Porous anodic alumina optimized as a catalyst support for microreactors

Article abstract of DOI:10.1016/j.jcat.2004.06.016

The optimization of process conditions for making porous anodic alumina as a catalyst support in monolithic microreactors was presented. The basic process involved direct current anodization of 1100 alloy aluminum in oxalic acid. Electrolyte concentration, temperature, and anodization potential were optimized with respect to oxidation efficiency and pore density via a Box-Behnken experimental design to the values of 0.6 M, 18°C, and 30 v, respectively. The results illustrated the effective application of anodic alumina films as catalyst supports in monolithic microreactors constructed from Al alloy. The films magnified the original metal surface area up to three orders of magnitude, and up to four orders of magnitude when a hydrothermal-thermal treatment of the film was employed after anodization. The films adhered well to the Al substrate and served as an effective catalyst support capable of dispersing transition metal catalysts to a reasonable extent. The performance of Al-alumina microreactors with optimized porous anodic films for the decomposition of anhydrous NH₃ using supported metal catalysts was demonstrated. Microreactors of this type could be used various heterogeneously catalyzed reactions carried out on a small scale, especially in mobile applications where monolithic structures are advantageous.

Full text of DOI:10.1016/j.jcat.2004.06.016

Journal of Catalysis 227 (2004) 26–32
www.elsevier.com/locate/jcat
Porous anodic alumina optimized as a catalyst support for microreactors
J.C. Ganley, K.L. Riechmann, E.G. Seebauer, R.I. Masel ∗
Department of Chemical and Biomolecular Engineering, University of Illinois at Urbana-Champaign, 213 Roger Adams Laboratory, Box C-3,
600 S. Mathews Avenue, Urbana, IL 61801-3792, USA
Received 15 April 2004; revised 24 May 2004; accepted 15 June 2004
Available online 20 July 2004
Abstract
This paper describes the optimization of process conditions for making porous anodic alumina as a catalyst support in monolithic microre-
actors. The basic process involves direct current anodization of 1100 alloy aluminum in oxalic acid. Electrolyte concentration, temperature,
and anodization potential are optimized with respect to oxidation efficiency and pore density via a Box–Behnken experimental design to the
◦
values of 0.6 M, 18 C, and 30 V, respectively. The effects of subsequent hydrothermal–thermal treatment on the surface area enhancement
and surface morphology of the porous oxide are also investigated and optimized. The resulting films are employed in the fabrication of active
catalytic aluminum–alumina microreactors for the decomposition of ammonia to hydrogen and nitrogen.
2004 Elsevier Inc. All rights reserved.
Keywords: Anodic alumina catalyst support; Surface area optimization; Ammonia reforming
1. Introduction
areas may be conformally located within confined microre-
actor geometries. Furthermore, the films offer very little re-
sistance to diffusive mass transfer, particularly compared to
extruded or sol-derived alumina [11]. Note, however, that
the specific surface area of anodic alumina lies far below
that of traditional alumina powders or pellets. Other attrac-
tive properties of anodized films include strong adhesion to
the substrate and nontrivial hardness [17,18].
Previous work from this laboratory has demonstrated the
use of anodized alumina in the context of microreactors
for ammonia decomposition [19]. However, the synthesis
procedure reported in that work yields films only 60 µm
thick with specific surface areas near 16 m2/g. Commer-
cially viable microreactors require thicker films and higher
specific surface areas, however. The present work develops
optimized procedures for accomplishing these increases—to
yield films from 50 to 90 µm thick with specific areas of over
150 m2/g. In addition, this work demonstrates a hydrother-
mal treatment method for anodized films that boosts their
specific surface areas by a factor of 10. When the alumina is
used to support metal catalysts such as Ru or Ni, this treat-
ment increases the catalyst dispersion by factors around 2
to 3.
The anodic oxidation of aluminum surfaces to form over-
laying films of porous aluminum oxide, also known as alu-
mite, is a common process employed in the metal finishing
industry as a first step to passivate and protect aluminum
surfaces from corrosion and abrasion. Under appropriate
conditions on properly prepared substrates, a regular array
of nanometer-scale pores can be formed [1–3]. This struc-
ture lends itself to a wide range of applications, including
forms for electroplating metal nanowires [4–6], templates
for the creation of carbon [7,8] or titanium dioxide [9,10]
nanotubes, and catalysts or catalyst supports [11–14]. This
last set of applications concerns the present work.
