- Email: [email protected]
Modeling and mitigation of pad scratching in chemical–mechanical polishing
G Model
CIRP-1022; No. of Pages 4 CIRP Annals - Manufacturing Technology xxx (2013) xxx–xxx
Contents lists available at SciVerse ScienceDirect
CIRP Annals - Manufacturing Technology jou rnal homep age : ht t p: // ees .e lse vi er . com /ci r p/ def a ult . asp
Modeling and mitigation of pad scratching in chemical–mechanical polishing S. Kim a,*, N. Saka a, J.-H. Chun (1)a, S.-H. Shin b a b
Laboratory for Manufacturing and Productivity, Massachusetts Institute of Technology, Cambridge, MA, USA Manufacturing Technology Center, Samsung Electronics Co., Ltd, Suwon, Republic of Korea
A R T I C L E I N F O
A B S T R A C T
Keywords: Defect Polishing Semiconductor
In the chemical–mechanical polishing (CMP) of semiconductor structures, such defects as micro- and nano-scale scratches are frequently produced on the surfaces being polished. Recent research shows that not only agglomerated abrasives but the softer pad asperities in frictional contact also scratch the relatively hard surfaces. Accordingly, pad scratching is modeled based on the topography and mechanical properties of pad asperities. Asperity radius, Ra, and the standard deviation of asperity heights, sz, are identified as the key topographical parameters. The theoretical models and experimental results show that pad scratching in CMP can be mitigated by increasing Ra/sz. ß 2013 CIRP.
1. Introduction The introduction of the dual damascene technique in the semiconductor industry over a decade ago has made possible the incorporation of copper interconnects in ultra-large-scale integrated (ULSI) circuits. To achieve both global planarization and local polishing of the Cu interconnects and dielectric layers, the chemical–mechanical polishing (CMP) process is widely used [1– 3]. A continuing problem in CMP, however, is the generation of the so-called ‘‘killer’’ scratches on the wafer surface. Such scratches are several hundred nanometers or even several micrometers wide. Moreover, in recent years as the size of the device features continued to shrink, and the usage of low-k, low-strength dielectrics increase, scratching has emerged as the main defectgeneration mechanism in CMP [4,5]. It is commonly believed that agglomerated hard, abrasive particles are the agents of scratching in CMP [6–8]. However, it has recently been found that under certain conditions even the pad, though soft, can scratch the relatively hard top layers being polished [9]. The width of the scratches produced by the pad is an order of magnitude greater than that of the particle-generated scratches [10]. Fig. 1 shows the scratches generated on Cu interconnects and low-k dielectric lines. The ‘‘polishing’’ experiments were conducted on Cu/low-k layers using a CMP pad but only with deionized water, i.e., without any abrasive particles. It is apparent that scratches produced by the pad asperities are far more critical than those produced by the agglomerated particles. In this paper, accordingly, pad scratching models are developed in terms of the topography and the mechanical properties of pad asperities. The key topographical parameters that promote pad scratching are identified. Based on the contact mechanics models, control of the topographical parameters of the polishing pads is found to be an effective method of scratch mitigation. Results of sliding experiments validate the theoretical prediction that
* Corresponding author.
Fig. 1. Scratches produced by pad asperities on damascened Cu/low-k layers. Only deionized water was used in the experiments.
scratching by the CMP pads can be mitigated by modifying their topography. It has been observed, additionally, that material removal rate is enhanced by topography modification. 2. Topography and properties of a typical CMP pad Surfaces of the CMP pads are porous, Fig. 2a, and relatively rough, Fig. 2b. The average pore size is about 50 mm and the average roughness is about 5 mm. To facilitate slurry flow and eliminate hydroplaning during CMP, the pad roughness is maintained by an in situ diamond conditioner. The profile of an
Fig. 2. SEM image and profile of the pad surface.
0007-8506/$ – see front matter ß 2013 CIRP. http://dx.doi.org/10.1016/j.cirp.2013.03.069
Please cite this article in press as: Kim S, et al. Modeling and mitigation of pad scratching in chemical–mechanical polishing. CIRP Annals - Manufacturing Technology (2013), http://dx.doi.org/10.1016/j.cirp.2013.03.069
G Model
CIRP-1022; No. of Pages 4 S. Kim et al. / CIRP Annals - Manufacturing Technology xxx (2013) xxx–xxx
2
Table 1 Topographical parameters and mechanical properties of an IC1000 CMP pad, and of a monolithic Cu layer. Material
Property
CMP Pad
za (mm) Ra (mm) la (mm) Ep (GPa) Hp (MPa)
Cu
El (GPa) Hl (MPa)
a
Std. Dev.
