Vous êtes sur la page 1sur 24

Tuesday , March 02, 2004

Application of composite hardness models to copper thin film hardness measurements

Application of composite hardness models to copper thin film

py

hardness measurement
D. Beegan1, 2, M.T. Laugier1, 2*
1

Department of Physics, University of Limerick, Limerick, Ireland

iew

Co

Materials and Surface Science Institute, University of Limerick, Limerick, Ireland

*Address for corresponding author:

Re
v

Dr. M.T. Laugier,

Department of Physics,

University of Limerick,
Limerick,

IRELAND.

Tel: +353-61-202257

Fax: +353-61-202423

e-mail: michael.laugier@ul.ie

1
Elsevier

1 of 24

Tuesday , March 02, 2004

Abstract
The hardness of copper thin films with thicknesses ranging from 25-500 nm on oxidised
silicon substrates is investigated. The Oliver and Pharr method is used to analyse the loaddisplacement curves obtained from nanoindentation. Composite hardness models are applied
suitability and limitations of these models are discussed.

py

to the range of hardness values in order to distinguish the film and substrate hardnesses. The

Co

Keywords: copper; nanoindentation; composite hardness models


1. Introduction

During hardness determination of thin films by indentation methods, the influence of the
substrate must be considered. The measured hardness is a complex value depending on the
relative indentation depth and the mechanical properties of both the film and the substrate.
Above a certain critical penetration depth the so-called composite hardness, which includes

iew

a component of the substrate hardness, will be measured.


The simplest and most widely applied solution to the problem of substrate effects is the 10%
rule [1, 2]. This rule proposes that the film properties can be measured for indentation depths
less than 10% of the total film thickness. This rule of thumb has several deficiencies. The
rule is too restrictive for soft coatings on hard substrates, while it may be not be restrictive

Re
v

enough for hard coatings on soft substrates. In addition, for nanometre scale films, it can be
experimentally difficult to perform and analyse indents that satisfy this requirement.
When the ranges of measured film hardness values are plotted against displacement for a
particular film/substrate system, three different regimes become evident. Such a graph is
shown in figure 1 for a soft film on a hard substrate. It is found that for shallow penetration
depths (region I) the response is that of the film only. As the depth increases (region II) the
hardness increases gradually and is associated with the mixed response of the coating and the
substrate. At high indentation depth in region III the response is mainly dominated by the
substrate.

2
Elsevier

2 of 24

Tuesday , March 02, 2004

1.1

Composite hardness models

To obtain the hardness of the coating alone from the experimental measurements, several
models exist [3-9]. In particular those models advanced by Jnsson and Hogmark [4], Burnett
and Rickerby [5, 6], Chicot and Lesage [7], Korsunsky et al.[8], and Puchi-Cabrera [9] will be

py

examined. These models operate on a number of different principles.


Bckle [3] first proposed a simple law of mixtures model, where the composite hardness Hc is
expressed as:
Hc = a Hf + b Hs

(1)

Co

with a + b = 1; a varies from 1, when the film-only hardness is measured, to 0, when the
thickness of the film is negligible compared with the indentation depth. For a film thickness t
and an indentation depth h, the parameter a is related to the ratio h/t. Various groups have
since developed this original model.

Jnsson and Hogmark [4] used a simple geometrical approach to separate the substrate and

iew

coating contributions to the measured hardness. They used a simple area law of mixtures:

Hc =

Af

As
Hs
A

Hf +

(2)

where Af is the area on which the mean pressure Hf acts, As is the area on which the mean
pressure Hs acts and the total area is then A = Af + As.

Re
v

From geometrical consideration, the area ratios are given by the following relations:
Af

t
= 2C
A
h

As
=1
A

t
h

(3)

Af
A

(4)

where t is the film thickness, h is the indentation depth and C is a constant dependent on the
indenter geometry.

A volume law of mixtures approach was originally suggested by Sargent [10] and
subsequently extended variously by Bull, Burnett, and Rickerby [5, 6]. In this model, the zone
of plastic deformation beneath the indenter is assumed to be hemispherical.

