Journal of Non-Crystalline Solids 151 (1992) 88-94

North-Holland

]OURNAL

OF

Concentration and size depth profile of colloidal silver particles

in glass surfaces produced by sodium-silver ion-exchange
A. B e r g e r
Center for Research in Electro-Optics and Lasers, University of Central Florida, 12424 Research Parkway, Suite 400,
Orlando, FL 32826, USA

Received 12 May 1992

Revised manuscript received 7 July 1992

The sodium-silver ion exchange of a commercial soda-lime silicate glass followed by an annealing process yield small
silver particles in a surface layer. Microspectrophotometryspectra and transmission electron microscopyshow that the mean
particle size increases with the penetration depth, whereas the volume concentration is constant. A model on the basis of a
homogeneous nucleation including ion exchange, diffusion, reduction and particle formation is given. The measured depth
dependence is described with the model by a non-linear system of differential equations.

I. Introduction
Staining of alkali-silicate glasses by exchanging silver ions from pastes or liquids with alkali
ions of the glass yields a yellow coloured surface
layer, which is caused by small silver particles
[1,6]. Following Weyl [1] it is usually assumed that
the formation of these particles is a result of the
complex combination of ion exchange, diffusion
of silver ions, their reduction to atoms and aggregation to particles.
Whereas ion exchange and silver ion diffusion
in glasses are well understood (see, for example,
refs. [2,3]), there are many unsolved questions
concerning reduction and particle formation. Especially the microscopic processes resulting in the
reduction of the silver ions to atoms and their
aggregation to crystalline particles are unknown.
A detailed model describing the formation of the
small silver particles has not yet been given.

Correspondence to: Dr A. Berger, Center for Research in

Electro-Optics and Lasers, University of Central Florida,
12424 Research Parkway,Suite 400, Orlando, FL 32826, USA.
Tel: + 1-407 658 3991. Telefax: + 1-407 658 3955.

Dependent on the glass composition in commercial soda-lime glasses, iron (concentration <<
1 wt%) generally can be found in the Fe 2+ as
well as in the Fe 3 state [4]. Using optical spectroscopy a relation of F e 2 + / F e 3= 0.43 was estimated for the investigated glass [5].
This ratio is important for the reduction of
silver ions (in a non-reducing atmosphere). The
following thermoreducing reaction is to be expected [1]:
Fe 2 + ~ Fe3++ e A g + + e---* Ag .
According to this assumption the maximum number of silver atoms has to be equal to or less than
the number of Fe 2+.
In ref. [6], a method was presented, from which
size as well as volume concentration of silver
particles can be determined from optical extinction spectra. From a measurement of optical
spectra with a microspectrophotometer as a function of penetration depth, an estimate of particle
size and concentration as a function of depth
from the exchanged surface can be made.

0022-3093/92/$05.00 1992 - Elsevier Science Publishers B.V. All rights reserved

A. Berger / Colloidal silver particles in glass surfaces

This paper presents experimental results of

size and volume concentration of silver particles
as a function of penetration depth after a low
temperature ion exchange ( T = 400C) followed
by an annealing process near the transition temperature of the glass, Tg. Size and concentration
were determined using the 'optical' method of
ref. [6] and by transmission electron microscopy.
The results are explained by a model on the basis
of a homogeneous nucleation. The system of
non-linear differential equations explains the formation of silver particles by ion exchange, diffusion, reduction and aggregation. A quantitative
description of the complex process is given.

2. Experiments and results

For all the investigations, samples of a commercial soda-lime glass (for composition, see
table 1) with a thickness of 2.2 mm were used.
The samples were subjected to an ion exchange
in a salt liquid of 2 wt% AgNO3-98 wt% NaNO 3
at 400C for 2 h and 2 wt% AgNO3-98 wt%
KNO 3 at 400C for 30 min, respectively. Electron-probe microanalysis detected a silver containing surface layer with a thickness of 55 and 25
txm, respectively, after the ion exchange [7]. In
the case of the AgNO3-NaNO 3 salt liquid, 80%
of the sodium ions are exchanged by silver ions at
the surface. The weight change of the sample
resulting from ion exchange gave a total amount
of silver of M = 1.85 txm -2, if the silver is assumed to be in an infinitesimal thin surface layer
(appropriate to eq. (20)). From optical spectra, I
conclude that nearly all the silver is in the ionic
state, because there is no absorption caused by
silver atoms or particles. Silver particles cannot
be found by electron microscopy.
The ion exchanged samples were subsequently
annealed at 550, 600 or 650C. By cutting, grinding and polishing, cross-sectional samples with a

89

thickness of 25 Ixm were produced. On these

samples, optical extinction spectra of the silver
particles could be measured with a microspectrophotometer [9] at different depths from the
surface exposed to the salt melt. The beam from
the microspectrophotometer had a width of 10
I~m.