C.V. a
5.20 23.48 102.38 2.21 290
3.96 10.69 70.09 1.49 220
0.76 0.46 0.69 0.67 0.76
126.50 1560
12.51 260
0.10 0.17
Avg.
1=2 Hp 2 2:38m2 þ 0:45m þ 0:04 > ; ðm 0:3Þ 3 Hl
(2-b)
where Hp is the hardness of the pad asperities, Hl is the hardness of the layer being polished and m is the coefficient of friction. As the asperity deformation exceeds the elastic limit, the mean pressure applied by the asperity will be greater than its yield strength. In the extreme case of fully-plastic deformation of the asperity, the traction distribution will be uniform and the contact pressure will be three or four times the yield strength, as shown in Fig. 3b. Then the corresponding scratch criteria are [9]:
Coefficient of variation (C.V.) = Std. Dev./Avg.
IC1000 CMP pad manufactured by the Dow Chemical Co. is shown in Fig. 2b. Such topographical parameters, as the asperity height, za, radius, Ra, and the spacing, la, were determined by a Tencor P16 stylus profilometer and are summarized in Table 1. Additionally, the Young’s modulus and hardness, determined by a Hysitron TI900 nano-indenter, are also listed in the table. 3. Theory of scratching hard layers by soft pad asperities At the typical polishing pressures employed in CMP, about 7 kPa, the real area of contact is about a percent of the nominal area [11,12]. Thus, only the tallest asperities of the pad surface contact the layer being polished, and thus the geometry and the mechanical properties of these asperities play a dominant role in pad scratching. Scratching is primarily by the plastically deformed asperities in the contact. The proportion of plastic asperities among those in contact, therefore, essentially determines the number of scratches. For a soft asperity to scratch the relatively hard layers, however, certain criteria must be satisfied. Such criteria for the limiting modes of asperity deformation, elastic and fully-plastic, have been developed [9].
Hp > 0:34; ð0 m 0:1Þ Hl
(3-a)
1=2 Hp 1 7:76m2 þ 0:76m þ 0:41 > ; ðm 0:1Þ 4 Hl
(3-b)
Based on Eqs. (2) and (3), scratch-regime maps for elastically and plastically deformed asperities can be constructed as in Fig. 4a and b, respectively. A pad asperity can scratch the surface layer if the combination of hardness ratio and the friction coefficient falls in the ‘‘scratch regime’’ of the map. For the IC1000 pad and a Cu layer, and a friction coefficient (between the two surfaces in water) of 0.4, elastically deformed asperities of IC1000 pad do not scratch the Cu layer, whereas plastically deformed asperities will.
3.1. Single-asperity scratching When a pad asperity is loaded against a hard layer and slid, in addition to the normal pressure tangential tractions due to friction are also developed. But if friction is small and the deformation is elastic the far-field displacement, dp, of the asperity would be less than that at the elastic limit, dy,p, [13]:
dy; p ¼
p2 H p 16
Ep
2
Ra
(1) 3.2. Multi-asperity scratching
where Hp and Ep, respectively, are the hardness and Young’s modulus of the asperities. In elastic deforamtion due to normal loading only, the contact pressure distribution is elliptical or Hertzian. But with friction it will be as shown in Fig. 3a. In the limiting case of elastic deformation, i.e., when the asperity is about to yield, the maximum contact pressure is about 1.5 times the yield strength. Then the condition for an elastically deformed asperity to scratch the top layer can be written as a function of the ratio of padto-layer hardness and the friction coefficient [9]: Hp > 1; ð0 m 0:3Þ Hl
Fig. 4. Scratch-regime maps for elastic and fully-plastic pad asperities.