3
Elsevier

3 of 24

Tuesday , March 02, 2004

Burnett and Rickerby proposed a mixing law similar to Jnssons relation, considering the
volumes of the plastic zones, Vf and Vs respectively, in the film and in the substrate:

Hc =

Vf
V

Hf +

Vs
Hs
V

(5)

with V = Vf + Vs.

the indentations. Using Lawns equation [11]:


d Ei
bi =
2 Hi

py

Generally it is assumed that the size of the hemispherical plastic zone is related to the size of
1 2

cot 1

where bi is the plastic zone radius, d is the indentation diagonal,

is the indenter semi-angle,

Co

E the Youngs modulus and H the hardness.

(6)

In order to take into account the interaction between the coating and substrate, Burnett
introduced a correction factor termed the interface parameter, , into the relation:

Vf
V

Hf +

iew

Hc =
Hc =

Vf
V

Vs
V

Hf +

H s for Hs < Hf

(7)

Vs
H s for Hs > Hf
V

(8)

In order to avoid the introduction of such a parameter Chicot and Lesage [7] developed a
model that takes into account the interaction between the substrate and the coating. This
model is also based on a volume law of mixtures for the plastic zone. The model supposes

Re
v

continuity between the coating and the substrate. The volume of the plastic zone of the
coating is approximated by a cylinder and that of the substrate by a hemispherical zone. The
respective radii are calculated using Lawns equation (equation 6 above) [11]. The final
model gives:

3t
Hc = HS +
2d

Hf
Ef

HS
+
ES

tan

(H

H S ) (9)

where is the indenter semi-angle.

4
Elsevier

4 of 24

Tuesday , March 02, 2004

The model developed by Korsunksy et al. [8] employs dimensionless parameters. This model
is applicable to either plasticity or fracture-dominated behaviour, with all scales measured
relative to the coating only thickness. This model is based on energy expenditure during an
indentation and the final equation is given by:

where

Hf

HS

1+ k

(9)

py

Hc = HS +

is the relative indentation depth and k is a constant related to the film thickness.

The Puchi-Cabrera model [9] is another geometrical model, which again gives the composite
hardness from a simple law of mixtures in terms of the volume fractions of coatings and

Co

substrate and their hardnesses, Hf and Hs as suggested by Bckle [3]:


H = HS X S + H f X f

(10)

The major difference between this model and the previous law of mixture models is that it
assumes that the contribution of the substrate to the composite hardness starts to be effective
before the indenter crosses the film thickness. The resultant model is given by:

H = H S + (H f

n
R

(11)

is the relative indentation depth and k and n represent material parameters that

iew

where

H S )exp

Re
v

characterise the change in hardness as the indenter passes from the film to the substrate.

5
Elsevier

5 of 24

Tuesday , March 02, 2004

2. Experimental details

A thin layer of SiO2 was then thermally grown on P-type silicon wafers (100) (Wacker,

2-30 cm) substrates. SiO2 film thickness was measured by ellipsometry and found to be

py

100 nm.
Copper films were deposited in a Leybold Lab 500 RF magnetron sputtering system (RF
13.56 MHz). The chamber base pressure was below 2.3 x10-5 mbar and the process pressure
was 4 x 10-3 mbar. A 99.9999% pure copper target was used for film deposition with an
applied RF power of 40 W. The copper deposition rate was 9.6 nm/min and film thicknesses

Co

were varied by varying deposition time. In this paper we look at copper films of thicknesses
ranging from 25 to 500 nm. An in-situ quartz crystal oscillator was used to monitor the film
thickness. Film thickness was measured subsequently by measuring a step height with a Zygo
white light interferometer.

The mechanical properties of the films were characterized using a Nano Hardness Tester

iew

(NHT) developed by CSM Instruments, Switzerland. Indents were performed over a range of
loads from 0.5 to 100 mN. We used a constant strain rate method [12] by setting 1/P(dP/dt) to
be 2 min-1 and a pause of 10 s was set at maximum load. A minimum of five indents was
performed per load. A Berkovich diamond indenter was used for all indents and the Oliver

Re
v

and Pharr method [13] was used to determine the hardness, H.

6
Elsevier

6 of 24

Tuesday , March 02, 2004

3. Results and discussion


3.1. Hardness behaviour

The measured hardness values for the range of films are shown in figure 2. The three distinct
stages discussed earlier are clearly visible. Firstly at low normalised depths we see what

py

appears to be the film-only hardness. As the depths increase the hardness increases until at
the higher normalised depths we reach the substrate hardness.

