Defect reduction in Cu chemical-mechanical polishing
The MIT Faculty has made this article openly available.
Please share
how this access benefits you. Your story matters.
Citation
Eusner, Thor, Nannaji Saka,and Jung-Hoon Chun. "Defect
Reduction in Cu Chemical-Mechanical Polishing." ISSM 2010
conference proceedings, the Eighteenth International Symposium
on Semiconductor Manufacturing, October 18-20, 2010, Hyatt
Regency Tokyo, Shinjuku, Tokyo. IEEE, 2010. © Copyright 2010 IEEE.
As Published
http://ieeexplore.ieee.org/stamp/stamp.jsp?
tp=&arnumber=5750240
Publisher
Institute of Electrical and Electronics Engineers
Version
Final published version
Citable link
http://hdl.handle.net/1721.1/78335
Terms of Use
Article is made available in accordance with the publisher's
policy and may be subject to US copyright law. Please refer to the
publisher's site for terms of use.
Defect Reduction in Cu Chemical-Mechanical Polishing
/PO-O-032 /
Thor Eusner, Nannaji Saka, Jung-Hoon ChunLaboratory for Manufacturing and Productivity Massachusetts Institute of Technology
Cambridge, MA, USA
teusner@mit. edu, nsaka@mit.edu, jchun@mit. edu
Abstract - The chemical-mechanical polishing (CMP) of Cu is a critical step in the manufacture of ultra-large-scale integrated (ULSl) semiconductor devices. During this process, undesirable scratches are formed on the surface being polished [I -3]. Recent research suggests that the "killer" scratches found on the Cu wafers are due to the soft pad asperities and not necessarily by the hard abrasives in the slurry [4,5]. Figure 1 shows examples of scratches on a Cu coating due to pad asperities. This paper presents the theory and experimental validation of scratching by soft pad asperities in Cu CMP. Based on the models and experimental results, practical solutions for mitigating scratching by pad asperities in Cu CMP are suggested.
INTRODUCTION
The chemical-mechanical polishing (CMP) of Cu is a critical step in the manufacture of ultra-large-scale integrated (ULSI) semiconductor devices. During this process, undesirable scratches are formed on the surface being polished [1-3] . Recent research suggests that the "killer" scratches found on the Cu wafers are due to the soft pad asperities and not necessarily by the hard abrasives in the slurry [4, 5 ] . Figure 1 shows examples of scratches on a Cu coating due to pad asperities. This paper presents the theory and experimental validation of scratching by soft pad asperities in Cu CMP. Based on the models and experimental results, practical solutions for mitigating scratching by pad asperities in Cu CMP are suggested.
Figure 1. Examples of "killer" scratches on eu coatings created by soft pad asperities.
SCRATCHING BY A SOFT PAD ASPERITY It was assumed that the deformation that occurs in the pad asperities is one of two extremes, i.e. elastic deformation at the onset of yielding or fully-plastic deformation. If the asperity deforms elastically and at the onset of yielding with friction present, then the contact pressure distribution is Hertzian, Figure 2(a), and the maximum pressure that the pad can apply to the Cu coating is about 1. 5 times the yield
strength of the pad, (Jy,p' Alternatively, if the pad deforms fully-plastically with friction present, then the contact pressure distribution is uniform, Figure 2(b), and the maximum pressure that the pad can apply to the Cu coating is 3(Jy,p, or equivalently, the pad hardness, Hp.
If a pad asperity deforms elastically and at the onset of yielding, as in Figure 2(a), then scratching of the Cu will not occur if [6] : H -p
<1.0 ,(0::::,u::::0.3)
He (I)�P
r (a) (b)Figure 2. (a) Elastic pad asperity at the onset of yielding with friction, (b) Fully-plastic pad asperity with friction.