Generally, asperity heights of CMP pads are normally or exponentially distributed. Though normal distribution may possibly give a better description of the topography, the exponential distribution has analytical advantages and gives similar results [13]. Thus, if the asperity heights are exponentially distributed, the probability-density function, f(za), of the asperity height may be written as 1 Z fðza Þ ¼ exp a (4)
sz
(2-a)
sz
where sz is the standard deviation of asperity heights. As the pad is pressed against a hard, flat layer, only the tall asperities on the surface will be in contact, as shown in Fig. 5. If the distance of the layer surface from the centerline is d, the number of asperities in contact, Nc, is given by Nc ¼ N
Z1
fðZ a Þdza ¼ Nexp
d
sz
(5)
d
Fig. 3. Surface tractions at a pad asperity/hard layer contact.
where N is the total number of asperities. In order for an asperity to deform plastically, (za d) should be greater than dy,p, Eq. (1). Therefore, assuming that the radius of all the asperities is the same, the number of plastically deformed asperities, Np, is
Please cite this article in press as: Kim S, et al. Modeling and mitigation of pad scratching in chemical–mechanical polishing. CIRP Annals - Manufacturing Technology (2013), http://dx.doi.org/10.1016/j.cirp.2013.03.069
G Model
CIRP-1022; No. of Pages 4 S. Kim et al. / CIRP Annals - Manufacturing Technology xxx (2013) xxx–xxx
3
Fig. 5. Modes of contact of exponentially distributed asperities. Fig. 7. Schematic of asperity flattening processes.
Np ¼ N
Z1
d þ dy; p fðZ a Þdza ¼ Nexp
sz
(6)
dþdy; p
From Eqs. (1), (5) and (6), the proportion of plastically deformed asperities in the contact can be estimated by ( ) Np 1 ¼ exp 2 (7) Nc c where c is the plasticity index, which is defined as
c¼
sz dy; p
1=2 ¼
4 E p s z 1=2 p H p Ra
(8)
The dependence of the probability of plastic contact on the plasticity index is shown in Fig. 6. For the IC1000 pad, for example, the proportion of plastically deformed asperities is estimated to be about 0.92. Thus almost all the asperities in a contact are expected to deform plastically.
the asperity radius and the asperity spacing, subsurface deformation of the pad would be negligible, and only the pad asperities will deform and flatten [14]. A roller is preferred to a flat plate because it requires much less normal load to initiate plastic flow. Furthermore, by increasing the temperature of the metal plate, or the roller, asperity flattening can be accelerated. Asperities of circular disks, 20 mm in diameter, of the IC1000 pad were flattened by pressing the specimens against a flat stainless steel plate, and by rolling or sliding a stainless steel roller, 4.8 mm in diameter, over the pad. In flattening by a plate, a normal load of 400 N was applied, or at an average pressure of 300 kPa. The hold time was 60 s. In flattening the asperities by rolling and sliding, a normal load of 1 kN/m was applied on the roller, or at an average pressure of 2.3 MPa [14]. The sliding velocity was 5 mm/s. In elevated-temperature processing, the plate and the roller were heated to 185 8C before flattening. Scratching experiments were then conducted on a reciprocating friction apparatus, Fig. 8, using a new CMP pad and five differently asperity-flattened pads. The circular pad disks were slid over 1 mm thick Cu layer. The normal load applied to the specimen was 2 N, which corresponds to an average pressure of 7 kPa (1 psi). The sliding velocity was 7 mm/s. Four experiments were conducted with each fresh pad, using deionized water as a ‘‘lubricant’’. The scratches on the Cu layer were characterized by optical and scanning electron microscopes.
Fig. 6. Proportion of plastically deformed asperities versus 1/c2.
For pad scratching is mainly due to the plastically deformed asperities, scratching can be mitigated, from Eq. (7), by decreasing the value of c. That is, from Eq. (8), by increasing Hp/Ep and Ra/sz. While Hp/Ep is the stronger parameter, in general Hp and Ep are proportional to each other for various polymers and thus cannot be varied independently. Therefore, the most effective way of scratch mitigation seems to be by increasing the value of Ra/sz, i.e., by modifying pad topography. 4. Experimental The simplest way of increasing the value of Ra/sz is by pressing, at high pressure, the pad asperities against a flat metal plate as shown in Fig. 7a. If the pressure is sufficiently high, the radius of the tall asperities will be increased and the height variation reduced, due to flattening, thus increasing the value of Ra/sz. Alternatively, the asperities can also be flattened by rolling or sliding a metal, or ceramic, roller over the pad surface at high enough load, Fig. 7b. If the radius of the roller is much greater than
Fig. 8. Reciprocating friction apparatus.