Anodic films can serve as effective catalysts or catalyst
supports within microreactors for several reasons. The films
may be grown to thicknesses over a hundred micrometers
[15,16] and exhibit significant specific surface areas (typi-
cally 10–40 m2/g) [12,14]. Thus, substantial total surface
*
Corresponding author.
E-mail address: r-masel@uiuc.edu (R.I. Masel).
0021-9517/$ – see front matter
doi:10.1016/j.jcat.2004.06.016
2004 Elsevier Inc. All rights reserved.

J.C. Ganley et al. / Journal of Catalysis 227 (2004) 26–32
27
2. Methodological background
2.1. Optimization of anodization conditions
Many previous experimental and modeling studies have
investigated the phenomenology and mechanisms by which
anodic alumina grows and develops. Key process variables
include the temperature of the electrolyte, the anodization
potential or current density, and the pH or concentration of
the electrolyte [2,14,15,20–22]. The present work employs
a Box–Behnken design [23] with these variables to maxi-
mize the pore density and the oxidation efficiency by which
the anodic layers form. This statistical experimental design
method applies to a wide range of experimental systems
[24–26] and generates a full quadratic empirical model be-
tween variables and experimental responses.
Fig. 1. Typical cross section of anodized 1100 alloy aluminum shim.
2.2. Treatment of anodized films
leading to severe aluminum dissolution [14,20,22]. The for-
mation of porous anodic alumina films in oxalic acid elec-
trolytes has been shown to be optimized between 30 and
60 V [4,21], and our laboratory studies show that aqueous
oxalic acid solutions become saturated above about 0.6 M at
low (near 0 ◦C) temperatures. Therefore, anodization poten-
tial was set to 30, 45, or 60 V; and oxalic acid concentrations
were 0.20, 0.40, and 0.60 M.
Sheets of 1100 aluminum alloy shim (ShopAid, Inc.)
were used as substrates. 1100 Al is a high-purity aluminum
alloy (minimum 99% Al, 0.65% Si and Fe, 0.1% Zn, 0.05%
Mn, 0.05% Cu, 0.15% other). All samples were first de-
greased in acetone, and then anodized for 30 min at the
appropriate process conditions to be tested. The thin an-
odic film was then removed by immersion in a 1.5 wt%
chromic acid and 6 wt% phosphoric acid solution at 60 ◦C
for 15 min [9]. Removal of this first anodic film allowed a
second, thicker film to be grown on a surface free of the
defects and surface scratches common to cold-worked alu-
minum foils. The samples were reanodized for 2 h and dried
in a convection oven at 150 ◦C for 4 h. The anodized films
were analyzed using high-resolution scanning electron mi-
croscopy (SEM, Hitachi S-4700). Plan-view images were
used to determine pore density, and cross-sectional images
yielded the thickness. A typical shim cross-section micro-
graph appears in Fig. 1.
A porous anodic oxide created for decoration or corrosion
protection is typically sealed after formation by treatment in
hot water or steam [27]. Hydration and expansion of the alu-
minum oxide cause the nanopores to close. Although the ex-
act nature of the hydrated oxide is not known, it is believed to
contain boehmite [28], psuedoboehmite [29], and physically
adsorbed water [27]. Hydrothermal treatment of α-alumina
has been shown to produce surface gibbsite–bayerite [30].
Importantly for the present work, the hydrated alumina
hydroxides boehmite, gibbsite, and bayerite may be con-
verted upon dehydration to the high-surface area δ- or
γ -alumina phases under the proper conditions [31]. Cor-
respondingly, dehydration of sealed porous alumina films
causes a significant increase in specific and total oxide
surface area; however, some oxide is lost with successive
treatments after the first [32]. This study subjects anodized
alumina in a microreactor configuration to hydrothermal–
thermal treatments to determine the suitability of this ap-
proach as a fabrication technique. Process conditions are
adjusted to yield the highest surface area of alumina, and the
effects of this procedure on the dispersion of several metal
catalysts supported by the alumina are examined.