Hp
2
[
(1-2V)2 (1-2v)(2-v),wr (16-4V+7V2),u27r2
l
'Y,
(2)-<- --
+
+
(,11203)H, 3 3
4
64
'
where Hp is the pad hardness, He is the Cu hardness, f1 is the coefficient of friction during polishing, and v is the Poisson's ratio of Cu. In order for the pad asperities to deform elastically, the radii of curvature of every pad asperity must be greater than a critical value, otherwise, to be conservative, it is assumed that at least some asperities deform plastically. Figure
3
shows Eqs. (1) and (2) labeled as "Hertzian Case. " If the process operating point is located above the line, then scratching of the Cu coating will occur. If, however, theoperating point is below the line, then scratching is not expected to occur.
For the case of a pad asperity that deforms fully plastically, finite element analyses (FEA) are used to analyze the uniform contact pressure profile when friction is taken into account. The FEA is used to locate the magnitude of the largest equivalent stress in the Cu coating. Then, the Mises yield criterion is used to obtain relations such that a fully plastic pad asperity will not scratch a Cu coating if [6] :
H -p
<0.34 , (0";,u";0.1)
He(3)
H1
[
J�
-p<- OA05+0.755,u+7.763,u2 2 , (,u�0.1)
He4
(4)
Figure
3
shows Eqs. (3) and(4),
labeled as "Plastic Case," plotted for different ratios of pad-to-Cu hardness and coefficients of friction. If the pad deforms plastically, and the process operating point is located above the respective line, then scratching of the Cu coating will occur. If, however, the operating point is below the line, then scratching is not expected to occur. 1.4 .,---, 1.2 � a�
1.0 t---\ :s, '" '" " c�
0.8 :c '" u B ;; 0.6 E ;.,;e
Hertzian Case]
0.4I
Plastic Case '" c:: --- - ---+--- ---0.2 0.0 +---.---,----.----;----1 0.0 0.2 0.4 0.6 0.8 1.0 Coefficient of Friction (p.)Figure 3. Regions of Cu scratching and no-scratching.
The primary parameter that affects the type of asperity deformation, i. e. elastic or plastic, is the asperity radius of curvature, Ra. A large radius of curvature will translate into a large contact area. As a result, for a given load, there is a greater probability that the asperity will not deform plastically. If every asperity has a radius of curvature greater than a critical value, Ra,crit. then the pad asperities will deform
elastically only. However, if any of the pad asperities have a radius of curvature smaller than Ra,crib then at least some of the asperities will deform plastically.
ASPERITY RADIUS OF CURVATURE
In order to determine Ra,crit, statistical methods were used to analyze a pad with multiple asperities, with a given distribution of asperity heights, in contact with a Cu coating. Multiple asperities, with a distribution of asperity heights, in contact with a planar surface were first examined by Greenwood and Williamson [7] . They assumed that the asperities had a height distribution, such as a Gaussian distribution. Furthermore, they assumed that every asperity had the same radius of curvature. Their primary concern was to calculate the real area of contact between the asperities and the planar surface. Although their work provides a theoretical framework to address the problem, the primary concern here is to determine the smallest asperity radius of curvature, given an asperity height distribution, such that the tallest asperity does not deform plastically. Therefore, it is the goal to determine Ra,crit such that the tallest asperity is at the onset of yielding. If every pad asperity has a radius of curvature Ra 2': Ra,crib then it can be guaranteed that every asperity in
contact with the planar Cu surface has a Hertzian contact pressure distribution. Once Ra,crit is determined, it is possible to properly design or break-in the pad.
1
1
1
111 11 11 14 1 1111
111
Figure 4. A CMP pad i n contact with a C u wafer.