Two thin sections of the glass (thickness = 40

Ixm) were cut perpendicular to the stained surface. These two thin sections were glued between
two aluminium supporting rings in such a way
that their ion exchanged surfaces were contiguous
along a ring diameter. The specimen was then
thinned by ion beam etching until there was a
hole in the centre with a radius greater than the
depth of the coloured region. Afterwards the
sample was covered with a thin carbon film by
vacuum evaporation to avoid charging effects in
the transmission electron microscope (TEM). Ion
beam thinning shaped the edge of the hole to a
wedge allowing electron transmission through the
thinner parts. Due to the specimen geometry,
TEM measurements along the edge of the hole
yield the possibility of imaging the sample at
different penetration depths [10].
Figure 1 shows two typical transmission electron micrographs of silver particles taken at different penetration depths. There are spherical
particles which electron diffraction identifies as
crystalline silver (see refs. [6,13]). The particles
are predominantly monocrystalline, as was shown
by high resolution electron microscopy [10]. In
each volume element, a distribution of the particle sizes occurs, which can be fitted by a Gaussian
distribution [6,13]. The mean particle size changes
with the penetration depth, whereas their volume
concentration is nearly constant, both in correspondence with fig. 3 and fig. 4, respectively.
In fig. 2 a typical change of the optical extinction spectra with the penetration depth can be
seen. Using the method described in ref. [6], the
dependence of particle radius (fig. 3) and volume

Table 1
Glass composition (wt%) (measured by X-ray fluorescence analysis [8])
SiO 2

Na20

CaO

MgO

ml203

KzO

SO 3

Fe203

72.20%

14.20

6.50

4.42

1.49

0.71

0.39

0.13

A. Berger / Colloidal siluer particles in glass surfaces

90

5o

5O
i

45

2Omin/6OO~

2h/600'C
17h/60~C
95h/600"C

40

45

lh/650~C
100h/650"C
100h/600'C
100h/550~

4O

"23o
0)

~o251

~ 20

~2i

o
&

'7

10

lO

,o

~^~o~ ~
qllF ' ' I

o o

q
v

p~v
r

~
v

ee&

vv

300 600 900 1 200

~enetrotion depth (/~m)

1 O0 200 500 400

penetration depth (p.m)

l~ig. 3. Particle radius in dependence on the pentration depth

for sample (a) exchanged in 2 wt% AgNO3-98 wt% KNO 3
and (b) exchanged in 2 wt% AgNO3-98 wt% NaNO 3 after
different annealing processes.

Fig. 1. TEM micrographs at penetration depths of (a) 24 i~m

and (b) 286 I~m after 30 min ion exchange in 2 wt% AgNO 3-98
wt% KNO 3 at 400C and 95 h annealing at 600C.

concentration (fig. 4) on the penetration depth

was determined. (Because of the Gaussian distribution this method yields mean partical sizes.)
The diameters and volume concentrations estimated, thus, are in a good agreement with the
electron microscopical mean partical sizes [6,13].
The results: can be summarized as follows.
(1) After the ion exchange, only ionic silver
occurs in a very small surface layer.

6Q

b
2O

70

I .........

/\
I t
I !.

"~.50

--

t /I

I-

,.C.

.~ 40
.u

x= 30,u.m
x=l 30p.m
x=230#m
x= 280,u.m
x=3OO,u,m

--- -

1 / I
I s 1 ~

x=31

20min/600*C
2h/600"c
17~/800'C
95h/600'c

16

, ioso
i00h/650"C

I
J

14
12

~t0

v0

Oe

10

~ +0

00

~> 4

20

"

16

~12

18

~14

O,u.m

x=3,30/zm
x=350p.m

----

2.0
i

18

ee

Ooo~ OoOo

0
v

2
0
0

550

370

390

410

430

450

470

wovelength (nm)

490

510

5.30

550

Fig. 2. Optical extinction spectra for different penetration

depths, x, after an ion exchange of 30 min in 2 wt% AgNO 398 wt% KNO 3 at 400C and 95 h annealing at 600C.