3. Experimental
3.1. Anodization optimization
3.2. Hydrothermal–thermal alumina treatment
Box–Behnken optimization of anodization process con-
ditions was carried out by varying the anodization potential
as well as the temperature and concentration of aqueous ox-
alic acid solutions. Each factor was tested at a preselected
low, moderate, or high value simultaneously with the other
two factors as prescribed by a standard three-factor Box–
Behnken experimental design [23–26]. Electrolyte tempera-
ture was set to 0, 10, or 25 ◦C. Other studies have shown that
anodization processes carried out at higher temperatures dra-
matically decreases the oxide formation current efficiency,
Aluminum microreactors were constructed from rolled
bar stock of 1100 aluminum (99+% Al). Electrical discharge
machining (EDM) was employed to cut 14 parallel channels
along the reactor’s length. Each channel was 300 µm wide
and 3 mm deep. The thickness of each wall separating two
channels was 300 µm. Reactors were 9.2 mm wide, 12.6 mm
long, and 4 mm thick. A photograph of a typical reactor is
shown in Fig. 2. Using the optimum anodization conditions
found using the above procedure (30 V, 18 ◦C, 0.6 M oxalic

28
J.C. Ganley et al. / Journal of Catalysis 227 (2004) 26–32
conductivity detector. The detector was calibrated by pass-
ing known mixtures of ammonia, hydrogen, and nitrogen
through the reactor bypass. Control experiments through-
out the temperature range of interest showed that the reactor
housing induced no conversion in the absence of catalyst.
The reactant stream consisted of technical grade (99.99%)
anhydrous ammonia flowing at 92 standard cubic centime-
ters per minute (sccm) and controlled with a calibrated mass-
flow meter. All experiments were carried out at atmospheric
pressure.
4. Results and discussion
Fig. 2. Photograph of an anodized channel microreactor beside a US penny.
4.1. Anodization optimization
acid), the reactors were anodized using the same two-step
procedure outlined in the previous section. Some reactors
were treated one or more times in the following manner: Im-
mersion in deionized water at 100 ◦C for 1 h, drying in a
convection oven at 150 ◦C for 30 min, and then dehydration
at 550 ◦C in a tube furnace under air for 16 h.
The total surface area of each reactor was determined us-
ing single-point BET with a commercial unit (Micromeritics
ChemiSorb 2705), with nitrogen physisorption at 77 K. The
oxide pore structure was examined using high-resolution
SEM (Hitachi S-4700). X-ray diffraction (Rigaku D-Max)
was used to assess the oxide crystallinity. The total pore vol-
ume was estimated by the weighing the reactor before and
after immersion in deionized water at room temperature.
Box–Behnken experiments showed that temperature ex-
erted virtually no influence upon pore density. Thus, a re-
sponse surface showing the influence of the two remaining
variables (anodization potential and oxalic acid concentra-
tion) could be constructed in simple two-dimensional form.
Fig. 3 shows this response surface using data taken at 10 ◦C.
The strong dependence of pore density on anodization po-
tential is apparent in the response surface. This behavior
agrees with the observations of others who have attempted
to “tune” pore diameter by adjusting either the anodiza-
tion potential or the associated anodization current density
[2,14,15,20–22]. The present work sought to obtain the high-
est density of pores possible, which occurred at the lowest
anodization potential (30 V) and highest oxalic acid concen-
tration (0.6 M) as the best process conditions.
3.3. Catalytic reactor tests
EDM-microstructured channel microreactors were an-
odized by the above two-step process under the conditions
determined to be optimum as described above, with the sec-
ond anodization step increased to 16 h to provide an oxide
coating of at least 60 µm thickness. Some reactors were sub-
jected to one hydrothermal–thermal treatment as outlined
in Section 3.2. Metal catalyst was then deposited by wet
impregnation with one of the following solutions: 0.20 M
RuCl3, 0.20 M RhCl3, or 0.44 M Ni(NO3)2 in a solvent of
75% acetone and 25% water. This procedure provided for a
catalyst loading of 1 wt% in each case. The reactors were
dried in a convection oven at 150 ◦C for 1 h, calcined in
a tube furnace under air for 4 h at 550 ◦C, and reduced in
hydrogen at the same temperature. Active metal dispersion
for each reactor was calculated using pulsed CO adsorption
at room temperature in a commercial unit (Micromeritics
ChemiSorb 2705), assuming a 1:1 ratio of adsorbed CO to
exposed catalytic metal.
