable Sb diffusion is ex-
pected. Conversely, both Sb-implanted and coimplanted
samples show large modifications in the concentration
profiles. The most striking feature is exhibited by the
Sb-implanted sample, where the diffusion proceeds out-
wards, i.e.,against the concentration gradient, and is ac-
companied by asignificant (7%%uo) Sb loss (see Table I). In
the case of the coimplanted sample, the diffusion
proceeds inwards and nearly to the same extent for Sb
and B. In this case no dopant loss is observed. The per-
centage of dopant loss and the concentration profile shift
(difference in the tail position at the reference concentra-
tion of 10' at. cm )are reported in Table III, where the
most important structural features of all the investigated
samples are summarized as well.
The weak beam dark-field (WBDF) XTEM micrograph
of sample 2SbL is reported in Fig. 3(a). It shows ahighly
defective region which extends from the surface down to
adepth of 80 nm and consists of vertical dislocations and
52 ANOMALOUS LG%'-TEMPERATURE DOPANT DIFFUSIVITY AND. . . 1863
V
o+~
o
e5 ~&D
M
+4
o
Q
g5 o
04~ 0
05
ce
ca
g
tQ
+
ce
~+m
p% Q
08
o
an
g.0
oba
Xgg
O
C/I
0
2—.
~~
tv
0
cn
P0
~~
P
6—~
Gn
6
~~
~~
ooA
o
a5 wa5
an
bG a5
0
4P
~~
b0
o
8
K, ~
05
-ocE~
~9p cA +g
Q~g
w=o 3~
~~
V
Nan
~WbG
a5
C5 ~e~
nca
&D
oQ
&D
(Q Q
an
O
a$ g~~ b0
0
o
~M~
0
6
~M
g6
0
6Ea
0
g~cn tD
Q
grQ
g
Q
(Q 46
~g
0
0
0
~W
0
CA
6Q
o0
G4
~W
0
g4 ce
+~~ cn
CP
g
CP
cn cn lQ
000
OOO~OOO
O
I
O
NO
VI
~aOO O OO
O
aa~4
0OOOOO
OO
O
mO
+
O
O
OO
O
O
OO
O
IO
OWIIv
OO~~OO
O
O
O
OO
O
O
O
O
O
O
CP
~~
0
G4
cn +
rA
O
0
+
05 0
~W
0
+
Ct
C/)
o
+tg
0
ba
&Q
a5
an 0
e5 O
IQ
6
~W
cA 0
II~II
OO~OO
O
O4O
rt- O~OOm
IIOII
OOa OOC
Ch
OO
O00
O
OO OO Ch
OO O OOOO
t
OO
OOO
OOO
O
OO
O
O
O
O
I
OO~OO~OO~OO
A. ARMIGLIATO et al. 52
/I RIIIBIU la
~9NSI
(Bl
small dots. The dislocations are probably related to a
columnar regrowth of the implanted layer, due to insta-
bilities in the solid-liquid interface, 'the column boun-
daries being decorated with Sb. The small dots are prob-
ably clusters of point defects. The existence of this defec-
tive region indicates that, even for the lower Sb dose, the
laser treatment is unable to fully recover the crystal
structure of silicon up to the surface of the wafer. It is
also possible to see, in Fig. 3(a), aband of small defects,
located at adepth of 180 nm. Acloser inspection by
high-resolution transmission electron microscopy
(HRTEM) reveals that they are rodlike defects (RLD),
i.e.,(113)-oriented Si platelets. '
In order to compare the e6'ect of the coimplantation, in
Fig. 3(b) the bright-field XTEM image of sample 2SbBL
is reported. No extended defects are observed, indicating
that the simultaneous presence of 8strongly favors the
crystal regrowth. The subsequent thermal annealing of
this sample does not induce any observable TEM change
in the structure.
The e6'ect of the thermal annealing on the Sb-
implanted and laser-annealed sample is shown by the
W8DF XTEM image of sample 2SbT reported in Fig.
3(c). The depth of the surface layer containing threading
dislocations is reduced to 45 nm, and the dislocation den-
sity is reduced with respect to the laser-annealed sample
[compare Fig. 3(a)]. At alarger depth, there is aband, 85
nm wide, consisting of alarge population of small defects,
which in part includes small Sb precipitates, as found by
channeling experiments in the corresponding depth re-
gion and also by EXAFS (see below). Evidence of these
precipitates is given by aweak polycrystalline ring in the
di6'raction patterns (not reported here) taken in that re-
gion, whose interplanar spacing corresponds to that of
(102) planes of hexagonal antimony. Deeper in the crys-
tal, there is ahigh concentration of similar small defects,
peaked around 145 nm. This depth corresponds to the
position of the Si defects observed by channeling (see Fig.
6), thus indicating that they probably consist of small
dislocation loops.
Let us start the description of the channeling analysis
with the best regrown samples, i.e.,the coimplanted sam-
ples. After laser annealing the depth-resolved channeling
analysis performed by employing a2.0-MeV He+ beam
does not show any significant depth feature, and the re-
sults are in perfect agreement with those obtained by us-
ing a0.63-MeV H+ beam, including the expected beam
energy scaling for the critical angle. The aligned spectra
closely resemble those of aperfect crystal. The channel-
ing analysis has been repeated for the three principal lat-
tice directions, i.e.,[001], [101],and [111]axes. The Sb,
B, and Si dips for the [101] axis of sample 2SbBL are
shown in Fig. 4, and compared to that obtained for virgin
Si. The channeling dips for Sb and 8are identical, and
they diAer from the dip relative to the Si matrix only for
aslightly higher minimum yield. The same result has
been obtained for all the investigated directions.
