Effects of Particle Concentration on Chemical Mechanical ...

<strong>Effects</strong> <strong>of</strong> <strong>Particle</strong> <strong>Concentration</strong> on Chemical Mechanical
Planarization
Kevin Cooper, a, * ,z Jennifer Cooper, b Johannes Groschopf, b John Flake, a, **
Yuri Solomentsev, a,c and Janos Farkas a, **
a Motorola Digital DNA Laboratories, b AMD-Motorola Technology Alliance, Austin, Texas 78721, USA
<strong>Effects</strong> <strong>of</strong> particle concentration on removal rates in chemical mechanical planarization CMP were investigated. Experimental
data shows the removal rate scales with the cubic root <strong>of</strong> weight percent solids and is proportional to mean separation distance
between particles in the slurry. Results show similar removal rate scaling with solid concentration for both silicon dioxide and
copper films. This work also identifies a critical solid concentration needed to initiate removal for oxide or copper surfaces. The
effects <strong>of</strong> particles on copper surface oxidation and roughness are further characterized by voltammetry and atomic force microscopy.
Results show copper films become progressively smoother with additional solids in the slurry.
© 2002 The Electrochemical Society. DOI: 10.1149/1.1517772 All rights reserved.
Manuscript submitted May 24, 2002; revised manuscript received August 21, 2002. Available electronically October 4, 2002.
Chemical mechanical planarization CMP is widely used in
various industrial applications ranging from optical lens polishing to
storage media and semiconductor manufacturing. An important
characteristic <strong>of</strong> any CMP process is the overall removal polishing
rate <strong>of</strong> the surface. Removal rate R is typically described in terms <strong>of</strong>
Preston’s law. 1,2 This is an empirical relationship that postulates proportionality
<strong>of</strong> the removal rate to applied pressure P and relative
velocity v <strong>of</strong> the surfaces in contact. That is
R APv 1
where A is the Preston coefficient. The coefficient is a function <strong>of</strong>
several parameters including reaction kinetics, solids concentration,
and surface properties. 2,3 Although many systems are considered to
follow Prestonion behavior, deviations are commonly observed and
reported in literature. 4-6 Various adaptations <strong>of</strong> Preston’s law have
been proposed and a generalized relationship may be written as
R AP v 2
where and are experimental constants. Quantities and are
considered as adjustable parameters and are used to account for
nonlinear variation <strong>of</strong> rate with kinetics or other factors. 7,8
In this work, the effect <strong>of</strong> solids concentration on the Preston
coefficient was studied using a linear model. As stated in the removal
mechanism proposed by Kaufman 9 and others, 10,11 accelerated
removal rates are due to mechanical abrasion <strong>of</strong> the oxidized
metal or hydrolyzed oxide surface by slurry particles. Previous reports
have also shown increasing concentrations <strong>of</strong> abrasive particles
in a CMP solution lead to increased removal rates. 11-13 The
nature <strong>of</strong> the relationship between increases in removal rate and
solids concentration is considered in further detail here. In dilute
slurries, similar to those commonly used in CMP, the mean particle
separation distance l is much larger than the particle size a. In these
conditions, it is reasonable to expect A to be proportional to the
frequency <strong>of</strong> collision between the particles and the polished surface
assuming fixed chemical composition. A general relationship for
the Preston coefficient can be written as follows
A D Bfc 3
where B is a constant, f is the frequency <strong>of</strong> collisions, c is solids
concentration, and D is a constant independent <strong>of</strong> concentration. The
* Electrochemical Society Student Member.
** Electrochemical Society Active Member.
c Current address: Texas Instruments Silicon Technology Development, Dallas,
Texas 75243.
z E-mail: kevin.cooper@motorola.com
Electrochemical and Solid-State Letters, 5 12 G109-G112 2002
0013-4651/2002/512/G109/4/$7.00 © The Electrochemical Society, Inc.