Additionally, polishing experiments were conducted on a faceup polisher to determine the material removal rates. Two different sets of experimental conditions were used. In the first set, a slurry comprising 5 vol.% of Al2O3 abrasives of average size 300 nm, at a pressure of 13 kPa (2 psi) and a velocity of 0.87 m/s. In the second set a commercial slurry (HS-BT815, Hitachi Chemical Co.) was used, the pressure was 7 kPa (1 psi) and velocity was 0.66 m/s. 5. Results and discussion Table 2 lists the Ra and sz values of an IC1000 pad before and after asperity flattening under different process conditions. Not
Please cite this article in press as: Kim S, et al. Modeling and mitigation of pad scratching in chemical–mechanical polishing. CIRP Annals - Manufacturing Technology (2013), http://dx.doi.org/10.1016/j.cirp.2013.03.069
G Model
CIRP-1022; No. of Pages 4 S. Kim et al. / CIRP Annals - Manufacturing Technology xxx (2013) xxx–xxx
4
Table 2 Radius, standard deviation of asperity heights, and their ratio of IC1000 pads. Also listed are the number of scratches generated by the pads on a Cu layer in sliding experiments. Parameter
Ra (mm) sz (mm) Ra/sz # of scratches
New pad
23.48 3.96 5.93 23
Flattened pad Compression
Rolling
300 kPa
2300 kPa
Sliding
25 8C
185 8C
25 8C
185 8C
25 8C
39.49 3.76 10.59 11
173.41 2.02 85.84 10
57.83 3.62 15.98 5
106.90 3.19 33.51 5
72.33 3.03 23.80 1
surprisingly, the increments in the Ra/sz values of asperityflattened pads using a roller are greater than those of the pad flattened by a flat plate––due to the higher average pressure in rolling. Moreover, the pad flattened by sliding the roller has greater Ra/sz value than that of the pad flattened by just pure rolling, due to interfacial friction. Therefore, asperity flattening could be made more effective by imposing sliding motion. However, it should be noted that sliding at high contact pressures may result in heavy wear of both the pad and the roller. The results of the scratch experiments show, Fig. 9, that the number of scratches indeed decreases as the Ra/sz value increases. While the Ra/sz value can be further increased by raising the process temperature and increasing the duration of loading, surprisingly scratch reduction with pads flattened at high temperatures was found to be less than that using the pads with asperities flattened at room temperature.
It is apparent from Fig. 10 that the material removal rate can be increased and the propensity for scratching decreased by controlling the topography of the CMP pads, i.e., by flattening the asperities. One way of controlling pad topography is by casting the pads in a mold with a micro-dimpled surface [15]. Then, the height, radius and spacing of the asperities, too, can be independently controlled. Unfortunately, however, since the pads used in CMP are rather large, 0.7 m in diameter, manufacture of large molds with high-precision surfaces may not be possible. Furthermore, since pad topography continuously changes during CMP, due both to conditioning and wear, mitigation of pad scratching by molding the pads with optimal initial geometry may not eliminate scratching. Only novel methods of maintaining high Ra/sz value during polishing, such as in situ asperity flattening, are expected to mitigate scratching. 6. Conclusions In this work, pad scratching and its mitigation in CMP have been investigated, and the following conclusions are drawn. 1. Contact mechanics models predict that the number of scratches produced in CMP can be reduced by decreasing the proportion of plastically deformed asperities for they are the primary source of pad scratching. 2. The ratio of asperity radius to the standard deviation of asperity heights, Ra/sz, is identified as the key parameter that determines the proportion of asperities in plastic contact. 3. Scratching experiments have shown that pad scratching is mitigated that by flattening the asperities. 4. Polishing experiments have shown that material removal rate is enhanced by flattening the pad asperities.
Acknowledgments The present study was funded by the Samsung Electronics Company, Ltd. References
Fig. 9. Normalized number of scratches versus 1/c2. The points represent the average values and the bars the standard error.
The results of polishing experiments are shown in Fig. 10. Material removal rates were higher with pads flattened by sliding rollers. It seems, then, that an added benefit of flattening the asperities is that the contact area and hence the material removal rate will be increased. As asperities are flattened to have more elastic contacts, the real area of contact for a given nominal pressure will be increased compared with that of the new pads. Therefore, the material removal rate with flattened pads is expected to be greater than that with new pads.
Fig. 10. Material removal rate using new and asperity-flattened IC1000 pads. The asperities were flattened by sliding a roller over the pad.