A rough pad with multiple asperities in contact with a planar coating is shown in Figure 4. The critical radius of curvature, Ra,crib is a complex function of many parameters:
(5 ) where
¢ ( z
s)
is the distribution of pad asperity heights, (Is isthe standard deviation of pad asperity heights, p is the pressure applied to the pad, Aa is the pitch between pad asperities, and E* is the equivalent modulus of the pad/coating contact. In this analysis, four different distributions will be considered for
¢(z,)
: the Gaussian distribution, the exponential distribution, the uniform distribution, and the delta "distribution", for which case all of the asperities are the same height. The Gaussian, exponential and uniform distributions, shown in Figure 5 , all have the same standarddeviation. Furthermore, the area under each of the curves from z" = 0 to z" = CXJ is the same for all four curves. It is
important to note that the delta distribution should have a height that stretches to infinity and should have a width that approaches zero. Only the results of the Gaussian and delta distributions are presented in this paper, but the value of Ra,crit will be shown for all four distributions in Table 1.
¢(Z,)
Asperity Height, Zs (�m)
Figure 5. Four different asperity height distributions: Gaussian, exponential, uniform and delta.
(6) where
By solving Eq. (6) for Ra,crito it can be assured that if every asperity has a radius of curvature Ra:>: Ra,crito then every pad asperity will deform elastically for a pad that has asperity heights Gaussian-distributed. Then the elastic deformation curve in Figure 3 can be used. The integral in Eq. (6) cannot be solved analytically. Therefore, it must be solved numerically in order to determine the value of Ra,crit.
If the asperity height distribution is the delta distribution, then all of the asperities are of the same height, and the distribution of asperity heights is then defined as:
¢(Z.)=oo
at z s
=
z max(7)
where Zmax is a constant and the height of every asperity is
the same and equal to Zmax. For this distribution, the value of
Ra,crit can be obtained by a force balance on a single asperity, since all of the asperities are identical.
Therefore, Ra,crit can be explicitly solved for:
(8)
The standard deviation does not appear in Eq.
(8)
since it was assumed that all of the asperity heights are identical.Table 1 shows the values for Ra,crit for the four different distributions. The second column of the table shows the value of the tallest asperity. For the exponential and Gaussian distributions, a truncated value had to be chosen because both distributions stretch to infinity. Therefore, the third column in the table shows the probability that an asperity is taller than the truncation point shown in the second column. The fourth column shows the value of the critical radius of curvature, Ra,crit. This value was obtained using Eq. (6) for the Gaussian distribution and Eq.
(8)
for the delta distribution. For each of the distributions, if all of the pad asperities are greater than the critical value, i. e. Ra:>: Ra,crito then the asperities are loaded elastically and the elastic deformation curve can be used in Figure 3. However, if any of the pad asperities have a radius of curvature less than Ra,crito then the plastic deformation curve must be used in Fig. 4 to analyze pad scratching.Distribution � max P(�., >�rrmx) R G, orit (11m) C RG, orit (11m) (For C =1)
Exponential lOu, 0.002% 2103 0.49 500
Gaussian 4lJs 0.003% 1012 0.70 500
Uniform 1.70'., 0 144 1.86 500
Delta 0 21
Table 1. Summary of the critical asperity radius of curvature.
EXPERIMENTAL
By an appropriate choice of polishing pressure, pad hardness, lubricant, and level of pad conditioning, it is possible to significantly reduce scratching defects during CMP. Figure 4 shows a schematic of a CMP pad in contact with a wafer during polishing. Two Cu wafers were polished under typical CMP conditions. Samples of Rohm & Haas
rclOOO pads were used to conduct sliding tests on a l-�m thick Cu coated wafers using a sliding friction apparatus. The samples of pads were either new, or had been conditioned. The pads slid over the Cu wafers with either no lubricant or a 0. 01 molar solution of Sodium Dodecyl Sulfate (SDS) present, in order to vary the coefficient of friction. A normal load of 2 N was applied to the pads, which had a diameter of 0.02 m. This corresponds to about 5 kPa, or 0. 73 psi, which is within a typical range of pressure for CMP. The relative velocity of the pad was 0.005 mls and the range of the linear stage was 0. 025 m. That means that for every cycle of sliding, the linear stage, which held the Cu wafer, moved 0. 025 m to the left and then 0. 025 m to the right, at a velocity in both directions of 0. 005 m/s.