The substitutional lattice location of an impurity leads
to channeling dips with the same width for the host ma-
trix and the impurity as in the present case. The sha-
dowed fraction of the impurity atoms is then easily ex-
tracted by comparing the minimum yield yof the two sig-
nals:
FIG. 3. TEM micrographs taken in the lower-dose samples
(a) 2SbL„WaDF;y) 2Sbm. ,eF, ~cl 2Sbr, WSDF.
where g~ and y~ are the normalized minimum yield of
the impurity and of the host matrix, respectively. For a
substitutional location the shadowed fraction is indepen-
52 ANOMALOUS LOW-TEMPERATURE DOPANT DIFFUSIVITY AND. ..
2tirijr
1.
o04
000.5
tilt angle (degree) 1.5
FIG. 4. 0.63-MeV H+ beam [101] angular scan of sample
2SbBL. X, RBS on avirgin Si crystal; A, RBS on the Si matrix
of the sample;, RBS on Sb; 0, NRA on B. The lines are only
drawn to guide the eye.
dent of the channeling direction and equal to the substi-
tutional fraction. From our data it appears that both Sb
and 8atoms are 98% substitutional in the Si lattice.
However, we must notice that the dip of the matrix
shows asmaller (about 10%%uo) critical angle and ahigher
minimum yield than the dip of virgin Si. This fact can be
interpreted by adistortion of the Si lattice caused by the
high concentration of dopants with different bond length
with respect to Si. The actual Si displacements could be
investigated by using extensive Monte Carlo simulations
which are out of the scope of this work. However, asern-
iquantitative estimate of the displacements can be made
in the framework of the Lindhard continuum potential
model. 'Let us consider the atomic distribution around
the lattice string in the transverse plane. In aperfect
crystal, the position of each atom has aCzaussian distri-
bution, centered on the lattice site, with astandard devia-
tion equal to the rms transverse thermal vibration ampli-
tude, p. Random static displacements in the atomic equi-
librium positions lead to afurther broadening of the
transverse distribution of the atom rows and to acorre-
sponding narrowing of the channeling dip as if the
thermal vibration amplitude was increased. By compar-
ing the experimental critical angles of the Si matrix and
of avirgin Si crystal, the static broadening ArM of the Si
atoms can be derived. Adopant dip narrower than that
of the Si matrix indicates that the dopant transverse dis-
tribution is broader than that of the matrix atoms. In
this case the distribution width hard, can be obtained by
the same procedure as above. The results of this pro-
cedure are reported in Table III for the [001] case, even
though limited to the crystal regions where extended de-
fects are not revealed by TEM.
It is worth noting that the static displacement values
are well above the error bar and of the order of p, i.e.,
about 0.1A, perpendicular to the [001] direction. If the
actual atomic displacements are assumed to be in the
(111) directions, along which the covalent bond is
directed, adisplacement of the order of 0.2Ais obtained.
This value agrees very well with the difference in bond
length between Si—
Si (2.35 A) and Si—
Sb (2.54 A) or
Si—
8(2.06 A) thus giving meaning to the above analysis.
By looking at the Ar values in Table III it appears that
nowhere but in sample 2SbBI. can lattice sites occupied
by Sb and Bbe considered perfectly substitutional.
Moreover, the Si matrix is always distorted so that the
concept itself of perfectly substitutional becomes some-
what ambiguous. Nevertheless, in most cases the dis-
placements are small so that, for simplicity, we shall refer
to these dopant off-center site locations as substitutional
sites (in the distorted matrix). As the eff'ect of the off-
center site occupation is not only to narrow the dip but
also to increase the minimum yield, it follows that the
substitutional fraction calculated through Eq. (1) must be
considered as alower estimate of the actual value. The
substitutional fractions calculated by using Eq. (1) are re-
ported in Table III.
By comparing the SIMS concentration profiles and the
results of the substitutional fraction, it appears that both
dopants are in substitutional position up to aconcentra-
tion of nearly 2X10 'cm .This huge amount of substi-
tutional dopants can be obtained by the nonequilibrium
process of the liquid-phase epitaxy by laser annealing
treatment.
As aconsequence of the thermal annealing on the
coimplanted sample (2Sb8T), the substitutional fraction
of both Band Sb does not change. However, acertain
reordering of the Si matrix accompanied by asignificant
displacement of the dopants is observed (see b,rdata in
Table III). It must be noted that dopant precipitation
does not occur despite the very large supersaturation of
both dopants; in fact, the solid solubility at 600'C is less
than 1X10' cm for both Sb (Ref. 22) and 8(Ref. 23).
The TEM investigation of the Sb-implanted samples
showed the presence of three layers with different defects.
As aconsequence, the channeling analysis has been cor-
respondingly depth resolved and the results are reported
in Table III. As far as the substitutional fraction is con-
cerned, the integral analysis of the whole implanted layer
is also reported for the sake of comparison with the
EXAFS data.
In Fig. 5(a) we report the [001]channeling dips of sam-
ple 2SbL recorded in the intermediate layer (80—
160 nm),
where TEM does not reveal any defect. The Sb and ma-
trix minimum yields are much higher than that of the
reference virgin Si because of the dechanneling in the sur-
face layer where ahigh density of extended defects is
present. Nevertheless, the result for the Sb substitutional
fraction is very close to that of the corresponding coim-
planted sample, indicating that aSb concentration of
2X 10 'cm in substitutional sites is achieved after laser
annealing even without coimplantation.
Adifferent behavior is shown in the region where TEM
suggests the presence of rodlike defects. The channeling
dip for the layer between 160 and 200 nm, shown in Fig.