[1] Singer P (1998) Tantalum, Copper and Damascene: The Future of Interconnects. Semiconductor International 21(6):90–98. [2] Saka N, Lai JY, Chun J-H, Suh NP (2001) Mechanisms of the Chemical Mechanical Polishing (CMP) Process in Integrated Circuit Fabrication. Annals of the CIRP 50(1):233–238. [3] Gambino J (2009) Copper Interconnect Technology For the 32 nm Node and Beyond. IEEE Custom Integrated Circuits Conference (CICC), San Jose, CA, 141–148. [4] Chandrasekaran N, Ramarajan S, Lee W, Sabde M, Meikle S (2004) Effects of CMP Process Conditions on Defect Generation in Low-k Materials. Journal of Electrochemical Society 151(12):G882–G889. [5] Teo TY, Goh WL, Lim VSK, Leong LS, Tse TY, Chan L (2004) Characterization of Scratches Generated by a Mutiplaten Copper Chemical–Mechanical Polishing Process. Journal of Vacuum Science and Technology 22(1):65–69. [6] Chandra A, Karra P, Bastawros AF, Biswas R, Sherman PJ, Armini S, Lucca DA (2008) Prediction of Scratch Generation in Chemical Mechanical Planarization. Annals of the CIRP 57(1):559–562. [7] Armini S, Whelan CM, Moinpour M, Maex K (2009) Copper CMP with Composite Polymer Core-Silica Slurry Abrasives: A Defectivity Study. Journal of the Electrochemical Society 156(1):H18–H26. [8] Saka N, Eusner T, Chun J-H (2008) Nano-Scale Scratching in Chemical– Mechanical Polishing. Annals of the CIRP 57(1):341–344. [9] Saka N, Eusner T, Chun J-H (2010) Scratching by Pad Asperities in Chemical– Mechanical Polishing. Annals of the CIRP 59(1):329–332. [10] Eusner T, Saka N, Chun J-H (2011) Breaking-in a Pad for Scratch-Free, Cu Chemical–Mechanical Polishing. Journal of the Electrochemical Society 158(4):H379–H389. [11] Greenwood JA, Williamson JBP (1966) Contact of Nominally Flat Surfaces. Proceedings of the Royal Society of London A 243:190–205. [12] Elmufdi CL, Muldowney GP (2006) A Novel Optical Technique to Measure PadWafer Contact Area in Chemical Mechanical Planarization. Materials Research Society Proceedings 914: 0914-F12-06. [13] Johnson KL (1985) Contact Mechanics, Cambridge University Press, Cambridge, UK. [14] Rajendrakumar PK, Biswas SK (1997) Deformation Due to Contact Between a Two-Dimensional Rough Surface and a Smooth Cylinder. Tribology Letters 3:297–301. [15] Lee S, Kim H, Dornfeld D (2005) Development of a CMP Pad with Controlled Micro Features for Improved Performance. IEEE International Symposium on Semiconductor Manufacturing (ISSM) 173–176.
Please cite this article in press as: Kim S, et al. Modeling and mitigation of pad scratching in chemical–mechanical polishing. CIRP Annals - Manufacturing Technology (2013), http://dx.doi.org/10.1016/j.cirp.2013.03.069
CIRP-1022; No. of Pages 4 CIRP Annals - Manufacturing Technology xxx (2013) xxx–xxx
Contents lists available at SciVerse ScienceDirect
CIRP Annals - Manufacturing Technology jou rnal homep age : ht t p: // ees .e lse vi er . com /ci r p/ def a ult . asp
Modeling and mitigation of pad scratching in chemical–mechanical polishing S. Kim a,*, N. Saka a, J.-H. Chun (1)a, S.-H. Shin b a b
Laboratory for Manufacturing and Productivity, Massachusetts Institute of Technology, Cambridge, MA, USA Manufacturing Technology Center, Samsung Electronics Co., Ltd, Suwon, Republic of Korea
A R T I C L E I N F O
A B S T R A C T
Keywords: Defect Polishing Semiconductor
In the chemical–mechanical polishing (CMP) of semiconductor structures, such defects as micro- and nano-scale scratches are frequently produced on the surfaces being polished. Recent research shows that not only agglomerated abrasives but the softer pad asperities in frictional contact also scratch the relatively hard surfaces. Accordingly, pad scratching is modeled based on the topography and mechanical properties of pad asperities. Asperity radius, Ra, and the standard deviation of asperity heights, sz, are identified as the key topographical parameters. The theoretical models and experimental results show that pad scratching in CMP can be mitigated by increasing Ra/sz. ß 2013 CIRP.