100 200 500 400

penetrotion depth (#m)

.300 600 900 1 200

lenetrotion depth (,u.m)

Fig. 4. Volume concentration of crystalline silver in dependence on the pentration depth for sample (a) exchanged in 2
wt% AgNO3-98 wt% KNO 3 and (b) exchanged in 2 wt%
AgNO 3-98 wt% NaNO 3 after different annealing processes.

91

A. Berger / Colloidal silver particles in glass surfaces

(2) After the annealing process near Tg, there

are spherical, monocrystalline silver particles
within a surface layer.
(3) In each volume element of this surface
layer exists a Gaussian distribution of particle
sizes.
(4) The mean particle size increases with
growing penetration depth.
(5) At the glass surface, the particle size increases with the duration of the annealing.
(6) After very long annealing processes, the
particles near the end of the surface layer have
very large diameters, which strongly increase with
growing penetration depth.
(7) After annealing, the volume concentration
of silver particles is nearly constant within the
coloured surface layer and decreases at their end
to zero.

3. Theory
After the ion exchange, there is only ionic
silver within a very small surface layer. Two independent processes, ion exchange and reaction
processes during annealing, will be discussed. Silver particles are formed by the annealing process.
Subsequently, three different scales will be assumed.
(1) Within scale 1, the diffusion of ionic silver
takes place. It is assumed to be slow compared
with the reaction processes.
(2) The reaction processes are discussed on a
smaller local scale. In small volume elements, the
concentration of ionic silver and reduction agent
are assumed to be constant. On this scale, mean
values for particle size and concentration are
used.
(3) On a still smaller scale, a spatial varying
concentration of ionic silver and reaction agent,
i.e., within scale 2 an inhomogeneous distribution
results in a size and concentration distribution of
silver particles (as discussed for example in ref.
[11]).
In the following, only scale 2 will be discussed.
Scale 1 is only of interest inasmuch as it gives the
conditions of scale 2 (i.e., the concentration of
ionic silver). Considering this, the simplest model

on the basis of a homogeneous nucleation including diffusion, reduction and particle formation is
given by rate equations (1)-(4):
0i

0 (D0i]_

Ot = d x ~

-~x ]

nklinrn

Or
- - -= - n k li"r n - k z i r N ,
Ot
ON
-- =kli"r",

-kzirN'

(1)
(2)
(3)

0t

0c
Ot = V a [ n k l i n r "

+ kzirU]"

(4)

Equation (1) describes the change of the concentration of ionic silver i (with the exception of the
volume concentration, all the concentrations used
subsequently are numbers of ions, atoms or particles per volume unit). Starting from the silver ion
concentration profile after the ion exchange, there
is a diffusion process (term 1 in eq. (1)) as well as
a decrease of the silver ion concentration by
reducing processes (formation of stable nuclei of
n silver atoms (term 2) and growth of nuclei or
particles (term 3)). The concentration of the reducing agent, r, which is constant, r = r0,
throughout the glass volume before the annealing
process, is decreased by this reducing process in
the same way (eq. (2)). The concentration of the
stable nuclei, which is equal to the concentration
of the particles, is given by N (eq. (3)), whereas c,
the volume concentration of silver particles (i.e.,
particle volume per glass volume, which is equal
the product of particle concentration and particle
volume), is given by eq. (4). Va is the volume of a
silver atom in the particle; k 1 and k 2 are the
rates of nuclei formation and addition of silver
atoms to nuclei or particles, respectively.
If it is assumed that the concentration of ionic
silver is much higher than the concentration of
the reducing agent, and therefore the 'reduced'
silver, the second and third terms in eq. (1) can
be neglected. Equation (1) reduces then to
0t

0x

(5)

i.e., eq. (1) (scale 1) and eqs. (2)-(4) (scale 2) are

separated.

92

A. Berger/ Colloidalsilverparticlesin glasssurfaces

Because of r = r 0 for t = 0 (in correspondence

with the constant concentration of Fe 2+ in the
glass before annealing) and r ~ 0 for t ~ oo (resulting from eq. (2)),
r = r 0 exp( - C l t )

The equilibrium concentrations (i.e., on scale 2:

t ~ oo) are
r ( t --* oo) = 0,

(14)

C ( t ~ oo) = roVa,

(15)

(6)

can be assumed. Since c = 0 for t = 0 and c ~ c 0

for t ~ oo (in correspondence with the measured
constant particle volume concentration), c is supposed to be
c = c o i l - exp( - C z t ) ] .

