Full text of DOI:10.1063/1.115133

The role of the surface in transient enhanced diffusion
D. R. Lim, C. S. Rafferty, and F. P. Klemens
Citation: Appl. Phys. Lett. 67, 2302 (1995); doi: 10.1063/1.115133
View online: http://dx.doi.org/10.1063/1.115133
View Table of Contents: http://apl.aip.org/resource/1/APPLAB/v67/i16
Published by the AIP Publishing LLC.
Additional information on Appl. Phys. Lett.
Journal Homepage: http://apl.aip.org/
Journal Information: http://apl.aip.org/about/about_the_journal
Top downloads: http://apl.aip.org/features/most_downloaded
Information for Authors: http://apl.aip.org/authors
Downloaded 16 Aug 2013 to 130.15.241.167. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://apl.aip.org/about/rights_and_permissions

The role of the surface in transient enhanced diffusion
D. R. Lim, C. S. Rafferty, and F. P. Klemens
AT&T Bell Laboratories, Murray Hill, New Jersey 07974
͑
Received 24 February 1995; accepted for publication 22 July 1995͒
The enhanced diffusion of impurities that is seen following ion implantation is rapidly quenched,
hence, the name transient enhanced diffusion ͑TED͒. The quenching of TED is associated with the
annealing of implant damage, either by the diffusion of point defects to the bulk or to the surface.
It is variously assumed that either the surface or the bulk is the predominant annealing site. In this
work, we explore these assumptions by observing the reduction of TED in a buried marker, when
the surface is etched to bring it closer to implanted damage. The results show a considerable
reduction in TED with surface etching, demonstrating that the surface plays a key role in annealing
implant damage. Numerical modeling allows extraction of the surface recombination length of
interstitials at 800 °C. Only a value of 0.1 ␮m is consistent with the data. All but a very small
fraction of implanted interstitials are recombined at the surface with this value. © 1995 American
Institute of Physics.
2
The enhanced diffusion of impurities that is seen follow-
ing ion implantation1 is rapidly quenched, hence, the name
transient enhanced diffusion ͑TED͒. The time constant for
TED has the same activation energy as silicon
self-diffusion, so the quenching of TED is associated with
the annealing of implant damage, either by the diffusion of
boron implant. Silicon damage was then created by implant-
–5
13
Ϫ2
ϩ
ing 2.7ϫ10 cm Si at an energy of 180 keV. The bo-
ron dose was chosen so that diffusion would be intrinsic at
800 °C, while the dose of the silicon implant was chosen so
that no stable extended defects would be formed. After the
silicon implant, 1 ␮m of tetraethylorthosilicate ͑TEOS͒ was
deposited at a temperature of 425 °C using a plasma en-
hanced chemical vapor deposition ͑CVD͒ process. The
2
point defects to the bulk or to the surface. It is variously
assumed in modeling TED6
–10
that either the surface or the
bulk is the predominant annealing site. The distinction has
important practical consequences, since it determines the lat-
eral extent of transient diffusion effects in submicron de-
vices. Griffin et al.11 showed that for the same implant en-
ergy, boron, phosphorus, and arsenic implants caused similar
amounts of TED in a boron marker layer. This result might
suggest that surface recombination is unimportant since the
implant range of arsenic is about a factor of 5 less than that
of boron. On the other hand, Kim et al.12 showed that when
TEOS was patterned and a Cl reactive ion etch ͑RIE͒ was
2
performed to etch back the silicon in the windows to depths
of 650, 1000, 1250, 1750, and 2500 Å. The remaining TEOS
was stripped in HF, and the samples annealed at a tempera-
ture of 800 °C for 2 h in Ar. Boron profiles of samples before
and after annealing were measured using secondary ion mass
spectroscopy ͑SIMS͒.
The depth of the etch was calibrated using three separate
techniques. First, the total oxide/silicon step was measured
immediately after reactive ion etching ͑RIE͒ with a profilo-
meter. As the TEOS thickness was known, it was possible to
ascertain the amount of silicon etched. Second, the etch steps
were measured directly after TEOS removal with a profilo-
meter. Third, the relative position of boron peaks to the sur-
face in unannealed samples was used to determine the etched
depth. The three techniques yielded etch depths which
agreed to within 100 Å of one another.