1. Introduction The introduction of the dual damascene technique in the semiconductor industry over a decade ago has made possible the incorporation of copper interconnects in ultra-large-scale integrated (ULSI) circuits. To achieve both global planarization and local polishing of the Cu interconnects and dielectric layers, the chemical–mechanical polishing (CMP) process is widely used [1– 3]. A continuing problem in CMP, however, is the generation of the so-called ‘‘killer’’ scratches on the wafer surface. Such scratches are several hundred nanometers or even several micrometers wide. Moreover, in recent years as the size of the device features continued to shrink, and the usage of low-k, low-strength dielectrics increase, scratching has emerged as the main defectgeneration mechanism in CMP [4,5]. It is commonly believed that agglomerated hard, abrasive particles are the agents of scratching in CMP [6–8]. However, it has recently been found that under certain conditions even the pad, though soft, can scratch the relatively hard top layers being polished [9]. The width of the scratches produced by the pad is an order of magnitude greater than that of the particle-generated scratches [10]. Fig. 1 shows the scratches generated on Cu interconnects and low-k dielectric lines. The ‘‘polishing’’ experiments were conducted on Cu/low-k layers using a CMP pad but only with deionized water, i.e., without any abrasive particles. It is apparent that scratches produced by the pad asperities are far more critical than those produced by the agglomerated particles. In this paper, accordingly, pad scratching models are developed in terms of the topography and the mechanical properties of pad asperities. The key topographical parameters that promote pad scratching are identified. Based on the contact mechanics models, control of the topographical parameters of the polishing pads is found to be an effective method of scratch mitigation. Results of sliding experiments validate the theoretical prediction that
* Corresponding author.
Fig. 1. Scratches produced by pad asperities on damascened Cu/low-k layers. Only deionized water was used in the experiments.
scratching by the CMP pads can be mitigated by modifying their topography. It has been observed, additionally, that material removal rate is enhanced by topography modification. 2. Topography and properties of a typical CMP pad Surfaces of the CMP pads are porous, Fig. 2a, and relatively rough, Fig. 2b. The average pore size is about 50 mm and the average roughness is about 5 mm. To facilitate slurry flow and eliminate hydroplaning during CMP, the pad roughness is maintained by an in situ diamond conditioner. The profile of an
Fig. 2. SEM image and profile of the pad surface.
0007-8506/$ – see front matter ß 2013 CIRP. http://dx.doi.org/10.1016/j.cirp.2013.03.069
Please cite this article in press as: Kim S, et al. Modeling and mitigation of pad scratching in chemical–mechanical polishing. CIRP Annals - Manufacturing Technology (2013), http://dx.doi.org/10.1016/j.cirp.2013.03.069
G Model
CIRP-1022; No. of Pages 4 S. Kim et al. / CIRP Annals - Manufacturing Technology xxx (2013) xxx–xxx
2
Table 1 Topographical parameters and mechanical properties of an IC1000 CMP pad, and of a monolithic Cu layer. Material
Property
CMP Pad
za (mm) Ra (mm) la (mm) Ep (GPa) Hp (MPa)
Cu
El (GPa) Hl (MPa)
a
Std. Dev.
C.V. a
5.20 23.48 102.38 2.21 290
3.96 10.69 70.09 1.49 220
0.76 0.46 0.69 0.67 0.76
126.50 1560
12.51 260
0.10 0.17
Avg.
1=2 Hp 2 2:38m2 þ 0:45m þ 0:04 > ; ðm 0:3Þ 3 Hl
(2-b)
where Hp is the hardness of the pad asperities, Hl is the hardness of the layer being polished and m is the coefficient of friction. As the asperity deformation exceeds the elastic limit, the mean pressure applied by the asperity will be greater than its yield strength. In the extreme case of fully-plastic deformation of the asperity, the traction distribution will be uniform and the contact pressure will be three or four times the yield strength, as shown in Fig. 3b. Then the corresponding scratch criteria are [9]:
Coefficient of variation (C.V.) = Std. Dev./Avg.