5(b), indicates that the Sb critical angle is only 60%%uo of
that of the matrix. Nevertheless, the minimum yield is
the same as the one of the matrix. These facts lead to the
conclusion that Sb precipitation, partially coherent with
the Si matrix, occurs in this layer. This fact could be
connected to the RLD decoration, but the present mea-
1866 A. ARMIGI. IATO et al.
0.8
0.6
0.4
0.2
0
1.2
~I~ia
I
0.8
0.6
04
0.2
-05 0.5
tilt angle (degree)
'I IfI1IIIIIsIIt1II I I
1.2
sss1r
a)
Si~
I
damage peak at about 1180 keV, corresponding to a
depth of 160 nm, i.e.,in the region where TEM suggests
the presence of small dislocation loops. Channeling angu-
lar yield analysis shows that at any depth the Sb dip is
narrower than that of the matrix, while the Sb minimum
yields are higher. These facts indicate that the effect of
the thermal annealing has been to displace most of the Sb
atoms from regular lattice sites. In fact, the Arsb values
suggest the presence of precipitation even in the layer
where no extended defects are present. Aclear indication
of Sb precipitates partially coherent with the Si matrix is
found again in the tail region of the concentration profile
(130—
180 nm).
Finally the measured channeling values of the parallel
strain are also reported in Table III. Surprisingly, it ap-
pears always to be positive (tensile) even when only Sb is
implanted and acompressive stress is expected. The ex-
pected surfaces strain values can be computed in the as-
sumption that extended defects causing plastic relaxation
are absent in the epitaxially regrown layer, and by consid-
ering the lattice expansion (contraction) caused by the in-
clusion in substitutional sites of atoms with larger (small-
er) covalent radius than the matrix atoms. It turns out
that
~SI ~dOP
E,
1f —
cdop
FIG. 5. 2.0-MeV He+ beam RBS depth-resolved [001]chan-
neling dips of sample 2SbL and of avirgin silicon crystal; (a)
80—
160-nm depth range; (b) 160—
200-nm depth range. The
meaning of the symbols is the same as in Fig. 4.
surements cannot exclude Sb precipitation in the crystal-
line regions between RLD's.
The Arsb value in this layer is very high, and probably
the corresponding analysis becomes too approximate.
The values are reported in Table III as amarker of the
precipitate presence.
After the thermal annealing (sample 2SbT) the situa-
tion is markedly different. The 2.0-MeV He random
and [001] channeling spectra of this sample are shown in
Fig. 6. The main feature in the aligned spectrum is the Si
0
1
~Random
0
700 900 1100 1300 1500 1700
Energy (keV) 1900
FIG. 6. 2.0-MeV He+ RBS spectra of sample 2SbT collected
in the random direction and in the [001] channeling condition.
The vertical arrows show the surface backscattering energy
from Si and Sb atoms.
where cd,pis the dopant relative concentration obtained
from the SIMS dopant profiles, and rd, and rs; are the
covalent radii of the dopant atom and of Si, respectively.
The covalent radius of Si is deduced by the lattice param-
eter of pure Si. For the covalent radius of Bthe empirical
value given by Pauling (0.88 A) has been used, while for
Sb the present EXAFS value (1.364 A) has been adopted.
It must be noted that this value is in good agreement with
that deduced from the linear expansion coefficient rnea-
sured in Ref. 25. The expected strain values are reported
in Table III for comparison with the experimental values
as well.
The EXAFS FT's of the laser-annealed samples are
shown in Fig. 1(b). The recrystallization of the implanted
layer is evident: aclear second-NN peak exists at 3.84 A.
This distance is characteristic of the second NN in mono-
crystalline Si. No Sb-Sb first-NN contribution has been
found. The EXAFS results are reported in Table II.
Whatever the sample, the Sb-Si distance is approximately
2.54 A. This value is consistent with the theoretical one
calculated by Bechstedt and Harrison' (2.52 A) and the
one measured by EXAFS by Van Netten, Stapel, and
Niesen' in a70-ppm Sb-doped Si crystal (2.53 A).
For the low-dose samples the first-NN number is ap-
proximately 4and, only in the case of sample 2SbT, a
peak corresponding to Sb-Sb couples, with avery low
coordination number (0.4), was found. These features, to-
gether with the information obtained by TEM and chan-
neling, on one hand justify the choice of sample 2SbBL as
the EXAFS reference sample for substitutional Sb; on the
other hand they show that, for the low-dose Sb-implanted
samples, the Sb lattice location is mainly substitutional in
the Si lattice, with some Sb precipitation in the case of
the thermal-annealed sample (2SbT). Amore detailed
52 ANOMALOUS LOW-TEMPERATURE DOPANT DIFFUSIVITY AND. ..1867
analysis will be presented below analyzing the high-dose-
implanted samples.
C. High-dose-implanted samples
The SIMS concentration pro61es of the dopants in the
laser-annealed high-dose samples are reported in Figs.
7(a) and 7(b) for Sb- and the Sb+B-implanted samples,
respectively. While in the case of Sb implantation (sam-
ple SSbL )the concentration profile extends to nearly the
same depth as in the case of the low-dose-implanted sam-
ples, the efFect of the coimplantation (sample 5SbBL )is
now to enhance the diffusion of both dopants greatly. To
our knowledge this effect has been never observed before,
and could be due to alower melting temperature, to a
smaller surface reflectivity, and/or to variations in the
thermal conductivity induced by the high-8 concentra-
tion. As amatter of fact, re6ectivity measurements in the
as-implanted specimens have shown that it decreases
from 55% for the Sb-implanted sample (5SbA )to 37%
for the coimplanted sample (5SbBA ). This fact implies
that adeeper melted front is achieved in the 5Sb8 Asam-
ple, and that the surface remains molten for alonger
time, allowing adeeper diffusion of the dopants in the
liquid phase.