The first wafer was polished with a new pad and no lubricant present, which corresponded to a coefficient of friction of 0. 5 5 . The average asperity radius of curvature was relatively small, about 15 /lm. The second wafer was polished with a conditioned pad and SDS as the lubricant, which corresponded to a coefficient of friction of 0. 19. The average asperity radius of curvature was relatively large, about 100/lm.
SEM images of the two wafers polished with a new and a conditioned pad are shown in Figure 6(a) and 6(b), respectively. Table 2 shows the maximum scratch semi-with, ac,rnax, as well as the density of micron-sized scratches. When the wafer was polished with a new pad and large coefficient of friction, the maximum scratch semi-width was about 20 /lm. However, when a small coefficient of friction was used with a conditioned pad, the largest scratch semi-width was about 25 0 nm. The data points in Figure 3 correspond to the
experiments. By lowering the coefficient of friction and by increasing the radii of curvature, it was possible to decrease the severity of pad scratching.
New Pad New Pad
fJ = 0.55 fJ = 0.55 (a)
Conditioned Pad Conditioned Pad
(b)
Figure 6. SEM images of scratches on Cu wafers polished with: (a) a new pad and fJ. = 0.55, and (b) a conditioned pad and fJ. = 0. 19.
CONCLUSIONS
Based on the developed scratching models, as well as experimental validation, scratching by pad asperities can be eliminated. For a typical CMP pad, with a maximum pad hardness of 0.3 GPa, pad scratching can be eliminated if the coefficient of friction between the pad and the wafer is less than 0. 2. Furthermore, the pad should be conditioned such that
the smallest pad asperity radius of curvature is greater than 200/lm.
Pad f1
New 0.55
Conditioned 0. 19
a c ,max (11m) No. Scratches a c :>: 1 11m in 0.03
mm2
20 0.25
> 10 o
Table 2, The largest scratch found on a Cu wafer surface, as well as the frequency of scratches found, after polishing under typical CMP conditions.
REFERENCES
[1] N. -S. Kim, K. -M. Kang, Y. -S. Lim, l-H. So, S. -T. Moon, D. -l Lee, S. -M. Jun, "Correlations of Ceria Abrasives Characteristics Between Calcination Temperature and Polishing Performance," in Proceedings of the International Symposium on Semiconductor Manufacturing, 2003, pp.447-45 0.
[2] S. -M. Jung, J. -S. Uom, W. -S. Cho, Y. -J. Bae, Y. -K. Chung, K-S. Yu, K. -Y. Kim, K. -T. Kim, "A Study of Formation and Failure Mechanism of CMP Scratch Induced Defects on ILD in W-Damascene Interconnect SRAM Cell," in Proceedings of the International Reliability Physics Symposium, 2001, pp. 42-47.
[3] T. L. Neo, S. Y. Shang, C. M. Chong, M. Huang, C. M. Chen, FJ. Hsu, "CMP Defect Reduction by Micro-Scratch Control using New Monitoring technique," in Proceedings of the International Symposium on Semiconductor Manufacturing, 2001, pp. 321-324.
[4] N. Saka, T. Eusner, lH. Chun, "Scratching by Pad Asperities in Chemical-Mechanical Polishing," CIRP Annals Manufacturing Technology, vol. 5 9, pp. 329-332, 2010. [5 ] S. Lee, H. Kim, D. Dornfeld, "Development of a CMP Pad with Controlled Micro Features for Improved Performance," in Proceedings of the International Symposium on Semiconductor Manufacturing, 2005 , pp. 173- 176.
[6] T. Eusner, "Multi-Scale Scratching in Chemical Mechanical Polishing," Ph. D. Thesis, MIT, Cambridge, MA,
2010.
[7] lA. Greenwood, lB. P. Williamson, "Contact of Nominally Flat Surfaces," in Proceedings of the Royal Society of London, Series A, Mathematical and Physical Sciences, vol. 295 , pp. 300-319, 1966.