frequency <strong>of</strong> collisions f is inversely proportional to the mean time
between collisions . is proportional to the mean particle separation
l and Eq. 3 can be rewritten as
A D F
l
F const 4
The volume <strong>of</strong> an individual particle, VP , and the total volume
occupied per particle, can be Vˆ
P written as
VP 4
3 a3 5
Vˆ
P VT
NP 4
3 l3 6
where a is the particle radius, V T is the total liquid volume, and N P
is the total number <strong>of</strong> particles. Furthermore, weight percent can be
defined as
wt % W P
W T
where W P is the weight <strong>of</strong> the particles
W P PV PN P
and W T is the total weight <strong>of</strong> the solution
W T WV T
where P and W are the density <strong>of</strong> the particle and the density <strong>of</strong>
solution, respectively. Substituting Eq. 8 and 9 into Eq. 7 yields
wt % WP
WT VP PNP VT W
Further substitution <strong>of</strong> Eq. 5 and 6 into Eq. 10 yields
wt % VP PNP Vˆ
PWN P
4a3 P 3 W 1
Vˆ
P
wt % B
Vˆ P
G109
7
8
9
10
11
12

G110
Figure 1. Removal rate as a function <strong>of</strong> P.
where B is
Rearranging Eq. 12 yields
4a3 P
3 W
Vˆ P B
wt %
Substitution <strong>of</strong> Eq. 6 into Eq. 13 yields
l
1
wt % 1/3
13
14
Because l is proportional to w1/3 where w is the solids weight
percent in the slurry Eq. 4 is reduced to the following relationship
A D Fw1/3 15
Equation 15 is valid for Prestonian systems with constant chemical
composition, pressure, and velocity.
Experimental
Silicon dioxide films 450 nm thick were prepared via chemical
vapor deposition CVD tetraethylorthosilicate TEOS on 200 mm
silicon substrates. Copper films 1.2 m thick were prepared by
electrochemical deposition from a commercial electrolyte on physical
vapor deposition PVD Cu seed 100 nm. Oxide wafers were
polished using an alkaline slurry with silica particles with a mean
size <strong>of</strong> 120 nm. Copper wafers were polished using proprietary
acidic slurries comprised <strong>of</strong> silica-based particles with an
approximate size <strong>of</strong> 40 nm. Prior to CMP, copper wafers were annealed
in a furnace for sufficient time and temperature for complete
recrystallization.
Oxide film thickness was measured before and after the CMP
experiments using a commercial spectrophotometer. Copper film
thickness was measured using a calibrated four-point probe. Data
was collected for 30 points per wafer and removal rates represent
the average values. The wafers were polished on a rotational tool for
60 s in each experiment. Experiments with varying solids concentration
were done at constant pressure and rotational velocity.
Results and Discussion
Experiments were first performed to verify Prestonian behavior
<strong>of</strong> the system. Figure 1 shows linear scaling <strong>of</strong> Cu removal rate with
the Pv product; similar behavior was observed for the oxide films.
Electrochemical and Solid-State Letters, 5 12 G109-G112 2002
Figure 2. Potentiodynamic polarization scans <strong>of</strong> the copper rotating disk
electrode polished with and without abrasion.
The rate data was normalized by maximum removal rate and Pv
product was normalized by the maximum as well. For both oxide
and copper films the experimental data was in good agreement with
the predictions in Eq. 1.
In comparing the removal mechanisms <strong>of</strong> oxide and copper
films, copper must undergo an additional oxidation step prior to
removal. As described in the work by Kaufman et al., 9 the steadystate
removal involves dynamic formation and mechanical removal
<strong>of</strong> oxides. The oxidation and removal <strong>of</strong> copper in these experiments
was examined using voltammetry under abrasive and nonabrasive
conditions. A Pt rotating disk electrode RDE electroplated with
approximately 10 m <strong>of</strong> Cu was rotated at 300 rpm and pressed
against a polyurethane pad. Down force was approximately 50 kPa
as the weight <strong>of</strong> the RDE apparatus rested entirely on the disk.
Figure 2 shows steady-state potentiodynamic scans <strong>of</strong> the electroplated
Cu electrode in an acidic solution with an oxidizing agent.
Scans were taken during polishing with 1 wt% and without
abrasives. As seen in Fig. 2, the exchange current density and anodic
slope increase when particles are present. In the absence <strong>of</strong> particles,
the anodic portion <strong>of</strong> the curve shows lower currents indicating the
Figure 3. Removal rate <strong>of</strong> SiO 2 as a function <strong>of</strong> solid content.

Figure 4. Removal rate <strong>of</strong> copper as a function <strong>of</strong> solid content.
presence <strong>of</strong> a passivating surface oxide. When particles are added to
the system, the rest potential is shifted approximately 50 mV cathodically
and the anodic current is significantly greater with overpotential.