In these laser-annealed samples, the maximum concen-
tration of the dopants is reached at asurface layer of 100
nm. The Sb concentration is roughly 4and 3.5X10 '
at. cm for samples 5SbL and 5Sb8I.,respectively, i.e.,
nearly twice the concentration in the low-dose samples,
while the 8concentration is about 2.5X10 'at. cm
owing to the lower implantation dose (Table I) and to the
deeper diffusion.
The thermal annealing for both types of samples leads
to an outward diffusion accompanied by asignificant
dopant loss. In the case of the Sb-implanted sample
(5SbT) the concentration tail shift is of about 35—
40 nm
and the Sb loss is 9%. In the inset of Fig. 7(a) the con-
centration profile is reported on alinear scale, allowing us
to show an anomalous peak occurring around 110 nm,
which is related to some kind of Sb segregation (see the
TEM analysis below), and anear-surface concentration
peak which is amarker of the outdid'usion process.
In the case of the coimplanted sample the concentra-
tion tail shift is of 30 and 45 nm for Sb and 8, respective-
ly. The higher-8 diffusivity is also consistent with the
IIIIiII I I[II I ItIIIItIIII[IIII]II I I
6
I1Q
1019
O
050 1QO 150 200 25Q 3QQ 35Q 4QQ
depth (nm)
IIIIiIIIIiIIII]IIIIIIIII)III I (IIII
FICi. 7. SIMS dopant concentration profile
of the high-dose samples: (a) samples 5SbL
and 5SbT, (b) samples 5SbBL and 5SbBT. The
meaning of the lines and symbols is the same
as in Fig. 2.
D
~vH
O
O
O
1Qzo
()19
5
2.5
00
50 1QO 150 200 250 300 350 400
depth (nm)
A. ARMIGLIATO et a1.
higher dopant outdid'usion, which is 19% for Band 8%
for Sb (see Table I). In the inset of Fig. 7(b) the concen-
tration profiles are reported on alinear scale in order to
better show the Bnear-surface (55 nm) peak, which is
again an important marker of the backward diffusion and
of the outdiffusion processes.
As already seen in the low-dose Sb-implanted sample,
laser annealing is unable to recover perfectly the lattice
structure owing to the high-Sb concentration. Even
more, TEM analysis confirms that this happens in the
high-dose Sb-implanted sample 5SbL„where the Sb con-
centration is nearly three times greater. In fact, in this
sample extended defects are present, as shown in the mi-
crographs of Figs. 8(a) and 8(b). They consist mainly of
twins up to adepth of 100 nm [Fig. 8(a)]. Ahigh density
of clusters of point defects is also present, extending from
:"&~Fp'0 'P~~?:!.
'S
'I gP:'-"'.'''~..+$pP~yP Py.g.,!pl%.%'..".';~.?Ikt?k""Q.'.-,'..:%!5k:
?.;
@+@I'~~~~QIIlggI")?e~gi~I@yg". .-''-',?? i-?i n~
lII IIIIIII
~il III L?&c
IL~~+~~W,g+4 "'~wi m@s!j!?iW)N.SRt'
~I!
~I!
-''~'IIIII')'p~IIIMIIIQ"P3
'"p&ii
~~wwa~IILe mal~itil~i ...,.„~I.i)III()L~ti~A~
!!,.I. 5?I .!I!I I!,!,.. ..?Nl!!?!Ill ..~? ..
' ' ~!!I.!» ..', ,ll..kiI!iA kiwi' Ul?Il(!?LH QH 85!RIA
FIQ. 8. TEM images taken on the higher-dose samples. (a) 5SbL, BF; (b) 5SbL, %'BDF; (c) 5SbBL, BF; (d) 5SbT, DF; and (e)
5SbBT DF. The last two micrographs have been obtained by including spots from the twins and the Sb precipitates in the objectiv
aperture.
ANOMALOUS LOW-TEMPERATURE DOPANT DIFFUSIVITY AND. ..1869
the surface to 120 nm, as shown by the WBDF micro-
graph in Fig. 8(b). Moreover, deeper in the sample,
there is aband (from 210 to 230 nm) of small defects
quite similar to the RDL observed in the sample 2SbL
[Fig. 3(a)].
When boron is coimplanted (sample 5SbBL ), the sur-
face band of defects evolves as shown in Fig. 8(c). It
spans over arange of about 90 nm and consists of thread-
ing dislocations and twins. The twin density is markedly
lower than the one observed in the case of Sb-only im-
plantation [see Fig. 8(a)], and shallower (maximum depth
of about 60 nm). This is once again due to the stress
compensation induced by the boron coimplantation.
Moreover, as in the low-dose case, the deep defects ob-
served in the 5SbL sample have disappeared.
When the high-dose samples are thermally annealed,
TEM reveals that the twins, which were formed during
the laser annealing, are still present, though, in the case
of the 5SbT sample, for areduced depth extension [100
nm instead of 120 nm; see Fig. 8(d)]. For the coimplant-
ed sample, the main effect has been to annihilate the
threading dislocations [Fig. 8(e)]. In addition, Sb precipi-
tation occurs in both samples, due to the high supersa-
turation [Figs. 8(d) and 8(e)]. This feature is evident in
the dark-field TEM micrograph in Figs. 8(d) and 8(e),
taken with an objective aperture, including spots from
both the twins and the hexagonal Sb precipitates. The
density of precipitates is somewhat lower in sample
5SbBT than in sample 5SbT, in agreement with the
EXAFS results (see below). Moreover, in sample 5SbT
their extension is deeper, reaching 150 nm.