N ( t ~ oo) = ~ kl--~nin-lrnk2

From c = NV, with V the particle volume, the

particle radius, R, can be calculated as

(7)

Equations (2) and (4) then give c I = c 2 (c 1 and c 2

are constants) and Var o = c o. Since diffusion is
assumed to be slower than reaction, the concentration of ionic silver is constant within the time
of particle formation. With N = 0 for t = 0, from
eqs. (3) and (6) it follows that
klinr~
N= --[1
- exp(-Qnt)].
(8)

C1n

~
R =

-roC 1 exp(-clt ) = -nkli"r ~ exp(-clnt )

kai"r ~
- k z i r o exp( - C l t ) - -

cln

[1 - exp( - c l n t ) ] .

(9)

For sufficiently large values of n and t, the terms

containing e x p ( - c ~ n t )
in eq. (9) can be neglected. Equation (9) then reduces to
klinr~
- r o C 1 e x p ( - C l t ) = - k 2 i r o exp(--Clt ) - -

cln

16"rr2k2Va2
9nkli,_lr~_

(17)

2 .

Since kl, k2, Va and r 0 are constant within the

entire glass volume, the particle radius depends
only on the concentration of ionic silver:
R = A i (1-n)/6,
(18)

~ 16-ffzk2Va2
A =

With this and eq. (6), eq. (2) gives

(16)

(19)

9nklr~_ 2

In case of long annealing times, the total penetration depth, xi, of ionic silver after the ion exchange (25 and 55 Ixm, respectively) is very small
compared with the width, x a, of the particle
containing layers resulting from annealing (350
and 1000 Ixm, respectively), so that the simple
model of diffusion from a plane source in a
semi-infinite medium can be applied. With the
assumption D = const., eq. (5) gives
i

vr~D~ exp - ~

(20)

(10)
With the assumption of constant silver ion concentration within the time of reaction and large
values of n and t, the system of eqs. (2)-(4) gives

[ ~]
r = r 0 exp

-klk2in+lr~
t
17
'

with M the total amount of ionic silver in an

infinitesimal thin layer [12]. Combining eqs. (18)
and (20), the dependence of particle radius on
the penetration depth and annealing time is then
given by

(11)

M
R=A'

Var o 1 - exp

'

(12)

x 2

]/(l-n)/6

e x p [ - ~D-~-]1

(21)

The logarithmic form of eq. (21) is

ln R = l n A

g=~nin-lr~{1-exp[-~klk2nin+lr~t]).

__(n-l)6 In[ ~ D t ] + ( n- - -l )
24

x2
Dt '

(22)

(13)

i.e., In R c~ (x2/t).

A. Berger / Colloidal silver particles in glass surfaces

4. Discussion

As an example, the experimental data of the

samples ion exchanged in 2 wt% AgNO3-98 wt%
NaNO 3 will be compared with the theoretical
description. The results for the samples exchanged in the A g N O 3 - K N O 3 liquid are similar.
For the theoretical discussion, a diffusion coefficient was used, which was extrapolated from the
temperature range 300-480C by an Arrhenius
function ( D O= 5.6 x 10 +3 ixm2/s; E , = 1.1 x
10-19 j) [7]). According to the weight change, the
total amount of silver in the glass assumed to be
in an infinitesimal thin surface layer is M = 1.85
~m 2.
The model (1)-(4) with solution (14)-(16) and
(18) gave the following qualitative results:
(i) high concentration of ionic silver (near surface, short times) results in many stable nuclei
and small size particles;
(ii) small concentration of ionic silver (great
depths, long times) results in few stable nuclei
and large size particles;
(iii) the volume concentration of particles is
constant within the silver containing layer;
(iv) for a given annealing time the particle size
increase with the depth.

a
50

2?