This behavior indicates the mechanism <strong>of</strong> chemical oxidation
followed by mechanical removal is experimentally apparent for
the dilute slurries 1 wt% considered here.
To examine the relationship between solids concentration and
removal rate Eq. 15, the polishing velocity and pressure were fixed
and the weight percent <strong>of</strong> solids in the slurries was varied. Figures 3
and 4 show the normalized removal rate as a function <strong>of</strong> the normalized
weight percent <strong>of</strong> solids for SiO 2 and copper substrates. The
y axis normalized removal rate was made dimensionless by the
maximum removal rate for each system. The x axis normalized wt
% was made dimensionless by the maximum weight percent used
in the each set <strong>of</strong> experiments. The error bars in Fig. 3 and 4 represent
the standard deviations in removal rates across all measured
points.
In Fig. 5 the removal rate data from Fig. 3 and 4 is plotted as
functions <strong>of</strong> wt % 1/3 . In Fig. 5, both x wt % 1/3 and y axis removal
rate were normalized by the maximum solid concentration and
Figure 5. Removal rate <strong>of</strong> copper and SiO 2 as a function <strong>of</strong> w 1/3 .
Electrochemical and Solid-State Letters, 5 12 G109-G112 2002 G111
Figure 6. Removal rate <strong>of</strong> copper and SiO 2 as a function <strong>of</strong> solid concentration
for a benzenetriazole-based chemistry solid line and a triazole-based
chemistry dashed line.
maximum removal rate in the copper system, respectively. As seen
in each case Cu or SiO 2) the removal rate scales linearly with
wt % 1/3 after a critical onset concentration is reached. The scaling in
this regime demonstrates the removal is directly proportional to the
frequency <strong>of</strong> collisions. At high solid concentrations once l is no
longer R), the scaling behavior no longer follows a collision frequency
and removal rate is independent <strong>of</strong> solid concentration. 13,14
The differences in critical onset concentrations reveal differences
in removal mechanisms and kinetics between the two systems.
While Cu may be removed efficiently at a relatively lower solids
concentration, oxide films require more energy thermal or mechanical
to reach the linear removal rate regime. Oxides <strong>of</strong> copper appear
to be more easily removed with relatively lower solids concentration
compared with silicon dioxide films. The nature <strong>of</strong> the relatively
lower critical concentration may be attributed to mechanical or
Figure 7. Three-dimensional AFM images <strong>of</strong> polished and unpolished copper
films.

G112
Figure 8. RMS <strong>of</strong> polished copper surfaces as a function <strong>of</strong> abrasive concentration.
chemical effects. Mechanically, the oxides <strong>of</strong> copper are known to
show relatively poor adhesion to the elemental copper. This is in
contrast to metals such as aluminum or tungsten where the oxides
show much better adhesion to the elemental metal. Further work
with other metals is needed to separate effects <strong>of</strong> adhesion and kinetics
on the critical solids concentration.
To better understand the impact <strong>of</strong> chemistry on removal rate,
two copper slurry chemistries were compared. The components and
concentrations <strong>of</strong> the slurries were equivalent in all respects with the
exception <strong>of</strong> the corrosion inhibitor, triazole TA or benzenetriazole
BTA. Figure 6 shows the normalized removal rate as a function <strong>of</strong>
the normalized weight percent for both chemistries. When particles
are absent the slurry with BTA shows significantly lower removal
rates 3.5 times than the slurry with triazole. As seen in Fig. 6, the
removal rates collapse to an equivalent relationship wt % 1/3 with
the introduction <strong>of</strong> solids. These results demonstrate the removal
mechanism in the regime remains linearly dependant on frequency
<strong>of</strong> collisions despite differences in slurry chemistry. No differences
in critical concentration were observed indicating mechanical effects
may dominate the critical onset concentration in these slurries.
In addition to considering the effect <strong>of</strong> solids on removal rates, it
is important to consider the effects <strong>of</strong> solid concentration on surface
morphology. Surface roughness should be minimized to promote
easy integration <strong>of</strong> subsequent metal levels and to maintain the integrity
<strong>of</strong> the Cu-dielectric interface. Figure 7 shows threedimensional
atomic force microscopy AFM images <strong>of</strong> the polished
and unpolished copper surfaces. As seen in Fig. 7A and B, the surface
obtained with an abrasive-free polish Fig. 7B shows grain
boundaries are attacked resembling the rough unpolished surface
Fig. 7A. With introduction <strong>of</strong> abrasive particles to the slurry sur-
Electrochemical and Solid-State Letters, 5 12 G109-G112 2002
face roughness was reduced by 3.5 times Fig. 7B, C, and D. Continuous
reduction in surface roughness was observed with increasing
particle concentration. Figure 8 plots the root mean square rms
roughness values as a function <strong>of</strong> abrasive concentration.