The location of the defects in the thermal-annealed
sample is consistent with the results of SIMS experiments
[see Fig. 7(a)]: the main peak in the profile corresponds
to the twin-rich region. Moreover, in sample 5SbT the
small peak at 110 nm indicates alocal enhancement of
the Sb concentration, due to the observed precipitates.
Finally aWBDF image (not reported here) taken in sam-
ple SSbT reveals the presence of a20-nm-wide band of
small defects, occurring beyond adepth of 220 nm, and
having asize abit larger than the ones observed before
the thermal treatment [Figs. 8(a) and 8(b)]. These are
clearly Si defects, as the dopant profile stops at 200 nm.
As far as the channeling analysis of the high-dose Sb-
implanted samples is concerned, the high density of de-
fects in the surface region makes it dificult, and in some
cases meaningless. In Fig. 9the random and [001]-
aligned RBS spectra for sample 5SbL are reported. It ap-
pears that nearly no channeling occurs in the surface re-
gion because of the high dechanneling rate caused by
twins. Moreover, it also affects the minimum yield
deeper in the crystal. This is the reason why the substitu-
tional fractions of sample 5SbL, reported in Table III, are
greatly underestimated, and why they must be considered
qualitative ones.
The situation becomes even worse after the thermal an-
nealing (sample 5SbT) because of Sb precipitation. As a
matter of fact, channeling occurs in aregion where only
20% of the dopant is present and with avery low substi-
tutional fraction. Nevertheless, by comparing the Sb sub-
stitutional fractions of the two samples beyond the sur-
Random
ySb
G
0
O
o
a
0900 1100 1300 1500
Energy (keV)
1700 1900
FIG. 9. Random and [001]-aligned RBS spectra of sample
5SbL.
face layer, aclear indication of Sb precipitation is ob-
tained.
In the case of coimplanted samples the crystalline qual-
ity is much better, allowing ameaningful channeling
analysis. First of all, the depth-resolved analysis for Sb
shows that the substitutional fraction in the defected sur-
face region is lower than the integral value, while in the
layer below the surface defects it is comparable to that of
the corresponding low-dose samples, in agreement with
the fact that the Sb concentration has fallen below
2X10 'cm, i.e.,the concentration in the low-dose
samples. Moreover, the H+ analysis over the whole
doped layer indicates the same substitutional fractions for
Sb and 8as in the low-dose samples, suggesting that the
8behavior is the same as that for Sb. The thermal an-
nealing appears to cause anearly 10% precipitation of
both dopants in the region where extended defects are
present. The 8precipitation is further supported by the
Arn value, which is equal to that of Arsb and higher than
the value before annealing.
The high values of the Sb and 8substitutional frac-
tions in the high-dose sample require some comments. In
fact, electrical measurement of carrier concentration
profiles of the 2SbL and SSbL samples (unfortunately,
donor-acceptor compensation in the coimplanted samples
prevents meaningful electrical analysis), previously per-
formed, shows that the maximum concentration of elec-
trically active Sb is about 2X10 'cm in both samples.
In the case of sample 5SbL, it is lower than that deduced
from the channeling substitutional fractions and from the
SIMS dopant concentration. This fact indicates that part
of the atoms in substitutional sites are electrically inac-
tive and that, therefore, some passivating dopant com-
plexes were formed.
The strain measurements in these samples are relative
to the region where the dopant concentration begins to
fall, so that the data are only qualitative. However, they
are reported because they show that experimental values
are again more tensile than expected. Finally, in this case
thermal annealing causes astrain reduction probably
connected to the dopant precipitation.
According to the EXAFS result, in all samples the Sb-
1870 A. ARMIGLIATO et aI,.
Si NN distance is consistent with that found in the case
of the low-dose samples, although the average value is
2.56 Ainstead of 2.54 A. The main difference is shown
by the coordination number, which is about 3after laser
annealing and even lower after the thermal annealing. In
the latter case there is also evidence of some Sb-Sb NN;
see Table II.
Considering sample 5SbL, as the Sb concentration is
nearly twice the maximum carrier concentration obtain-
able by the laser-induced liquid-phase recrystallization, it
can be argued that only one-half of the Sb atoms can be
tetrahedrally coordinated with Si atoms. From the
EXAFS average coordination number it then follows that
nearly one-half of the Sb atoms must be surrounded by
two Si atoms and two vacancies. These Si2SbV2 com-
plexes have already been suggested by Nylandsted-Larsen
et al. from Mossbauer experiments. The EXAFS re-
sults do not change in the case of the corresponding
coimplanted sample. To explain the data in this case,
complexes involving B(SizSbB2 or SiSbBV) can also be
invoked, as suggested by Culbertson and Pennycook
and Margesin et al.
Even though in principle the EXAFS technique cannot
supply information about the atomic lattice site location,
by considering all possible NN configurations the Sb lat-
tice location can be inferred. In our analysis, by consid-
ering previous results, three possible configurations are
considered: (i) Sb in asubstitutional site (four Si NN at
2.54 A), (ii) Sb in acomplex (two Si NN at 2.54 A), and
(iii) Sb in metallic precipitates (three Sb NN at about 2.90
A). The respective Sb fractions are called T(tetrahedral),
C(complex), and P(precipitate), and can be obtained by
solving the following set of equations:
&Sb-Sb
4T+2C =Xsb s&
T+C+I' =1,
where the experimentally determined average coordina-
tion numbers are used. The overall error on the obtained
fraction values is on the order of 0.1—
0.2. The results of
this analysis are reported in Table III, where they can be
compared to the channeling results.
rv. DZSCUSSrox
Before discussing the most anomalous of our results,
i.e.,the large and unexpected dopant diffusivity, let us
summarize the m.ain structural properties of the samples.