After each catalytic microreactor was prepared, reactivity
was measured in a quartz tube heated by a temperature-
controlled tube furnace. A few alumina pellets placed up-
stream of the reactor housing served as a reactant pre-
heater. Concentrations in the product stream were monitored
by passing all reactor effluent through an on-line thermal
Fig. 3. Response surface plot of anodic oxide pore density as a function
of oxalic acid concentration and anodization potential. Values shown corre-
spond to an electrolyte temperature of 10 C.
◦

J.C. Ganley et al. / Journal of Catalysis 227 (2004) 26–32
29
Fig. 4 illustrates this surface at the intermediate acid con-
centration (0.4 M). The surface has the form of a hyperbolic
paraboloid. This form confirms several known principles
[20–22] of the anodization process. First, as electrolyte tem-
perature increases, the rate of film formation increases for
any anodization potential. However, temperatures that are
too high also lead to a reduction in current efficiency for
oxide formation by increasing the rate of aluminum oxide
dissolution into the acid electrolyte, reducing the final film
thickness ratio. Also, higher potentials yield correspond-
ingly higher current densities, but these higher currents can
cause local temperature gradients across the film surface,
leading again to reduced current efficiency for the formation
of the oxide.
We sought to obtain the highest anodization efficiency
possible, so Fig. 4 suggests the selection of the lowest an-
odization potential (30 V), which fortunately is consistent
with the requirements demanded by pore density. Fig. 4 also
suggests a temperature of 18 ◦C; pore density was insensi-
tive to this parameter. Pore density demanded use of 0.6 M
oxalic acid concentration, but the thickness ratio does not
depend on this parameter. In summary, the best set of an-
odization conditions for pore density and thickness ratio is
30 V anodization potential, 0.6 M oxalic acid concentration,
and 18 ◦C electrolyte temperature. These conditions we used
for all following anodizations.
Fig. 4. Response surface plot of oxidation efficiency as a function of elec-
trolyte temperature and anodization potential. Values shown correspond to
an oxalic acid concentration of 0.4 M.
Fig. 4 shows a response surface plot for the oxide thick-
ness. Fig. 4 reports thickness data in terms of a ratio that
permits assessment of the efficiency of film formation. The
oxidation efficiency, R, is defined as
4.2. Hydrothermal–thermal treatment
Subjecting microreactors anodized using the optimized
conditions described above to hydrothermal–thermal treat-
ments had a significant effect on the total available surface
area. Fig. 5 shows that untreated anodized reactors had
an average total surface area of about 2.5 m2. After one
hydrothermal–thermaltreatment, this area increased dramat-
ically to about 25 m2. Further treatments caused a gradual
decrease, however.
To better explain the trend observed here, consider the
surface morphology of the treated oxide films (Fig. 6). Grad-
ually, as the number of treatments increases, the pore struc-
ture of the anodic film disintegrates. All that remains after
six successive hydrothermal treatments is a sheet of mi-
crocrystalline alumina similar to that obtained by sol–gel
processes. In contrast to sol–gel processing, X-ray diffrac-
tion analysis of the present oxide did not reveal a dominant
alumina crystal phase (δ, γ , etc.). This observation may be
due to inadequate sample size or interference from the alu-
minum metal substrate.
O
R =
,
I − M
where O represents the thickness of each oxide layer, I rep-
resents half the initial thickness of the aluminum shim, and
M represents half the thickness of the metal remaining be-
neath the film. I and M are halved thickness values so that
we consider only the film grown on one side of the aluminum
shim. The ratio R can be greater than unity in certain cases
due to the lower overall density of the porous oxide film
compared to the aluminum substrate. Such cases represent
a high anodization current efficiency with respect to oxide
formation current, and we therefore sought to maximize R.
Achieving a high anodization current efficiency is important
to help avoid possible significant decreases in reactor fea-
ture size due to excessive aluminum dissolution. In addition,
a high anodization current efficiency promotes the conser-
vation of aluminum metal in the “core” beneath the oxide
layer, which allows for improved heat transfer throughout
the reactor.
Like pore density, the anodization efficiency responded
significantly to anodization potential. Unlike pore density,
however, which responded to oxalic acid concentration but
not temperature, the thickness ratio responded to tempera-
ture but not oxalic acid concentration. Thus, the response
surface for R could still be represented in two-dimensional
form with temperature replacing oxalic acid concentration.