35

2.5

, h/650"C

45

le0nleO0.c
100~/ss0.c
A=5 e . 1 0 - " ; . - i 8

40

A~5 6.10-11; n = l 8 5
--

35

A = I S.10-'; .-I 0.8

^=1 S-,O-';n=

eo

3O

20
o

15
z V ,

05~
0

10

~Z , 0 o . / e s - ~

1
2/t

2
5
in/~mZ/s

93

The logarithmic plot (fig. 5(a)) shows that for

long annealing times all values calculated from
experiments can be fitted to a linear function,
which is in agreement with eq. (22) and which
allows an estimate of n and A. The value of n,
the number of silver atoms building a stable nucleus, increases from 6.0 to 18.5 in the temperature range from 550 to 650C. The value of A
reduces from 1.5 x 10 -5 i~m(3-n)/2 to 5.6 x 10 -11
ixm(3-")/2, i.e., the ratio k2/k 1 decreases (eq.
(19)). Since, however, the concentration of the
particles, N, depends on the product k,i"r", the
number of particles per volume decreases with
increasing temperature. With r o = rFe2+, then c(t
--->oo) = 1.2 x 10 -4, which is in a good agreement
with the experimental data (see fig. 4).
A plot of particle radius versus penetration
depth (fig. 5(b)) shows that there is agreement
between experimental and theoretical values for
long annealing times.
In the logarithmic plot, the data for shorter
annealing times also fits a linear function. These
have, however, a different slope. As fig. 5(b)
shows, experimental and theoretical values differ
in the depth dependence of the particle size. This
difference is, however, not a general failure of
the model, but is to be expected, because the
assumption x i << x a is not valid for small annealing times. To solve the diffusion eq. (5), a constant diffusion coefficient was assumed. For long
annealing times resulting in small silver ion concentrations, this assumption is good. For small
annealing times, i.e., larger silver ion concentrations, however, the concentration dependence of
the diffusion coefficient has to be taken into
account. To give a correct description for short
annealing times, it seems to be necessary to carry
out numerical calculations for the diffusion of the
silver ions taking into account the measured silver
depth profile after ion exchange and a concentration dependent diffusion coefficient.

5
0
300

600
900
X in/zm

1200

Fig. 5. Comparison of experimental data (ion exchange in 2

wt% AgNO3-98 wt% NaNO 3 at 400C for 2 h, different
annealing processes) with the theoretical description according eqs. (21) and (22), respectively. (A is in Ixm(3-n)/2).

5. Conclusions

The proposed model of particle formation on

the basis of a homogeneous nucleation in which
ion exchange, silver ion diffusion, reduction and

94

A. Berger / Colloidal silver particles in glass surfaces

agglomeration are the parameters gives a qualitative explanation of the dependence of particle
size and concentration on the penetration depth.
For long annealing times at temperatures in the
range 550-650C, even a quantitative description
of the dependence is possible. The number of
silver atoms building a stable nucleus was estimated to be in the order of 6-18 in the temperature range from 550 to 650C.
When this work was done, the author was with
the Martin-Luther-Universit~it, Halle-Wittenberg,
Fachbereich Physik, Germany. He thanks Dr.
Herbert Hofmeister from the Max-Planck-Institut
fiir Mikrostrukturphysik Halle for carrying out
the electron microscopical investigations.

References
[1] W.A. Weyl, Coloured Glasses (Dawsons, London, 1959).
[2] G.H. Frischat, Ionic Diffusion In Oxide Glasses (TransTech, Aedermannsdorf, 1975).
[3] H. Garfinkel, in: Membranes, ed. G. Eisenman, Vol. 1
(Dekker, New York, 1972) p. 179.
[4] C.R. Bamford, Colour Generation and Control in Glass
(Elsevier, Amsterdam, 1977).
[5] D. Ganka, dissertation (A), Martin-Luther Universitiit
Halle, in preparation.
[6] K.-J. Berg, A. Berger and H. Hofmeister, Z. Phys. D20
(1991) 309.
[7] M. Petzold, unpublished work.
[8] S. Thiel, Praktikumsbeleg (Torgau, 1989).
[9] K.H. Brauer and F. Frfhlich, Exp. Techn. Phys. 6 (1958)
216.
[10] M. Dubiel, H. Hofmeister, K.-J. Berg and A. Berger,
Ver6ffentlichungen zur 12. Tagung Elektronen
mikroskopie, Dresden, 1988, p. 424.
[11] S.C. Jain, A.E. Hughes, J. Mater. Sci. 13 (1978) 1611.
[12] J. Crank, The Mathematics of Diffusion (Oxford University Press, 1956).
[13] A. Berger, thesis, Martin-Luther Universitiit Halle (1988).