Atomic structure of thin dysprosium-silicide layers on Si(1 1 1)

Article abstract of DOI:10.1016/j.susc.2005.11.029

We report on scanning tunneling microscopy results of thin dysprosium-silicide layers formed on Si(1 1 1). In the submonolayer regime, both a 23×23R30° and a 5 × 2 superstructure were found. Based on images taken at different tunneling conditions, a structure model could be developed for the 23×23R30° superstructure. For one monolayer, a 1 × 1 superstructure based on hexagonal DySi₂ was observed, while several monolayers thick films are characterized by a 3×3R30° superstructure from Dy₃Si₅.

Full text of DOI:10.1016/j.susc.2005.11.029

Surface Science 600 (2006) 755–761
www.elsevier.com/locate/susc
Atomic structure of thin dysprosium-silicide layers on Si(111)
*
I. Engelhardt, C. Preinesberger, S.K. Becker, H. Eisele, M. Dahne
¨
Technische Universita¨t Berlin, Institut fu¨r Festko¨rperphysik, Hardenbergstraße 36, D-10623 Berlin, Germany
Received 4 July 2005; accepted for publication 25 November 2005
Available online 27 December 2005
Abstract
We report on scanning tunneling microscopy results of thin dysprosium-silicide layers formed on Si(111). In the submonolayer
pffiffi
pffiffi
regime, both a 2 3 ꢀ 2 3 R30ꢁ and a 5 · 2 superstructure were found. Based on images taken at different tunneling conditions, a struc-
pffiffi
pffiffi
ture model could be developed for the 2 3 ꢀ 2 3 R30ꢁ superstructure. For one monolayer, a 1 · 1 superstructure based on hexagonal
pffiffi pffiffi
DySi2 was observed, while several monolayers thick films are characterized by a 3 ꢀ 3 R30ꢁ superstructure from Dy3Si5.
ꢀ 2005 Elsevier B.V. All rights reserved.
Keywords: Lanthanides; Silicides; Silicon; Thin film structures; Scanning-tunneling microscopy
pffiffi pffiffi
1. Introduction
monolayers thick films form a 3 ꢀ 3 R30ꢁ superstruc-
ture. Using a variety of tunneling conditions, we were able
pffiffiffi
pffiffiffi
Silicides of the trivalent rare earth metals grown as thin
films on silicon surfaces are currently of high interest be-
cause of their low Schottky-barrier heights on n-type sub-
strates [1–5], their epitaxial growth on Si(111) [6–18] and
the self-organized formation of nanowires on Si(001)
[19–23]. These silicide films can be prepared by rare-earth
deposition and subsequent annealing. In several publica-
tions, the atomic structure of films on Si(111) has been
studied using scanning tunneling microscopy (STM), show-
ing a variety of structures mainly depending on the rare-
earth element and its coverage. Most of these studies were
related to erbium silicides. The detailed atomic structure of
the different structures occurring for dysprosium-silicide
films on Si(111), however, is not completely understood
up to now.
to derive the atomic structure of the 2 3 ꢀ 2 3 R30ꢁ
pffiffi pffiffi
submonolayer and the 3 ꢀ 3 R30ꢁ multilayer films.
2. Experimental
Si(111)7 · 7 substrates were prepared by repeated flash-
ing of n-type Si(111) wafers followed by slow cooling
down in order to enable the formation of a defect-free
reconstruction. The silicide films were grown in situ by
depositing dysprosium on the clean Si(111)7 · 7 surface
held at room temperature, and afterwards annealing at
500 ꢁC for 1 min to form the silicide. The base pressure
was lower than 5 · 10ꢂ11 mbar and did not exceed
2 · 10ꢂ10 mbar during preparation. Such vacuum condi-
tions are prerequisite for studying lanthanide silicides in or-
der to prevent oxidation. Dysprosium exposures were
determined using a quartz-crystal microbalance with an
absolute accuracy of 20%. One monolayer dysprosium,
defined using the density of silicon atoms at the unrecon-
structed Si(111) surface, corresponds to a dysprosium
Here we present detailed scanning tunneling microscopy
results on the growth of dysprosium-silicide films in the
range from submonolayer coverages up to a few monolay-
ers. Submonolayer silicide layers are characterized by do-
pffiffi
pffiffi
mains of 2 3 ꢀ 2 3 R30ꢁ and 5 · 2 superstructures. At
one monolayer, a 1 · 1 structure is found, and several
˚
exposure of 2.5 A. Annealing temperatures were controlled
by an infrared pyrometer with an accuracy of about
20 ꢁC. The STM experiments were performed using a
non-commercial instrument. The tunneling tips were
*
Corresponding author. Tel.: +49 30 314 24698; fax: +49 30 314 26181.