4.3. Microreactor kinetics
Table 1 shows how important physical and kinetic char-
acteristics of the catalysts in each microreactor vary with
the number of hydrothermal–thermal treatments. Figs. 7–
9 show related data for the conversion. For each catalyst,

30
J.C. Ganley et al. / Journal of Catalysis 227 (2004) 26–32
Fig. 5. Total surface area of anodized microreactors after various hydrothermal–thermal treatment cycles.
(a)
(c)
(e)
(b)
(d)
(f)
Fig. 6. Surface morphology of anodized microreactors following (a) zero, (b) one, (c) two, (d) three, (e) four, and (f) five hydrothermal–thermal treatments.

J.C. Ganley et al. / Journal of Catalysis 227 (2004) 26–32
31
Table 1
Catalytic microreactor characteristics
Catalyst
Hydrothermal–
Catalyst
E
a
thermal treatments
dispersion
(kcal/mol)
Ru
Ru
Rh
Rh
Ni
0
1
0
1
0
1
14%
23%
38%
45%
8%
8.2
8.2
8.4
8.4
13.5
12.0
Ni
22%
Fig. 9. Fractional conversion of 92 sccm anhydrous ammonia over
Ni-catalyzed microreactors following (") zero and (2) one hydrother-
mal–thermal treatment.
Fig. 7. Fractional conversion of 92 sccm anhydrous ammonia over
Ru-catalyzed microreactors following (") zero and (2) one hydrother-
mal–thermal treatment.
Fig. 10. Turnover frequencies of ammonia decomposition over (") Ru,
(2) Rh, and (Q) Ni catalysts on microreactors with no hydrothermal–
thermal treatment, plotted in Arrhenius form.
Fig. 8. Fractional conversion of 92 sccm anhydrous ammonia over
Rh-catalyzed microreactors following (") zero and (2) one hydrother-
mal–thermal treatment.
hydrothermal–thermal treatment slightly increased the dis-
persion. Rh catalysts exhibited the highest dispersions (38
and 45%), while Ni catalysts showed the largest increase in
dispersion as a result of treatment (8 to 22%). Reactant con-
versions increased in response to treatment for all catalysts,
probably due to the increases in both the dispersion and the
total surface area of the supporting oxide.
Figs. 10 and 11 show turnover frequencies derived from
the conversion and dispersion data for the various catalysts.
Table 1 shows the activation energies (Ea) derived from the
low-temperature data in these figures, where the rates are
limited by catalyst activity rather than by gas-phase trans-
port. Thermal–hydrothermal treatment did not significantly
Fig. 11. Turnover frequencies of ammonia decomposition over (") Ru,
(2) Rh, and (Q) Ni catalysts on microreactors with one hydrothermal–
thermal treatment, plotted in Arrhenius form.
affect the activation energies, thereby supporting the notion
that treatment altered only the availability of metal atoms
rather than something intrinsic to the reactions themselves.
The magnitudes of the activation energies reported here are
comparable to those reported for ammonia decomposition on
supported catalysts at conditions near atmospheric pressure
[33,34].

32
J.C. Ganley et al. / Journal of Catalysis 227 (2004) 26–32
5. Conclusions
[4] Z. Zhang, D. Gekhtmann, M. Dresselhaus, J. Ying, Chem. Mater. 11
(1999) 1659.
[5] Z. Wang, M. Kuok, S. Ng, H. Fan, D. Lockwood, K. Nielsch,
R. Wehrspohn, Mater. Phys. Mechan. 4 (2001) 22.
[6] T. Gao, G. Meng, J. Zhang, Y. Wang, C. Liang, J. Fan, L. Zhang, Appl.
Phys. A 73 (2001) 251.
[7] J. Li, C. Papadopoulos, M. Xu, M. Moskovits, Appl. Phys. Lett. 10
(1998) 260.
[8] G. Che, B. Lakshmi, E. Fisher, C. Martin, Nature 393 (1998) 346.
[9] A. Michailowski, D. Almawlawi, G. Cheng, Chem. Phys. Lett. 349
(2001) 1.