From the results presented it clearly appears that, under
laser annealing, 8coimplantation leads to amuch better
regrowth behavior with respect to the Sb-implanted sam-
ples. This fact, of course, is aconsequence of the strain
reduction induced by B. As aconsequence of the re-
growth from the liquid phase, aboxlike dopant profile is
obtained with anearly constant concentration in asur-
face layer. The dopant concentrations largely overcome
the solid solubility limit at room temperature for both
implanted doses. The successive low-temperature
thermal annealing produces some extended defect reduc-
tions, an increase of the dopant static displacements from
substitutional lattice position, and an increase of the
dopant complex fraction and/or of dopant precipitation,
owing to the large supersaturation.
The channeling results indicate that in the coimplanted
samples the substitutional fraction and the Ar values are
the same for the two dopants, indicating some dopant
coupling. Moreover, in the low-dose-coimplanted sam-
ples acomplete substitution is obtained after the laser an-
nealing, and maintained after the thermal annealing,
despite the large 8supersaturation. This result is not
surprising in view of the nonequilibrium process involved
in the laser annealing, while the pairing hinders the pre-
cipitation of both Sb and 8in the 2SbBT sample. In fact,
after thermal annealing aperfect lattice regrowth is ob-
tained as well (2SbBT). For the corresponding high-dose
samples, where extended defects are present, some Sb and
8precipitation, increasing after thermal annealing, is evi-
dent also after laser annealing. In the case of Sb implan-
tation at low dose agood regrowth, accompanied by a
high-Sb substitutional fraction, is obtained after laser an-
nealing, the successive thermal treatment causing alarge
precipitation. In the case of the high-dose samples, un-
fortunately, the occurrence of alarge density of twins
makes it impossible to achieve ameaningful channeling
analysis.
Strain measurements always give higher values than
expected. Moreover, in the case of Sb implantation,
where acompressive stress is expected, atensile strain is
found. In order to explain these results it is necessary to
assume that alarge number of vacancies have been incor-
porated into the crystal during the liquid-phase regrowth.
On the other hand, in the case of samples 2SbL and
2Sb7; EXAFS results indicate that no Sb-V complexes
are formed. Thus it appears reasonable to assume that
the vacancies are related to some Si-V complex. This hy-
pothesis could also explain the high distortion of the Si
matrix observed by channeling.
The most important EXAFS result is the clear indica-
tion of Sb complex formation, particularly evident in the
high-dose samples. By comparing the EXAFS and chan-
neling results it can be concluded that the Sb atom in a
complex is in asubstitutional site. In fact, the sum of the
Tand CEXAFS fractions fits very well the channeling
substitutional fraction within the respective errors. As
for the structure of the Sb complexes, it clearly appears
that they are Sb V2 in samples 5SbL and 5SbT, where Bis
not present.
In the coimplanted samples the possible formation of
SbB2 or Sb VB complexes could be inferred from the evo-
lution of the complex fraction and from the increase of
the dopant hr values after thermal annealing. In fact by
looking at the EXAFS data for the high-dose Sb-
implanted samples, it appears that the effect of the
thermal annealing is to cause the precipitation of the
former substitutional fraction, leaving the complex frac-
tion unaffected. Conversely, in the case of the corre-
sponding coimplanted samples, amuch lower Sb precipi-
tation accompanied by an increase of the complex frac-
tion is observed. However, the lack of sensitivity of the
EXAFS technique in the discrimination between vacan-
52 ANOMALOUS LOW-TEMPERATURE DOPANT DIFFUSIVITY AND. ..1871
cies and 8atoms and the impossibility, for the channeling
technique, to recognize the mutual coordination of the
dopants do not allow us to reach adefinite conclusion.
The most striking feature of the present experiment is
undoubtedly the significant Sb difFusivity at 600 C(Figs.
2and 7). This is not accounted for by the extrapolation
at 600 Cof the Sb diffusion coeKcients reported in litera-
ture. ''The extent of the profile shifts and of the
outdiffusion observed in the various samples after the
thermal treatment at 600 Chave been reported in Table
III. From this table a profile shift of 50 nm (at 10' at
cm )is deduced for specimen 2SbT. In addition, this
shift is backward oriented, i.e.,toward the specimen sur-
face; this latter phenomenon will be discussed below.
To explain the occurrence of ashift, which is arather
surprising result, we have tried to simulate the experi-
mental profile assuming the usual value of the diffusion
coefficient for Sb (Ds& )that can be computed by extrapo-
lating to 600'C the Fair's formula
2
Dsb =D; +D; l
(4)
where D; and D;, respectively, are the neutral and ion-
ized intrinsic diffusion coefficients due to the neutral V
and to the charged Vvacancies, while nand n; are the
electron and the intrinsic carrier concentration, respec-
tively. For Sb-doped samples investigated in this work,
Ds& depends mainly on the factor (n/n, ).In fact the
maximum Sb concentration in the samples before the
thermal treatment is n=1.8X10 'cm, while
n; =3.54X 10' cm at 600 C, so that one gets
(n/n;) =2.6X10 and Ds&=6X10 'cm /s.
This value introduced in the simulation of the carrier
concentration profile after thermal annealing leads to a
precipitation Sb fraction consistent with the one deter-
mined experimentally. However, it is two orders of mag-
nitude lower than that required to explain the experimen-
tal shift in the concentration profile. Moreover, electrical
measurements performed on the coimplanted samples
gave scattered results, due to the strong compensation,
yet with apositive Hall factor, which indicates apre-
valence of holes over electrons; as such, the Fermi level is
below the band-gap center. Nevertheless, these samples
exhibit more or less the same value of the diffusion
coe%cient, thus indicating that the position of the Fermi
level is not the reason for the huge difFusivity observed.