While it appears that we can estimate the film-only hardness directly from this graph, it is

Co

also clear that this is only the case due to the presence of the 400 nm and 500 nm films. If we
plotted the thinner films individually no clear value for the film hardness would be evident.
Similarly for the thicker films the substrate-only hardness would not be known directly
unless very high loads were applied.

The hardness data for these films have been fit using the five composite hardness models. The
following nomenclature is used in the fitting of the composite hardness models and are used

iew

when discussing these models:

1) J-H: Jonsson and Hogmark model


2) B-R: Burnett and Rickerby model
3) C-L: Chicot and Lesage model
4) K: Korsunksy model

Re
v

5) P-C: Puchi-Cabrera model

Figure 3 shows the fit of the data using the J-H model. The figure clearly shows that this
model doesnt fit the experimental data. At high normalised depths the model fit is too low; as
the depths decrease the fit does pass through some the data points. The model breaks down at
lower depths however and is seen to increase rapidly. This has previously been observed [14]
and the breakdown was attributed to the polynomial form that is prescribed for the expression
of the film fraction under the indenter.
The values obtained by fitting the data with the J-H are given in Table 1 with the results from

the other models given in Tables 2-5. The C values given in these Tables relate either to a
7
Elsevier

7 of 24

Tuesday , March 02, 2004

constant as given in the model (e.g. C in the J-H model, or k in the K and P-C models) or a
constant related to the fitting of hc/t as opposed to d/t where d relates to the indentation
diagonal of a Vickers indent. A value for n the power exponent in the P-C model is also given
in Table 5 .
The next model employed in the analysis was that of Burnett and Rickerby (B-R). For this
is 75 and the values of Es and Ef are taken to be 107 and 130 GPa,

py

model the value of

respectively. It can be clearly seen that this model does not fit the data to any extent (fig. 4).
The values from this fit are given in Table 2 and we can see from the very large percentage

Co

errors that this model is not suitable.

Following this the C-L model is fit (fig. 5) and we can see that it gives a curve very similar to
the B-R model. In fact these two models are seen to practically overlap and it is mostly
impossible to distinguish between the two. The errors associated with the C-L fit are also very
large. It is found that while the fits are the same for these models the film hardness values

iew

differ, with those predicted by the C-L model greater than those of the B-R model.
The K model is next to be examined. This model is seen to give a very good fit to the data
(fig. 6). The curve follows the shape of the experimental data at all points and levels out at
low normalised depths towards a value representing Hf and similarly at high normalised
depths for Hs. The errors are quite low for the three fitted values.
Finally the most recently developed model of Puchi-Cabrera is analysed. This model also

Re
v

gives a very good fit to the experimental data (fig. 7). It has nearly the same shape as the K

model although it doesnt level out as much at low normalised depths and predicts slightly
lower values of Hf. There is little difference between the Hs values predicted by these two

models. The percentage errors are also quite low.


Figure 8 shows the fit of all the models to the data. Thus it may be seen that both the K and P-

C models fit the data with the best results. There is little to distinguish the two models, with
the only difference being a separation at low normalised depths resulting in different values of
Hf. It is hard to say if either of the models is the better but the K model does give slightly
lower percentage errors.

8
Elsevier

8 of 24

Tuesday , March 02, 2004

If all the films are fit to the models individually there is a difference in the results obtained.
As an example the K model is applied to each film thickness and the Hf and Hs values for each
film are obtained. In fig. 9 the fits of three of the films are shown (i.e. 25, 200 and 500 nm)
and it can be seen that the curves are different for each set of data. Table 6 shows the Hf and
Hs values predicted by fitting the K model to each film. Not all of the data for each individual
sample covers a large enough range to adequately describe the change in behaviour from

py

near-substrate to film only. Hence the Hs data in Table 9 is not accurate or obtainable in all
cases, i.e. it is accurate for the thinner films in the range where sufficient data above hc/t = 10
is available. For similar reasons, there are insufficient data points below hc/t = 1 to be sure
that the Hf data presented is accurate. Thus at low film thickness the film hardness is

Co

overestimated whereas for very thick films the value of substrate hardness may be
underestimated. The overall effectiveness of the values predicted by the fit is dependent on
the range hc/t values over which the composite hardness is measured.
4. Conclusion:

The hardness of copper films deposited on oxidised silicon substrates with film thicknesses

iew

ranging from 25 to 500 nm was investigated. The Oliver and Pharr method was used to
analyse the nanoindentation data. A range of hardness data typical for a soft film on a hard
substrate was obtained. This data had three distinct stages ranging from film-only to
substrate-only hardness.