Conclusions
CMP removal rates viz. Preston coefficient were found to scale
linearly with wt % 1/3 for dielectric and metal substrates. Experimental
results show good agreement with theory in dilute solutions,
where the cubic root <strong>of</strong> weight percent is roughly proportional to
mean separation distance between abrasive particles in the slurry.
As seen in Fig. 5, the critical particle concentration before significant
removal rate is achieved was approximately sixfold greater
for oxide (SiO2) than for copper surfaces. The difference in critical
onset concentrations reveals differences in removal mechanisms and
activation between the two systems. In addition to relative differences
in removal rates <strong>of</strong> oxide and copper films, solid concentrations
have a tremendous impact on overall planarity and interconnect
performance on patterned wafers where dielectric and metal
films are exposed simultaneously. Considering that both dishing and
erosion are functions <strong>of</strong> bulk removal rates <strong>of</strong> both dielectric and
metal, the planarity may be minimized by tailoring the chemistry to
provide either selective or nonselective removal pr<strong>of</strong>iles.
Acknowledgments
The authors thank the Motorola APRDL pilot line and the
PALCOE lab for supporting this work. The authors are especially
grateful to Susan Backer and Charles Stager for their significant
laboratory analysis and technical discussion, respectively.
Motorola assisted in meeting the publication costs <strong>of</strong> this article.
References
1. F. Preston, J. Soc. Glass Technol., 11, 247 1927.
2. L. M. Cook, J. Non-Cryst. Solids, 120, 152 1990.
3. Y. Gotkis and R. Kistler, Abstract 496, The Electrochemical Society Meeting Abstracts,
Vol. 2000-2, Phoenix, AZ, Oct 22-27, 2000.
4. W.-T. Tseng and Y.-L. Wang, J. Electrochem. Soc., 144, L15 1997.
5. F. Zhang and A. Busnaian, Electrochem. Solid-State Lett., 1, 184 1998.
6. B. Zhao and F. Shi, Electrochem. Solid-State Lett., 2, 145 1999.
7. E. Paul, J. Electrochem. Soc., 148, G359 2001.
8. J. M. Steigerwald, S. P. Mararka, and R. J. Gutmann, Chemical Mechanical Planarization
<strong>of</strong> Microelectronic Materials, p. 153, John Wiley & Sons, New York
1997.
9. F. B. Kaufman, D. B. Thompson, R. E. Brodie, M. A. Jaso, W. L. Guthrie, D. J.
Pearson, and M. B. Small, J. Electrochem. Soc., 138, 3460 1991.
10. S. M. Lee, U. Mahajan, Z. Chen, and R. V. Singh, in Chemical Mechanical Planarization
in IC Device Manufacturing III, R. L. Opila, I. Ali, Y. A. Arimoto, Y.
Homma, C. Reidsema-Simpson, and K. B. Sundaram, Editors, PV 99-37, p. 187,
The Electrochemical Society Proceedings Series, Pennington, NJ 2000.
11. R. J. Gutmann, D. J. Duquette, P. S. Dutta, and W. N. Gill, in Proceedings <strong>of</strong> the
6th International CMP-MIC Conference, p.812001.
12. U. Mahajan, M. Bielmann, R. K. Singh, S. V. Babu, S. Danyluk, M. Krishnan, and
M. Tsujimura, Mater. Res. Soc. Symp. Proc., 566, 272000.
13. U. Mahajan, S.-M. Lee, and R. Singh, in Chemical Mechanical Planarization in IC
Device Manufacturing III, R. L. Opila, I. Ali, Y. A. Arimoto, Y. Homma, C.
Reidsema-Simpson, and K. B. Sundaram, Editors, PV 99-37, p. 396, The Electrochemical
Society Proceedings Series, Pennington, NJ 2000.
14. R. Singh, S.-M. Lee, and W. Choi, Paper I 1.6 presented at the 2002 MRS Conference
on Chemical Mechanical Planarization, San Francisco, CA, April 1–5,
2002.