This conclusion is further supported by the fact that in
the simulations of 2SbBT, i.e.,the only sample where the
diffusion shift after thermal annealing is forward, i.e.,
normal, the shape of the experimental tail (Fig. 2) sug-
gests that the diffusivity is concentration independent; in
fact an increased ionized diffusion coe%cient D; is not
able to reproduce it. Conversely, increasing the neutral
intrinsic diffusion coefficien D,-by seven orders of mag-
nitude reproduces the shape of the experimental concen-
tration profiles. The reason for this anomalous behavior
can be the large amount of vacancies incorporated into
the crystal lattice after laser annealing, and demonstrated
by our strain measurements. In fact the diffusion activa-
tion energy is in large part determined by the vacancy
formation energies, which are 2.6eV for V" (Ref. 29) and
3.38 eV for V.As aconsequence, in our experimental
conditions the activation energy should be strongly re-
duced (about 1.2eV), leading to the seven orders of mag-
nitude increase of the diffusivity. Moreover, in astrongly
compensated sample only the state of charge of Vis
stable, the energy level of the charge state of Vbeing
higher than V, meaning that the main fraction of the va-
cancies are of the V" type. This fact is confirmed by our
EXAFS results after laser annealing, indicating that the
vacancies do not form complexes with Sb where they take
acharged state but, on the contrary, must form com-
plexes only with Si in the neutral state of charge, the oth-
er possible complex 8-V being energetically unfavorable.
The presence of alarge fraction of V" can inhuence only
D;, explaning the concentration independence of the
dlffuslon.
As far as other thermal-annealed samples are con-
cerned, the most surprising efFect is the backward
diffusion (i.e.,against the concentration gradient). From
acomparison between tail diffusion and defect type and
density, it comes out that the presence of interfaces (e.g.,
the precipitate surfaces) as well as of extended defects
close to the tail of the concentration profile are the origin
of the backward diffusion. For instance, in sample 2SbT,
where ahigh density of defects and 10%%uo of Sb precipi-
tates have been found, the profile shifts backwards by 50
nm, whereas in the defect-free sample 2SbBT the
diffusion is normal (i.e.,onwards) by 50 nm for Sb and 40
nm for B. Therefore, the observed backward diffusion is
anet effect of two phenomena (a usual thermal and an
anomalous one) which move the atoms in opposite direc-
tions along the depth of the specimen. At the higher
dose, the Sb backward diffusion is the same 30—
40 nm at
10' cm both for 5SbTand 5SbBTsamples. This indi-
cates that (i) boron does not infiuence the anomalous Sb
diffusivity, and (ii) Sb diffuses not only towards the pre-
cipitates, whose fraction decreases from 50 to 10%%uo, but
also through the complexes, whose density increases by
adding boron. The backward diffusivity of boron is still
larger in the 5SbBT sample (—
45 nm); this suggests that
8diffuses in connection with SbBVcomplexes.
Another phenomenon which has been observed in this
work, and confirms the previous discussion, is the
significant outdifFusion of both Sb and 8after thermal an-
nealing (Table III). Again, it cannot be explained
without assuming the strongly enhanced diffusivity of the
dopant at 600 'C. The outdiffused fraction of Sb
moderately increases (from 7%%uo to 9%%uo) by increasing the
Sb dose, and is strongly related to the crystal defects. In
fact, in sample 2SbBT, where no defect is present, there is
no outdiffusion of the two dopants; on the other hand,
twins and threading dislocations seem to affect more
strongly the outdiffusivity of boron than of Sb. This is
evident in sample 5SbBT (the defects remaining after the
laser treatment are visible in Fig. 8), where the
outdiffusion is 19%for boron and 9% for Sb. From these
data it is impossible to determine the behavior of the 8
outdiffusion with dose and defect density, but it seems to
be larger than the one of antimony.
In overall, the experimental results indicate that the
1872 A. ARMIGLIATO et al. 52
position of the Fermi level appears to affect the move-
ment of the atoms around aprecipitation site, thus result-
ing in lower Sb precipitation in coimplanted samples, but
not a long-range diffusion. Conversely, the very high car-
rier concentration present after laser annealing and the
inclusion of ahigh density of vacancies produces strong
internal stresses inducing the structure to react, on one
side, with extended defects, and, on the other, with adis-
torted lattice matrix where the dopants have ahigh prob-
ability to join in complexes. These two facts guarantee a
huge increase of the dopant mobility during aheat treat-
ment at temperatures as low as 600'C, giving rise to all
the observed phenomena, i.e.,precipitation, outdiffusion,
and backward diffusion.
V. CONCLUSIONS
In this work it has been demonstrated that when high
doses of antimony (2 or 5X10' cm )are implanted into
silicon and then pulsed laser annealed, thus achieving a
very high electron concentration in the heavily doped re-
gion, the Sb difFusivity is much larger than expected even
for thermal treatments in furnace at temperatures as low
as 600 C. This also occurs in the case of silicon wafers
codiffused with the same doses of Sb and B, and thermal-
ly processed in the same way. In this case, after laser an-
nealing, astrongly compensated silicon is obtained in the
heavily doped zone. Moreover, in some cases antimony
and boron outdifFuse from the surface and, still more
surprisingly, backdiffuse toward the surface in the tail re-
gion of the concentration profile. This behavior is strong-
ly related to the amount and type of crystal defects in-
duced by the different thermal processes. From the ex-
perimental results the following conclusions can be
drawn.