The roughness at the bottom of the window, as measured
by the profilometer, was less than 100 Å, peak to peak, over
a profile which was 2 mm long. Furthermore, there are no
visible differences between the annealed boron SIMS pro-
files of the sample which was etched by 650 Å and the
sample which was not etched. Thus the surface roughness
had a negligible effect on measurement of the resulting bo-
ron profiles. More importantly, this implies that the RIE etch
did not itself cause transient diffusion.
The SIMS profiles for four samples are shown in Fig. 1;
before anneal, and annealed after 0, 650, and 2500 Å etch.
The other samples lie between the unannealed and the un-
etched cases and have been omitted for clarity. To quantify
the near surface region of an amorphizing BF implant was
2
etched, transient enhanced diffusion was virtually sup-
pressed. However, this experiment was not conclusive since
most of the implant damage was etched away at the same
time the surface was brought closer to the damage.
In the present work, we determine the effectiveness of
the surface as an interstitial recombination sink by the use of
etching to vary the position of the surface in relation to the
damage, before annealing out the transient. The effect of
etching away damage was decoupled from the effect of mov-
ing the surface closer to the damage by using a sequence of
etches, all shallower than the damage. The results indicate
that surface proximity greatly reduces transient diffusion.
The surface can be characterized in terms of a surface recom-
bination length ␭ϭD /k for interstitials, where D is their
I
I
I
diffusivity and k is their recombination velocity. A higher
I
recombination velocity results in a shorter ␭. This work es-
tablishes a value of ␭Ϸ0.1 ␮m for the surface recombination
length of interstitials at 800 °C.
A
diffusion marker was formed by implanting
Ϫ2
1
013 cm boron at an energy of 200 keV into 10–20 ⍀ cm
p-type FZ silicon, followed by an anneal at 1000 °C for 1
min in Ar. This is sufficient to remove damage due to the
the amount of diffusion, boron diffusion lengths
extracted by optimizing the boron diffusivity DB so that
ͱ
Dt were
2302
Appl. Phys. Lett. 67 (16), 16 October 1995
0003-6951/95/67(16)/2302/3/$6.00
© 1995 American Institute of Physics
Downloaded 16 Aug 2013 to 130.15.241.167. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://apl.aip.org/about/rights_and_permissions

FIG. 1. Comparison of SIMS profiles of boron, before anneal and after
anneal with 0, 650, and 2500 Å surface etch ͑shifted by etch depth͒. The
silicon damage profile used to drive TED is also shown for reference.
Gaussian broadening with that diffusivity transformed the
measured implanted profile into the measured diffused pro-
files, as previously described.13 Two-parameter optimization
on the diffusivity and the offset between the profiles was
used to minimize any error due to lack of precision in the
etch depth measurement.
FIG. 3. Calculated depth profile of the integrated interstitial supersaturation
(I/I*)dt for several model assumptions. The result is independent of the
values of D and C* , provided their product is held to a value consistent
͐
I
I
with silicon self-diffusion.
Figure 2 plots the boron diffusion length Dt as a func-
ͱ
tion of etch depth. Although the total broadening is quite
small, the statistically based procedure allows extractions of
a diffusion length of 500Ϯ30 Å for the unetched sample, and
out further calculation it can already be concluded that the
surface played an important role in removing damage.
A full simulation of the diffusion of point defects from
the silicon damage site was carried out using the
PROPHET10 simulator. Silicon interstitial diffusion was
4
30Ϯ30 Å in the 650 Å etched sample. The slight reduction
confirms the visual impression that the RIE process itself did
not contribute to diffusion. The error increases somewhat as
the amount of broadening decreases, so that the 2500 Å
etched sample shows 200Ϯ50 Å of diffusion. That reduction
in diffusion length corresponds to reducing the average bo-
ron diffusivity by at least a factor of 6, even though the etch
treated as Fickian with a diffusivity D in the bulk and a
I
recombination rate k at the surface. Similar equations were
I
solved for vacancies. The initial damage distribution was
taken from TRIM and normalized to give good fits to the
transient diffusion in the unetched sample. The result of
simulations with a recombination length ␭ϭD /k of 10 ␮m
I
I
has removed less than half the silicon damage, according to a
transport of ions in matter ͑TRIM͒ calculation.14 Even with-
͑slow recombination͒ are shown in Fig. 2. This fits the data
poorly. Simulations with recombination lengths of 1 ␮m, 0.1
␮m, 100 and 10 Å are also shown in Fig. 2. Only a recom-
bination length of about 0.1 ␮m is consistent with the data.