Five composite hardness models (i.e those of Jnnson and Hogmark, Burnett and Rickerby,

Re
v

Chicot and Lesage, Korsunksy et. al. and Puchi-Cabrera) were applied to the experimental
data in an attempt to determine the film and substrate hardnesses separately. It was found that
both the Korsunksy and Puchi-Cabrera models gave very good fits to the data and predict film
and substrate hardnesses that would be expected for this system.
By fitting the data for individual films with the Korsunksy model we see that the effectiveness
of this model depends on the range of normalised depths that is represented by the data. If the
range of data is in the lower normalised depths a reasonable value of film hardness may be
found but not substrate hardness. Similarly for high normalised depths a reasonable value of
substrate hardness may be found but not film hardness.

9
Elsevier

9 of 24

Tuesday , March 02, 2004

Both the Korsunksy and the Puchi-Cabrera composite hardness models will return good
values of film and substrate hardness provided a wide range of normalised depths are

Re
v

iew

Co

py

represented in the experimental data.

10
Elsevier

10 of 24

Tuesday , March 02, 2004

References

1. D.E. Kramer A.A. Volinsky, N.R. Moody, W.W. Gerberich, J. Mater. Res. Vol. 16. no.
11. Nov 2001, 3150
2. A.C. Fischer-Cripps, Vacuum, 58 (2000) 569
3. H. Bckle, in J.W. Westbrook and H. Conrad (eds.), The Science of Hardness Testing and

py

its Research Applications, American Society for Metals, Metals Park, OH, 1973, p.453
4. B. Jnsson and S. Hogmark, Thin Solid Films, 114 (1984) 257

5. P.J. Burnett and D.S. Rickerby, Thin Solid Films, 148 (1987) 41

6. S.J. Bull and D.S. Rickerby, Surf. Coat. Technol., 42 (1990) 149

Co

7. D. Chicot and J. Lesage, Thin Solid Films, 254 (1995) 123

8. A.M. Korsunsky, M.R. McGurk, S.J. Bull, T.F. Page, Surf. Coat. Technol., 99 (1998) 171
9. E. S. Puchi-Cabrera, J.A. Berrios, D. G. Teer, Surf. Coat. Technol. 157, (2002) 185
10. P.M. Sargent, PhD Thesis, University of Cambridge, 1979

11. B.R. Lawn, A.G. Evans, D.B. Marshall, J. Am. Ceram. Soc., 63 (1980) 574
12. B.N. Lucas, W.C. Oliver, G.M. Pharr, J-L. Loubet, Mater. Res. Soc. Symp. Proc., 436
(1997) 233

iew

13. W.C. Oliver, G.M Pharr, J. Mater. Res.7 (1992) 1564

Re
v

14. E.S. Puchi-Cabrera, Surf. Coat. Technol., 160 (2002)177

11
Elsevier

11 of 24

Tuesday , March 02, 2004

List of Figures

Fig. 1 Variation in composite hardness with normalised depth for a soft film on hard
substrate
Fig. 2 Plot of hardness vs normalised depth for all film thicknesses
Fig. 3 Fit of Jonnson and Hogmark (J-H) model to experimental data

py

Fig. 4 Fit of Burnett and Rickerby (B-R) model to experimental data


Fig. 5 Fit of Chicot and Lesage (C-L) model to experimental data
Fig. 6 Fit of Korsunksy model to experimental data

Fig. 7 Fit of Puchi-Cabrera (P-C) model to experimental data

Co

Fig. 8 Fit of all composite hardness models to experimental data

Re
v

iew

Fig. 9 Fit of Korsunksy model to experimental data for 25, 200 and 500 nm thick films

12
Elsevier

12 of 24

Tuesday , March 02, 2004

List of Tables

Table 1: Predicted values from the fitting of Jonnson and Hogmark model
Table 2: Predicted values from the fitting of Burnett and Rickerby model
Table 3: Predicted values from the fitting of Chicot and Lesage model
Table 4: Predicted values from the fitting of Korsunksy model

py

Table 5: Predicted values from the fitting of Puchi-Cabrera model

Table 6: Predicted values from the fitting of the Korsunksy composite hardness model to the