E-mail address: daehne@physik.tu-berlin.de (M. Da¨hne).
0039-6028/$ - see front matter ꢀ 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.susc.2005.11.029

756
I. Engelhardt et al. / Surface Science 600 (2006) 755–761
prepared by electrochemical etching of tungsten wires [24]
and subsequent annealing in ultra-high vacuum.
At coverages exceeding one monolayer (Fig. 1(b)), the
substrate surface is not visible any more, and two differ-
ently shaped flat terrace structures appear. As will be dem-
onstrated further below, these terraces are related to the
1 · 1-reconstructed DySi2 monolayer, which is character-
3. Results and discussion
ꢀ
ized by linear terrace edges pointing in ½101ꢃ direction,
pffiffi pffiffi
3.1. The general growth behavior
and a multilayer film with a 3 ꢀ 3 R30ꢁ superstructure
Representative overview images of the typical silicide
structures are presented in Fig. 1. At submonolayer cover-
ages (Fig. 1(a)), the surface is covered by domains of the
and rounder terrace edges consisting of Dy3Si5.
pffiffi
pffiffiffi
3.2. The 2 3 ꢀ 2 3 R30ꢁ superstructure
bare Si(111)7 · 7 and occasionally also the Si(111)5 · 5
pffiffiffi
pffiffi
pffiffi
surface [25] as well as by thin silicide films with 2 3ꢀ
Detailed images of the 2 3 ꢀ 2 3 R30ꢁ superstructure
taken at different polarities of the tunneling voltage are
shown in Fig. 2. This silicide structure is formed exclusively
pffiffi
pffiffi
pffiffiffi
2 3 R30ꢁ and 5 · 2 superstructures. The 2 3 ꢀ 2 3 R30ꢁ
superstructure is characterized by triangular domains sepa-
rated by linear dislocations. At sufficiently high coverages,
also the rather flat monolayer structure becomes visible,
which is shown in a detailed analysis to be characterized
by a 1 · 1 reconstruction.
˚
at coverages up to 0.2 A dysprosium, is dominant at half a
monolayer, and can be observed at higher coverages up to
pffiffiffi
pffiffi
Fig. 1. STM overview images of the different types of dysprosium-silicide
Fig. 2. STM images of the 2 3 ꢀ 2 3 R30ꢁ submonolayer structure at
pffiffiffi
pffiffiffi
sample voltages (a) of +0.6 V and (b) of ꢂ1.4 V. The 2 3 ꢀ 2 3 R30ꢁ
˚
thin film structures (a) at a submonolayer coverage of 1.5 A and (b) at a
˚
coverages of 7 A corresponding to a few monolayers.
unit cell is indicated.

I. Engelhardt et al. / Surface Science 600 (2006) 755–761
757
almost one monolayer. The size and orientation of the sur-
face unit cell can be determined by direct comparison with
that of the neighboring Si(111)7 · 7 surface. The structure
˚
is characterized by about 50 A wide domains separated by
ꢀ
linear dislocations aligned in ½112ꢃ direction, which are
forming a triangular network. From a closer inspection
pffiffi
pffiffiffi
of the structure it is found that the 2 3 ꢀ 2 3 R30ꢁ unit
cell rotates by 60ꢁ at the dislocations.
The general appearance of this submonolayer structure
is found to depend strongly on the tunneling polarity. At
positive sample voltages, ring-like structures are observed
in the image (Fig. 2(a)). In this case, the empty states are
imaged, which are more related to dysprosium atoms be-
cause of their lower electronegativity as compared with sil-
icon atoms. At negative sample voltages, the occupied
states are imaged, which are hence mostly associated with
silicon atoms. Here the images are characterized by wind-
mill-like structures (Fig. 2(b)).
pffiffi
In the following, a structure model of the 2 3ꢀ
pffiffi
2 3 R30ꢁ superstructure and its dislocation network will
be presented. It is based on a direct comparison of images
of the same surface area taken at different tunneling polar-
ities, where the presence of the dislocation lines allowed
determining the correct alignment of the observed features
in both images. In this way it was found that the unit cells
displayed in Fig. 2(a) and (b) coincide, enabling to derive
the relative alignment of the dysprosium and silicon atoms.