At both the longest and shortest recombination lengths, the
result becomes insensitive to the value of ␭, as the surface
tends towards a perfect mirror or perfect sink, respectively.
This extraction of recombination length is remarkably
resistant to changes in the model. The interstitial diffusivity
and background concentration can be varied over the entire
reasonable range without changing the results whatsoever,
provided the D C* product is held constant at the value set
I
I
15
by silicon self-diffusion. As an illustration of this, the dis-
tribution of the integrated interstitial supersaturation
͐
(I/I*)dt, which determines the influence of the damage on
the boron diffusion, is plotted in Fig. 3 for three sets of
model parameters. The first is an estimate based on metal
diffusion,15 the second a minimum value set by the require-
ment that the interstitials must diffuse at least to the boron to
2
cause TED, and the third an upper limit of D ϭ␯a , where ␯
I
is the silicon Debye frequency and a is the lattice constant.
The supersaturations are almost indistinguishable over eight
decades in diffusivity. Intuitively, as the diffusivity is in-
creased, interstitials disperse faster, but the influence of each
FIG. 2. Variation of the square of the boron diffusional displacement Dt
with etch depth. Also shown are simulations with various values of the
surface recombination length ␭ϭD /k . Only ␭Ϸ0.1 ␮m is consistent with
I
I
the observed reduction in TED.
is increased by the concomitant reduction in C* , so that the
I
Appl. Phys. Lett., Vol. 67, No. 16, 16 October 1995
Lim, Rafferty, and Klemens
2303
Downloaded 16 Aug 2013 to 130.15.241.167. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://apl.aip.org/about/rights_and_permissions

TABLE I. Interstitials remaining in bulk as a function of surface recombination length.
Interstitials remaining in bulk
after anneal of
2.7ϫ1013 cmϪ2
180 keV Si implant
Fraction of implanted
interstitials recombined at
surface during TED anneal
Surface recombination
length ␭
1
1
0
1
1
1
0 Å
.3ϫ109 cmϪ2
.5ϫ109 cmϪ2
99.9999%
99.9999%
99.9998%
99.5%
2
2
00 Å
.1 ␮m
␮m
ϫ109 cmϪ2
5
.4ϫ1011 cmϪ2
.3ϫ1012 cmϪ2
.8ϫ1013 cmϪ2
1
3
1
0 ␮m
00 ␮m
87.89%
35.2%
changes self-compensate. In addition, a model was intro-
duced to limit the maximum concentration of free intersti-
tials to represent the effect of interstitial clustering. The in-
tegrated supersaturation again overlays the unclustered
results. The extracted value of ␭, therefore, does not depend
on the other model parameters.
It is interesting to calculate the number of interstitials
which recombine at the surface, as a fraction of the total
number of injected interstitials. In the previous calculations,
starting distributions of vacancies and interstitials were de-
termined from TRIM. If instead we simply introduce a num-
ber of interstitials equal to the implanted dose ͓the ϩ1 model
dose in the surface peak is within 13% of that for ␭ϭ0.1 ␮m.
Therefore, the possibility of a surface impurity pileup driven
by defect gradients is supported by the present work.