[10] D. Gong, C. Grimes, O. Varghese, W. Hu, R. Singh, Z. Chen,
E. Dickey, J. Mater. Res. 16 (2001) 3331.
The results presented here illustrate the effective applica-
tion of anodic alumina films as catalyst supports in mono-
lithic microreactors constructed from aluminum alloy. The
films magnify the original metal surface area up to three or-
ders of magnitude, and up to four orders of magnitude when
a hydrothermal–thermal treatment of the film is employed
after anodization. The films adhere well to the aluminum
substrate and serve as an effective catalyst support capable
of dispersing transition metal catalysts to a reasonable ex-
tent. Furthermore, we have demonstrated the performance
of aluminum–alumina microreactors with optimized porous
anodic films for the decomposition of anhydrous ammonia
using supported metal catalysts. Microreactors of this type
could be used for a variety of heterogeneously catalyzed re-
actions carried out on a small scale, especially in mobile
applications where monolithic structures are advantageous.
[11] S. Ihm, E. Ruckenstein, Ind. Eng. Chem. Process Design Dev. 17
(1978) 100.
[12] G. Patermarakis, C. Pavlidou, J. Catal. 147 (1994) 140.
[13] G. Wiessmeier, D. Hönicke, J. Micromechan. Microeng. 6 (1996) 285.
[14] G. Patermarakis, N. Nicolopoulos, J. Catal. 187 (1999) 311.
[15] G. Patermarakis, H. Karayannis, Electrochim. Acta 40 (1995) 2647.
[16] T. Xu, R. Piner, R. Ruoff, Langmuir 19 (2003) 1443.
[17] X. Zhou, G. Thompson, P. Skeldon, G. Wood, H. Habazaki, K. Shi-
mizu, Corrosion 55 (1999) 561.
[18] G. Alcalá, P. Skeldon, G. Thompson, A. Mann, H. Habazaki, K. Shi-
mizu, Nanotechnology 13 (2002) 451.
[19] J.C. Ganley, E.G. Seebauer, R.I. Masel, AIChE J. 50 (2004) 829.
[20] V. Parkhutik, V. Shershulsky, J. Phys. D: Appl. Phys. 25 (1992) 1258.
[21] A. Li, F. Müller, A. Birner, K. Nielsch, U. Gösele, J. Appl. Phys. 84
(1998) 6023.
[22] G. Patermarakis, K. Moussoutzanis, Corrosion Sci. 43 (2001) 1433.
[23] G. Box, D. Behnken, Technometrics 2 (1960) 455.
[24] D. Ludlow, H. Schulz, J. Erjavek, J. Eng. Educ. 84 (1995) 351.
[25] Y. Abdel-Fattah, Biotechnol. Lett. 24 (2002) 1217.
[26] M. Muthukumar, D. Mohan, M. Rajendran, Cement Concrete Com-
posites 25 (2003) 751.
Acknowledgments
This work was supported by the Department of Defense
Multidisciplinary University Research Initiative (MURI)
program administered by the Army Research Office under
Contract DAAD19-01-1-0582. Any opinions, findings, and
conclusions or recommendations expressed in this publica-
tion are those of the authors and do not necessarily reflect the
views of the Department of Defense or the Army Research
Office.
[27] G. Patermarakis, P. Kerassovitou, Electrochim. Acta 37 (1992) 125.
[28] R. Spooner, W. Forsyth, Plating 55 (1968) 336.
[29] K. Wefers, Aluminum 49 (1973) 553.
[30] S. Desset, O. Spalla, P. Lixon, B. Cabane, Colloids Surf. A 196
(2002) 1.
[31] F. Shüth, K. Unger, in: G. Ertl, H. Knözinger, J. Weitkamp (Eds.),
Preparation of Solid Catalysts, Wiley, New York, 1999.
[32] G. Patermarakis, K. Moussoutzanis, J. Chandrinos, Appl. Catal. A 180
(1999) 345.
References
[1] H. Masuda, K. Fukuda, Science 268 (1995) 1466.
[2] O. Jessensky, F. Muller, U. Gösele, J. Electrochem. Soc. 145 (1998)
[33] G. Papapolymerou, V. Bontozoglou, J. Mol. Catal. A 120 (1997) 167.
[34] T.V. Choudhary, A.K. Santra, C. Sivadinarayana, B.K. Min, C.-W. Yi,
K. Davis, D.W. Goodman, Catal. Lett. 77 (2001) 1.
3735.
[3] D. Almawlawi, K. Bosnick, A. Osika, M. Moskovits, Adv. Mater. 12
(2000) 1252.