In particular it turns out that the dysprosium-related ring-
like features do not coincide in position with the windmill-
like features from the silicon atoms.
pffiffi
pffiffi
Since the 2 3 ꢀ 2 3 R30ꢁ structure is predominantly
observed at coverages around 0.5 monolayers, it can be as-
sumed that six dysprosium atoms contribute to every unit
cell. Furthermore it has to be taken into account that a sat-
uration of the dangling bonds of the substrate is required
to minimize the energy. This means for the present case
that all twelve dangling bonds of the unreconstructed
Fig. 3. Atomic structure model of the submonolayer film with the
pffiffiffi
pffiffiffi
2
3 ꢀ 2 3 R30ꢁ superstructure within the dislocation network (a) over-
pffiffi
pffiffi
Si(111) surface in the 2 3 ꢀ 2 3 R30ꢁ unit cell should
laid on the filled-state image and (b) together with the substrate structure.
be saturated.
The resulting structure model, as shown in Fig. 3, is
based on hexagons consisting of six dysprosium atoms ad-
sorbed at the unreconstructed Si(111) surface, similar to
sium neighbors. This gives rise to an asymmetric bonding
configuration of the silicon atoms, as shown by the filled
blue arrows in Fig. 3(b),1 which indeed matches well the
observations in the filled-state image (Fig. 3(a)) when
inspecting the position and shape of the single protrusions
from the windmill-like features.
pffiffi
pffiffi
the case of the erbium-induced 2 3 ꢀ 2 3 R30ꢁ super-
structure [14,16]. In this way, each of the twelve silicon
dangling bonds in the unit cell can bond to dysprosium
atoms, so that every dysprosium atom saturates two silicon
dangling bonds. Therefore the remaining third electrons of
the generally trivalent dysprosium atoms are assumed to
form a benzene-like hybrid orbital, as indicated by the or-
ange hexagon in Fig. 3(b). Such an electronic configuration
is assumed to stabilize the ring-like structure [16].
In this way, also the other features observed in the filled-
states images can be explained: the windmill-like appear-
ance with a rotational direction of the three protrusions
can now be associated with the tangential alignment of
the blue arrows indicating the bonding asymmetry within
one windmill. Also the observation of deep holes in the cor-
ners of the unit cell and in the center of the windmills can
be related to triple silicon atoms, each of them having only
Among the 12 surface silicon atoms, six atoms are bond-
ing to two dysprosium atoms, while the other six atoms
only have one dysprosium atom in the close neighborhood.
In particular the region in between two dysprosium hexa-
gons is characterized by two silicon atoms with only one
neighboring dysprosium atom and one with two dyspro-
1
For interpretation of color in Figs. 3, 6 and 7 the reader is referred to
the web version of this article.

758
I. Engelhardt et al. / Surface Science 600 (2006) 755–761
one dysprosium neighbor. In contrast, the faint protrusions
in the centers of the dysprosium hexagons stem from triple
silicon atoms, each of them having two dysprosium
neighbors.
dysprosium structure results in an alternating chain of
asymmetric bonding configurations of the silicon atoms,
as marked by the open blue arrows, and therewith in the
slight undulation observed in the filled-state image. These
bonding configurations are structurally different from
those responsible for the windmill-like features, since only
It should be noted that the symmetry of this structure
model should in principle also allow windmill positions ro-
tated by 60ꢁ around the center of the dysprosium hexagon.
However, the filled-state images clearly show three wind-
mills surrounding each dysprosium hexagon. Since the
positions of the three windmills relative to the surrounded
dysprosium hexagon rotates by 60ꢁ when crossing a dislo-
cation line, a tip-related artifact can be excluded. Thus it
has to be assumed that a further polarization of the asym-
metric bonding configurations with a radial shift towards
the center of the windmill occurs, resulting in the observed
images.