In summary, the present experiment shows a strong re-
duction in transient diffusion when the surface is etched to
bring it into proximity to damage created by a silicon im-
plant. It shows that the surface plays a key role in annealing
damage during transient diffusion. Numerical modeling
showed that a surface recombination length ␭Ϸ1000 Å is
necessary to predict the observed reduction in transient dif-
fusion. The surface was calculated to recombine Ͼ99% of
implanted interstitials even for as deep an implant as 180
ϩ
͑Ref. 5͔͒ and follow their destiny using the above model, the
keV Si .
results are given in Table I. For any recombination length
giving reasonable fits to the experimental data, Ͼ99% of
implanted interstitials are removed at the surface during the
anneal. Unlike the extraction of recombination length, the
It is a pleasure to acknowledge many stimulating discus-
sions with H.-J. Gossman and experimental assistance from
R. C. Kistler and C. G. Fleming. The authors also thank our
reviewer for suggesting improvements in the analysis of the
experimental profiles.
exact numbers depend on the separate values of D and
I
C* . Table I was calculated using parameters from Ref. 15.
I
At the lower limit of possible diffusivities, up to 5% of in-
terstitials might escape the surface during the anneal, with
1
T. E. Seidel, D. J. Lischner, C. S. Pai, R. V. Knoell, D. M. Maher, and D.
C. Jacobson, Nucl. Instrum. Methods B 7/8, 251 ͑1985͒.
A. E. Michel, Nucl. Instrum. Methods B 37/38, 379 ͑1989͒.
R. Angelucci, F. Cembali, P. Negrini, M. Servidori, and S. Solmi, J. Elec-
trochem. Soc. 134, 3130 ͑1987͒.
N. E. B. Cowern, K. T. F. Janssen, and H. F. F. Jos, J. Appl. Phys. 68, 6191
95% being recombined ͑using any ␭ that fits the data͒.
2
Higher diffusivities than those used in Table I lead to even
higher estimates of the proportion recombined at the surface.
These results may now be used to explain earlier obser-
vations. As previously mentioned, Griffin et al. showed that
the amount of TED in a boron marker was essentially inde-
pendent of the implant species used to create damage.11 We
believe the interpretation is that for a given energy, As pro-
duces much more damage than B. However, the As damage
is much shallower than B damage. The present work shows
that proximity to the surfaces reduces TED so that these two
effects compensate. Second, in Ref. 10 the assumption of
rapid surface recombination of interstitials was used to ex-
plain how impurities might be drawn to the surface and af-
fect transistor threshold voltage. This work confirms that the
surface is a strong sink for interstitials. Simulations show
that for any ␭ less than 1 ␮m, the effect of the surface on
impurity profiles is similar; even at ␭ϭ10 ␮m, the impurity
3
4
͑
1990͒.
5
6
M. D. Giles, J. Electrochem. Soc. 138, 1160 ͑1991͒.
N. E. B. Cowern, H. F. F. Jos, and K. T. F. Janssen, Mater. Sci. Eng. 4, 101
͑1989͒.
M. Hane and H. Matsumoto, Proc. IEDM 1991, 701.
H. Park and M. E. Law, J. Appl. Phys. 72, 3431 ͑1992͒.
D. Tsoukalas, J. Appl. Phys. 70, 7309 ͑1991͒.
C. S. Rafferty, H.-H. Vuong, S. A. Eshraghi, M. D. Giles, M. R. Pinto, and
S. J. Hillenius, Proc. IEDM 1993, 311.
P. B. Griffin, R. F. Lever, R. Y. S. Huang, H. W. Kennel, P. A. Packan, and
J. D. Plummer, IEDM 1993, 293.
7
8
9
0
1
11
12
Y. Kim, H. Z. Massoud, and R. B. Fair, J. Electron. Mater. 18, 144 ͑1989͒.
H.-J. Gossman, A. M. Vredenberg, C. S. Rafferty, H. S. Luftman, F. C.
Unterwald, D. C. Jacobson, T. Boone, and J. M. Poate, J. Appl. Phys. 74,
13
3150 ͑1993͒.
1
1
4
5
J. F. Ziegler, J. P. Biersack, and U. Littmark, The Stopping and Range of
Ions in Solids ͑Pergamon, New York, 1987͒.
G. Bronner and J. D. Plummer, J. Appl. Phys. 61, 5286 ͑1987͒.
2304
Appl. Phys. Lett., Vol. 67, No. 16, 16 October 1995
Lim, Rafferty, and Klemens
Downloaded 16 Aug 2013 to 130.15.241.167. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://apl.aip.org/about/rights_and_permissions