the ratio between the number of Si and O atoms involved in
the oxide layer is ͑Si: 2 ML͒/͑O: 5 ML͒. This means that the
number of additional Si atoms required for the complete de-
composition of the oxide layer is equal to 1.5 bilayers ͑3
ML͒. Since these Si atoms are supplied by means of surface
mass transport, surface roughness should result after the void
nucleation ͓Fig. 4͑b͔͒. This simple model is consistent with
our STM results, where Si atom consumption corresponding
to 1.5 bilayers leads to hole formation with an area density of
about 50% on the exposed Si surface. ͑This explanation is
suitable for the oxidation of ideal Si͑111͒-7ϫ7 surfaces.͒
Also, after the removal of thicker oxide layers, the surface
morphology becomes rougher due to the formation of multi-
step holes ͓Fig. 4͑c͔͒. This explains the SREM image after
the removal of the 0.9-nm-thick oxide layer ͓Fig. 1͑b͔͒. In
addition, because the same SREM features were previously
observed on the Si surfaces prepared by the Ishizaka and
Shiraki method,13 the decomposition process described
above can be applied for an oxide layer formed by a wet
chemical treatment.
In summary, we have investigated the thermal decompo-
sition of ultrathin oxide layers on Si͑111͒ surfaces by using
in situ SREM and STM. Void nucleation occurs indepen-
dently of interfacial steps, but the void density increases as
the oxide layer becomes thinner. The void growth is domi-
nated by surface mass transport within the exposed Si sur-
faces, where atomic-height holes are generated to supply Si
adatoms. This reaction results in surface roughening after the
thermal decomposition.
FIG. 4. Schematic illustration of the thermal decomposition of oxide layers
mediated by surface mass transport: ͑a͒ describes one bilayer oxidation of
Si͑111͒ surfaces, ͑b͒ shows the oxide layer removal, where single atomic-
height holes are left, and ͑c͒ shows the formation of multistep holes after
removal of the thicker oxide layer.
forming preferential steps determined by the crystal orienta-
tion ͓Fig. 3͑b͔͒. After the complete decomposition of the
oxide layer, the diameter of these holes was over 100 nm
͓Fig. 3͑c͔͒, which is very consistent with the size of the dark
objects in the SREM image ͓Fig. 2͑c͔͒. Moreover, we recog-
nized many 7ϫ7 domain boundaries, which connect these
holes.
Previous work has shown that oxide layers inhomoge-
neously decompose by forming a volatile SiO phase.14 It is
remarkable that, even for 1-ML-thick oxide films, the nucle-
ation sites are independent of the interfacial atomic steps.
Although we cannot identify the origin of the void nucleation
sites, we found that the void density depends on the oxide
layer thickness. These two results mean that the interfacial
structure alone does not determine the void nucleation site.
After forming voids in the oxide layer, thermal decomposi-
tion is dominated by the surface reactions illustrated in Fig.
4. That is, as shown in our STM results, the removal of the
residual oxide layer takes place by means of the edge retrac-
tion of the oxide layers. Since atomic-height holes in the
clean Si surface grow as the oxide area retraction continues,
the Si adatoms, which are needed to form a volatile SiO
phase, appear to migrate to the edge of the residual oxide.
͑Another possible mechanism, though, is vacancy diffusion
on the Si surface from the edge of the residual oxide.͒ In
either case, we can conclude that surface mass transport on
the exposed Si surface governs the void growth kinetics.
When an initial Si surface is not highly oriented, atomic
steps on the surfaces can supply adatoms to the reaction site.
As a result, atomically flat narrow terraces remain between
the steps after the oxide layer is removed.10 However, for a
highly oriented surface, Si atoms for the decomposition are
supplied by creating atomic-height holes within the exposed
Si areas. When we consider one bilayer oxidation ͓Fig. 4͑a͔͒,
The authors are grateful to Dr. Akitoshi Ishizaka of
JRCAT for his valuable comments and discussions, and they
also thank Dr. Shinobu Fujita and Dr. Shigemitsu Maruno
͑JRCAT͒ for their useful information. This work, partly sup-
ported by NEDO, was performed at JRCAT under the joint
research agreement between NAIR and ATP.
1 A. Ishizaka and Y. Shiraki, J. Electrochem. Soc. 133, 666 ͑1986͒.
2 S. Fujita, S. Maruno, H. Watanabe, and M. Ichikawa, Appl. Phys. Lett. 69,
638 ͑1996͒.
3 J. M. Gibson and M. Y. Lanzerotti, Nature ͑London͒ 340, 128 ͑1989͒.
4 M. Liehr, J. E. Lewis, and G. W. Rubloff, J. Vac. Sci. Technol. A 5, 1559
͑1987͒.
5 Y. Kobayashi and K. Sugii, J. Vac. Sci. Technol. A 10, 2308 ͑1992͒.
6 K. E. Johnson and T. Engel, Phys. Rev. Lett. 69, 339 ͑1992͒.
7 Y. Wei, R. M. Wallace, and A. C. Seabaugh, Appl. Phys. Lett. 69, 1270
͑1996͒.
8 Ph. Avouris, I-W. Lyo, and F. Bozso, J. Vac. Sci. Technol. B 9, 424
͑1991͒.
9 H. Watanabe, K. Fujita, and M. Ichikawa ͑unpublished͒.
10 K. Fujita, H. Watanabe, and M. Ichikawa ͑unpublished͒.
11 H. Watanabe and M. Ichikawa, Rev. Sci. Instrum. 67, 4185 ͑1996͒.
12 A. Ishizaka, S. Iwata, and Y. Kamigaki, Surf. Sci. 84, 355 ͑1979͒.
13 M. Ichikawa and T. Doi, in Reflection High Energy Electron Diffraction
and Reflection Electron Imaging of Surfaces, Vol. 188 NATO ASI Series,
edited by P. K. Larsen and P. J. Dobson ͑Plenum, New York, 1988͒, p.
343.
14 See, for example, Y.-K. Sun, D. J. Bonser, and T. Engel, J. Vac. Sci.
Technol. A 10, 2314 ͑1992͒.
Appl. Phys. Lett., Vol. 70, No. 9, 3 March 1997 Watanabe, Fujita, and Ichikawa 1097
129.105.215.146 On: Sat, 20 Dec 2014 16:45:20