Re
v

iew

Co

NHT hardness data of the individual copper films

13
Elsevier

13 of 24

Hardness (GPa)

Tuesday , March 02, 2004

14
12

8
6

II

2
0

0.01

0.1

III

Co

py

10

10

100

Normalised depth

Re
v

iew

Fig. 1

14
Elsevier

14 of 24

Tuesday , March 02, 2004

Hardness
(GPa)
Hardness (GPa)

14
12
10

py

8
6

25 nm
50 nm
100 nm
200 nm
300 nm
400 nm
500 nm

Co

2
0
0.1

1
10
Normalised depth

100

Re
v

Fig. 2

iew

Normalised Depth

15
Elsevier

15 of 24

Tuesday , March 02, 2004

12
10
8

py

6
4

Hardness data
J-H Hardness
model data

J-H model

0
0.1

Co

Hardness (GPa)

Hardness (GPa)

14

10

100

Normalised depth

Normalised Depth

Re
v

iew

Fig. 3

16
Elsevier

16 of 24

Tuesday , March 02, 2004

12
10

py

8
6
4

Hardness data
B-R model
Hardness data

Co

Hardness
(GPa)
Hardness (GPa)

14

B-R model

0
0.1

10

100

Normalised depth

Normalised Depth

Re
v

iew

Fig. 4

17
Elsevier

17 of 24

Tuesday , March 02, 2004

12
10

py

8
6
4

Hardness data
C-L model
Hardness data

C-L model

0
0.1

Co

Hardness
(GPa)
Hardness (GPa)

14

10

100

Normalised depth
Normalised
Depth

Re
v

iew

Fig. 5

18
Elsevier

18 of 24

Tuesday , March 02, 2004

12
10

py

8
6
4

Hardness data
K model

Hardness data
Korsunksy model

0
0.1

Co

Hardness
(GPa)
Hardness (GPa)

14

10

100

Normalised depth

Normalised Depth

Re
v

iew

Fig. 6

19
Elsevier

19 of 24

Tuesday , March 02, 2004

12
10

py

8
6
4

Hardness data
P-C
model data
Hardness

Co

Hardness
(GPa)
Hardness (GPa)

14

P-C model

0
0.1

10

100

Normalised depth

Normalised Depth

Re
v

iew

Fig. 7

20
Elsevier

20 of 24

Tuesday , March 02, 2004

12
10

py

Hardness data
J-H model
B-R model
Hardness data
J-H model
C-L model
B-R model
K model C-L model
Korsunksy model
P-C model
P-C model

6
4
2
0
0.1

Co

Hardness
(GPa)
Hardness (GPa)

14

10

100

Normalised depth

Normalised Depth

Re
v

iew

Fig. 8

21
Elsevier

21 of 24

Tuesday , March 02, 2004

12
10

py

25 nm data
25 nm fit
200 nm 25
data
nm data
25
fit
200 nm200
fit nmnm
data
200 nm fit
500 nm data
500 nm data
500 nm fit500 nm fit

6
4
2
0
0.1

Co

Hardness
(GPa)
Hardness (GPa)

14

10

100

Normalised depth

Normalised Depth

Re
v

iew

Fig. 9

22
Elsevier

22 of 24

Tuesday , March 02, 2004

Model

J-H

% error

Hs

11.23

2.58

Hf

0.49

3.99

0.30

5.64

Model

B-R

% error

Hs

9.32

3.86

Hf

2.54

6.35x10

0.50

5.07x10

12
13

Model
Hs
Hf
C

Model

Re
v

Table 4

C-L

% error

9.32

3.75

13

3.04

1.07x10

0.34

3.46x10

iew

Table 3

Co

Table 2

py

Table 1

Kor

13

% error

Hs

12.64

1.51

Hf

2.76

2.35

0.19

9.09

Model

P-C

% error

Hs

12.64

1.82

Hf

1.71

4.79

0.32

13.27

1.07

9.58

Table 5

23
Elsevier

23 of 24

Tuesday , March 02, 2004

Film thickness (nm)

25

50

100

200

300

400

500

Hs

13.18

12.48

12.15

12.21

14.49

10.14

9.93

Hf

7.20

4.18

2.56

2.73

2.90

2.33

2.46

0.10

0.17

0.17

0.15

0.14

0.61

0.24

Re
v

iew

Co

py

Table 6

24
Elsevier

24 of 24

Vous aimerez peut-être aussi