The structure model and the assignment of the observed
features are further supported when inspecting the disloca-
tion lines and their intersection nodes in Fig. 2. At positive
sample bias, the dislocation lines show a meander-like
structure, while they appear at negative bias in form of
slightly undulating lines with a considerably lower tip
height than the windmill-like features. This behavior nicely
agrees with a meander-like arrangement of the dysprosium
atoms on top of the unreconstructed silicon surface, as
indicated by the green meander line in Fig. 3(b). Such a
˚
˚
˚
Fig. 4. STM images of the 5 · 2 submonolayer structure at sample
voltages (a) of +1.4 V and (b) of ꢂ1.4 V. The 5 · 2 unit cell is indicated.
Fig. 5. STM images of the surface structure at (a) 1.2 A, (b) 2 A, (c) 3 A,
and (d) 5 A dysprosium exposure.
˚

I. Engelhardt et al. / Surface Science 600 (2006) 755–761
759
one silicon atom bonding to one dysprosium atom is in-
3.4. The 1 · 1 superstructure
volved, explaining the observed difference in contrast.
Using the same arguments, the intersection nodes of the
dislocation lines can be attributed to one dysprosium hexa-
gon, marked in green in Fig. 3(b), where six meander lines
are starting in non-radial directions. However, the different
bonding configuration at the node with all silicon atoms
bonding to two dysprosium atoms results in an electronic
configuration with an average of only 3/2 electrons of every
dysprosium atom necessary to saturate the silicon dangling
bonds. In this way, 3/2 electrons remain at every dyspro-
sium atom making the structure more instable, as indeed
frequently observed in form of defective intersection nodes.
In Fig. 5, the development of the surface appearance as
a function of coverage is shown. Already at half a mono-
layer dysprosium, characteristic flat terraces with straight
ꢀ
edges in the ½101ꢃ direction of the silicon substrate are ob-
served, which are related to the silicide monolayer. This
behavior indicates a significant lateral diffusion of dyspro-
sium atoms during silicide formation. The terraces domi-
nate around one monolayer corresponding to
a
˚
dysprosium exposure of 2.5 A. As will be shown in the fol-
lowing, these terraces have a 1 · 1 unit cell and consist of a
monolayer of hexagonal DySi2. At coverages exceeding one
monolayer, terraces with rounder edges become dominant,
pffiffi pffiffi
3.3. The 5 · 2 superstructure
which show a 3 ꢀ 3 R30ꢁ superstructure characteristic
of hexagonal Dy3Si5.
˚
At coverages above 0.3 A dysprosium, also the 5 · 2-
reconstructed submonolayer structure becomes visible
In order to derive the atomic structure of these terraces,
rather extreme tunneling conditions with high tunneling
currents up to 7 nA and low tunneling voltages around
0.3 V had to be chosen. In this way, the tunneling images
showed sufficient corrugation.
pffiffi
pffiffi
and dominates as compared with the 2 3 ꢀ 2 3 R30ꢁ
structure at coverages close to one monolayer. Detailed
scanning tunneling microscopy results are shown in
Fig. 4. The appearance of this chain-like structure in
In Fig. 61 such high-resolution images are shown for the
silicide monolayer together with a structure model. This
structure is already known to consist of a surface with buck-
led silicon hexagons, while the dysprosium atoms are found
underneath the surface [10]. In this way, the images mainly
show the dangling bonds of the uppermost silicon atoms. At
positive sample bias (Fig. 6(a)), also the electronic states
ꢀ
½101ꢃ direction also varies considerably when changing
the tunneling polarity. However, a closer analysis of this
structure and the development of a structure model is pres-
ently not feasible because of the much smaller available
data basis and possible tip-shape related artifacts in the
images in Fig. 4.
Fig. 6. STM images of the 1 · 1 monolayer structure at sample voltages and tunneling currents (a) of +0.3 V and 7 nA and (b) of ꢂ0.3 V and 3 nA. The
1 · 1 unit cell is overlaid. (c) and (d) show the structure model of the DySi2 monolayer.

760
I. Engelhardt et al. / Surface Science 600 (2006) 755–761
ꢀ
from the silicon atoms in the lower surface layer are ob-
served, so that the complete structure of the surface bilayer
with its buckled hexagons [10] becomes visible.
dominantly aligned in ½101ꢃ direction. The triple layer (3)
has more or less arbitrarily formed edges without any spe-
cific orientation.