(i) The increased Sb diffusivity cannot be explained on
the basis of avariation of the Fermi-level position, but by
ahuge increase, by seven orders of magnitude, in the
concentration-independent preexponential Do term in the
equation of the diffusion coefIIicient. This is due to the
lattice defects, including the high concentration of vacan-
cies which must be generated by laser annealing, as indi-
cated by the strain measurements.
(ii) The dopant outdiffusion is favored by the incom-
plete regrowth of the silicon lattice after laser annealing,
even for the lower dose. Dislocations and twins seem to
provide apreferential path for this phenomenon.
(iii) The backward diffusion (i.e.,against the concentra-
tion gradient) occurs whenever lattice defects are present
in the region of the profile's tail. Sb precipitates seems to
be most effective as asink for this anomalous difFusion.
In the case where no extended defect is detected, the
dopants difFuse inwards, as expected.
(iv) In the codiffused samples, antimony and boron
difFuse together, which is in part related to the formation
of Sb-B pairs, previously reported in literature and
detected by EXAFS in our samples, particularly for the
higher implantation dose.
ACKNOWLEDGMENTS
The authors are indebted to S. Solmi for many stimu-
lating discussions and for critically revising the text. The
technical assistance of F. Corticelli, A. Garulli, and D.
Govoni is also gratefully acknowledged.
~Th. Wicher, M. L. Swanson, and A. F. Quennville, Phys. Rev.
Lett. 57, 1759 {1986).
R. J. Culbertson and S. J. Pennycook, in Beam-Solid Interac-
tions and Transient Processes, edited by M. O. Thompson, S.
T. Picraux, and J. S. Williams, MRS Symposia Proceedings
No. 74 (Materials Research Society, Pittsburgh, 1987), p. 391.
3R. B.Fair, M. L. Manda, and J.J. Wortman, J. Mater. Res. 1,
705 (1986).
4N. E.B.Cowern, Appl. Phys. Lett. 54, 703 (1989).
B. Margesin, R. Canteri, S. Solmi, A. Armigliato, and F.
Baruffaldi, J.Mater. Res. 6, 2353 (1991).
P. E. Andersen, A. Nylandsted-Larsen, P. Tidemand-
Petersson, and G. Weyer, J.Appl. Phys. 71, 755 (1988).
7J. L. Allain, A. Bourret, J. R. Regnard, and A. Armigliato,
Appl. Phys. Lett. 61, 264 (1992).
J. W. Mayer and E. Rimini, Ion Beam Handbook for Material
Analysis (Academic, New York, 1977).
J. U. Andersen, E. Laegsgaard, and L. C. Feldman, Radiat.
Eff. 12, 219 (1972).
C. Cohen, J.A. Davis, A. V. Drigo, and T. E. Jackman, Nucl.
Instrum. Methods Phys. Res. 218, 147 (1983).
iA. Armigliato, G. G. Bentini, G. Ru%ni, G. Battaglin, G.
Dalla Mea, and A. V. Drigo, Nucl. Instrum. Methods 149,
653 (1978).
A. Camera and A. V. Drigo, Nucl. Instrum. Methods Phys.
Res. Sect. B44, 357 (1990).
i3T. J. Van Netten, K. Stapel, and L. Niesen, J. Phys. (Paris)
Colloq. 8, C-1049 (1986).
R. W. G. Wyckoff, Crystal Structures (Krieger, Huntington,
NY, 1982).
i5F. Bechstedt and W. A. Harrison, Phys. Rev. B39, 5041
(1989).
S. J. Gurman, N. Binsted, and I. Ross, J. Phys. C17, 143
(1984).
~7A. Garulli, A. Armigliato, and M. Vanzi, J. Microsc. Spec-
trosc. Electron. 10, 135 (1985).
~sJ. Narayan and J. Fletcher, in Defects in Semiconductors, edit-
ed by J. Narayan and T. Y. Tan, MRS Symposia Proceedings
No. 2(Materials Research Society, Pittsburgh, 1981),p. 191.
A. Parisini and A. Bourret, Phys. Mag. 67, 605 {1993).
20L. C. Feldman, J. %'. Mayer, and S. T. Picraux, Materials
Analysis by Ion Channeling (Academic, New York, 1982).
J. Lindhard, K. Dan. Vidensk. Selsk. Mat. Fys. Medd. 34,
NO. 1(1965)~
2~D. Nobili, Properties ofSilicon, EMIS Datareviews Series No.
4(INSPEC, London, 1988), p. 401.
23D. Nobili, Properties ofSilicon (Ref. 22), p. 384.
24L. Pauling, The Nature ofthe Chemical Bond (Cornell Univer-
sity Press, Ithaca, NY, 1960),p. 246.
25G. Celotti, D. Nobili, and P. Ostoja, J. Mater. Sci. 9, 821
52 ANOMALOUS LOW-TEMPERATURE DOPANT DIFFUSIVITY AND. ..1873
(1974).
A. Nylandsted-Larsen, F. T. Pedersen, G. Weyer, R. Galloni,
R. Rizzoli, and A. Armigliato, J.Appl. Phys. 59, 1908 (1986).
2R. J. Culbertson and S. J. Pennycook, in Beam-Solid Interac-
tions and Transient Processes (Ref. 2), p. 391.
S. Solmi, F. Baruffaldi, and M. Derdour, J. Appl. Phys. 71,
697 (1992).
~~R. P. Messmer, in Lattice Defect in Semiconductors, 1974 (The
Institute of Physics and Physical Society, Bristol, 1975), p. 44.
R. B. Fair, in Impurity Doping Processes in Silicon, edited by
F. F. Y. Wang (North-Holland, Amsterdam, 1981), pp.
317-442.