Fig. 7(a)1 shows high-resolution data of the multilayer
structure. This structure is known to have a surface consist-
ing of buckled silicon hexagons, while the silicon hexagon
ꢀ
The observed edges in ½101ꢃ direction are in nice agree-
ment with the structure model, since the density of silicon
dangling bonds is minimum for this edge orientation. Sim-
ilar observations were already made for the case of erbium-
silicide monolayers [11–13].
layers underneath are planar with every sixth silicon atom
pffiffi pffiffi
missing, giving rise to the 3 ꢀ 3 R30ꢁ superstructure
[8,17,18]. These vacancies in the second silicon layer have
a strong influence on the tunneling probability, resulting
in the observation of bright triangles with triple protru-
sions and dark holes in between.
pffiffi pffiffi
3.5. The 3 ꢀ 3 R30ꢁ superstructure
In Fig. 5(d) a multilayer dysprosium-silicide film is
shown, where the number of layers is marked. While the
terraces of the DySi2 monolayer (marked 1) still show very
A detailed analysis of the microscopy results leads to a
structure model shown in Fig. 7(b) and (c). The only fea-
pffiffi pffiffiffi
ꢀ
straight edges in the ½101ꢃ direction of the substrate and
ture in the image with the 3 ꢀ 3 R30ꢁ periodicity and
the threefold symmetry of the silicon vacancies is the center
of the bright triangle, which is thus representing the posi-
tion of the vacancy. It is interesting to note that the vacan-
cies are located underneath the lower silicon surface atom.
This behavior, observed here for thin Dy3Si5 films, is sim-
ilar to observations at thick Er3Si5 films [15], but in con-
trast to thin Er3Si5 films where the vacancies were found
underneath the top silicon surface atoms [13]. Thus the
atomic structure of thin Dy3Si5 and Er3Si5 films is found
to be slightly different.
sharp corners, the Dy3Si5 double layer (2) is already char-
acterized by softer corners, although the edges are still pre-
The form of the terrace edges can again be related to the
density of the silicon dangling bonds at the edges: Since bulk
Dy3Si5 is characterized by an ordered arrangement of
vacancies in the planar silicon layers [8], there is almost no
anisotropy in the density of the dangling bonds produced
during edge formation, in contrast to the buckled silicon
layer at the surface of Dy3Si5 or DySi2 that is free of vacan-
cies. At thick Dy3Si5 layers, where the relative influence of
the surface decreases, the terraces thus become rounder.
4. Summary
Using high-resolution scanning tunneling microscopy,
we were able to observe different atomic structures during
growth of dysprosium silicides on Si(111). At submono-
pffiffi
pffiffiffi
layer coverages, structures with 2 3 ꢀ 2 3 R30ꢁ and
5 · 2 periodicity were found. A detailed structure model
pffiffi
pffiffi
could be derived for the 2 3 ꢀ 2 3 R30ꢁ superstructure
as well as for the dislocation network separating the sin-
gle-crystalline domains. The monolayer is formed from
hexagonal DySi2 with a 1 · 1 periodicity, and both silicon
layers in the buckled silicon surface could be resolved. Fi-
nally, also the structure of thin multilayer Dy3Si5 films
pffiffi
could be derived, which is characterized by a
3ꢀ
pffiffi
3 R30ꢁ superstructure induced by silicon vacancies in
the silicide bulk.
Acknowledgements
pffiffiffi pffiffiffi
Fig. 7. (a) STM filled state image of the 3 ꢀ 3 R30ꢁ multilayer
The authors would like to thank the Deutsche For-
schungsgemeinschaft, Projects Da408/5 and Da408/11 for
financial support.
structure at sample voltage and tunneling current of ꢂ0.3 V and 4 nA. The
pffiffiffi pffiffiffi
3 ꢀ 3 R30ꢁ unit cell is overlaid. (b) and (c) show the structure model of
the Dy3Si5 multilayer in top view and side view, respectively.

I. Engelhardt et al. / Surface Science 600 (2006) 755–761
761
[14] T.P. Roge, F. Palmino, C. Savall, J.C. Labrune, C. Pirri, Surf. Sci.
383 (1997) 350.
References
´
´
´
[15] J.A. Martın-Gago, J.M. Gomez-Rodıguez, J.Y. Veuillen, Surf. Sci.
[1] K.N. Tu, R.D. Thompson, B.Y. Tsaur, Appl. Phys. Lett. 38 (1981)
626.
366 (1996) 491.
[16] P. Wetzel, C. Pirri, G. Gewinner, S. Pelletier, T.P. Roge, F. Palmino,
J.C. Labrune, Phys. Rev. B 56 (1997) 9819.
[17] M. Lohmeier, W.J. Huisman, E. Vlieg, A. Nishiyama, C.L. Nicklin,
T.S. Turner, Surf. Sci. 345 (1996) 247.
[18] M. Lohmeier, W.J. Huisman, G. Horst, P.M. Zagwijn, E. Vlieg, C.L.
Nicklin, T.S. Turner, Phys. Rev. B 54 (1996) 2004.
[19] C. Preinesberger, T. Kalka, S. Vandre, M. Da¨hne-Prietsch, J. Phys.
D: Appl. Phys. 31 (1998) L43.
[20] Y. Chen, D.A.A. Ohlberg, G. Medeiros-Ribeiro, Y.A. Chang, R.S.
Williams, Appl. Phys. Lett. 76 (2000) 4004.
´
[2] S. Vandre, T. Kalka, C. Preinesberger, M. Da¨hne-Prietsch, Phys.
Rev. Lett. 82 (1999) 1927.
´
[3] S. Vandre, C. Preinesberger, W. Busse, M. Da¨hne, Appl. Phys. Lett.
78 (2001) 2012.
´
[4] S. Vandre, T. Kalka, C. Preinesberger, M. Da¨hne-Prietsch, J. Vac.
Sci. Technol. B 17 (1999) 1682.
´
´
[5] M. Da¨hne, S. Vandre, C. Preinesberger, S.K. Becker, W. Busse, T.
Kalka, Adv. Solid State Phys. 41 (2001) 227.
[6] J.A. Knapp, S.T. Picraux, Appl. Phys. Lett. 48 (1986) 466.
[7] F. Arnaud d’Avitaya, A. Perio, J.-C. Oberlin, Y. Campidelli, J.A.
Chroboczek, Appl. Phys. Lett. 54 (1989) 2198.
[8] R. Baptist, S. Ferrer, G. Grenet, H.C. Poon, Phys. Rev. Lett. 64
(1990) 311.
[9] P. Wetzel, S. Saitenoy, C. Pirri, D. Bolmont, G. Gewinner, Phys. Rev.
B 50 (1994) 10886.
[10] M.H. Truilier, P. Wetzel, C. Pirri, D. Bolmont, G. Gewinner, Phys.
Rev. B 50 (1994) 2333.
[11] T.P. Roge, F. Palmino, C. Savall, J.C. Labrune, P. Wetzel, C. Pirri,
D. Bolmont, G. Gewinner, Phys. Rev. B 51 (1995) 10998.
[12] T.P. Roge, F. Palmino, C. Savall, J.C. Labrune, S. Saintenoy, P.
Wetzel, C. Pirri, D. Bolmont, G. Gewinner, Surf. Sci. 352–354 (1996)
622.
´
[21] C. Preinesberger, S.K. Becker, S. Vandre, T. Kalka, M. Da¨hne, J.
Appl. Phys. 91 (2002) 1695.
[22] C. Preinesberger, S.K. Becker, M. Da¨hne, AIP Conf. Proc. 696 (2003)
837.
[23] C. Preinesberger, G. Pruskil, S.K. Becker, M. Da¨hne, D.V. Vyalikh,
S.L. Molodtsov, C. Laubschat, F. Schiller, Appl. Phys. Lett. 87 (2005)
083107.
[24] J.P. Ibe, P.P. Bey Jr., S.L. Brandow, R.A. Brizzolara, N.A. Burnham,
D.P. DiLella, K.P. Lee, C.R.K. Marrian, R.J. Colton, J. Vac. Sci.
Technol. A 8 (1990) 3570.
[25] U. Ko¨hler, J.E. Demuth, R.J. Hamers, J. Vac. Sci. Technol. A 7
(1989) 2860.
[13] P. Wetzel, S. Saitenoy, C. Pirri, D. Bolmont, G. Gewinner, T.P. Roge,
F. Palmino, C. Savall, J.C. Labrune, Surf. Sci. 355 (1996) 13.