IC1000 CMP pad manufactured by the Dow Chemical Co. is shown in Fig. 2b. Such topographical parameters, as the asperity height, za, radius, Ra, and the spacing, la, were determined by a Tencor P16 stylus profilometer and are summarized in Table 1. Additionally, the Young’s modulus and hardness, determined by a Hysitron TI900 nano-indenter, are also listed in the table. 3. Theory of scratching hard layers by soft pad asperities At the typical polishing pressures employed in CMP, about 7 kPa, the real area of contact is about a percent of the nominal area [11,12]. Thus, only the tallest asperities of the pad surface contact the layer being polished, and thus the geometry and the mechanical properties of these asperities play a dominant role in pad scratching. Scratching is primarily by the plastically deformed asperities in the contact. The proportion of plastic asperities among those in contact, therefore, essentially determines the number of scratches. For a soft asperity to scratch the relatively hard layers, however, certain criteria must be satisfied. Such criteria for the limiting modes of asperity deformation, elastic and fully-plastic, have been developed [9].
Hp > 0:34; ð0 m 0:1Þ Hl
(3-a)
1=2 Hp 1 7:76m2 þ 0:76m þ 0:41 > ; ðm 0:1Þ 4 Hl
(3-b)
Based on Eqs. (2) and (3), scratch-regime maps for elastically and plastically deformed asperities can be constructed as in Fig. 4a and b, respectively. A pad asperity can scratch the surface layer if the combination of hardness ratio and the friction coefficient falls in the ‘‘scratch regime’’ of the map. For the IC1000 pad and a Cu layer, and a friction coefficient (between the two surfaces in water) of 0.4, elastically deformed asperities of IC1000 pad do not scratch the Cu layer, whereas plastically deformed asperities will.
3.1. Single-asperity scratching When a pad asperity is loaded against a hard layer and slid, in addition to the normal pressure tangential tractions due to friction are also developed. But if friction is small and the deformation is elastic the far-field displacement, dp, of the asperity would be less than that at the elastic limit, dy,p, [13]:
dy; p ¼
p2 H p 16
Ep
2
Ra
(1) 3.2. Multi-asperity scratching
where Hp and Ep, respectively, are the hardness and Young’s modulus of the asperities. In elastic deforamtion due to normal loading only, the contact pressure distribution is elliptical or Hertzian. But with friction it will be as shown in Fig. 3a. In the limiting case of elastic deformation, i.e., when the asperity is about to yield, the maximum contact pressure is about 1.5 times the yield strength. Then the condition for an elastically deformed asperity to scratch the top layer can be written as a function of the ratio of padto-layer hardness and the friction coefficient [9]: Hp > 1; ð0 m 0:3Þ Hl
Fig. 4. Scratch-regime maps for elastic and fully-plastic pad asperities.
Generally, asperity heights of CMP pads are normally or exponentially distributed. Though normal distribution may possibly give a better description of the topography, the exponential distribution has analytical advantages and gives similar results [13]. Thus, if the asperity heights are exponentially distributed, the probability-density function, f(za), of the asperity height may be written as 1 Z fðza Þ ¼ exp a (4)
sz
(2-a)
sz
where sz is the standard deviation of asperity heights. As the pad is pressed against a hard, flat layer, only the tall asperities on the surface will be in contact, as shown in Fig. 5. If the distance of the layer surface from the centerline is d, the number of asperities in contact, Nc, is given by Nc ¼ N
Z1
fðZ a Þdza ¼ Nexp
d
sz
(5)
d
Fig. 3. Surface tractions at a pad asperity/hard layer contact.
where N is the total number of asperities. In order for an asperity to deform plastically, (za d) should be greater than dy,p, Eq. (1). Therefore, assuming that the radius of all the asperities is the same, the number of plastically deformed asperities, Np, is
Please cite this article in press as: Kim S, et al. Modeling and mitigation of pad scratching in chemical–mechanical polishing. CIRP Annals - Manufacturing Technology (2013), http://dx.doi.org/10.1016/j.cirp.2013.03.069
G Model
CIRP-1022; No. of Pages 4 S. Kim et al. / CIRP Annals - Manufacturing Technology xxx (2013) xxx–xxx
3
Fig. 5. Modes of contact of exponentially distributed asperities. Fig. 7. Schematic of asperity flattening processes.
Np ¼ N
Z1
d þ dy; p fðZ a Þdza ¼ Nexp
sz
(6)
dþdy; p
From Eqs. (1), (5) and (6), the proportion of plastically deformed asperities in the contact can be estimated by ( ) Np 1 ¼ exp 2 (7) Nc c where c is the plasticity index, which is defined as
c¼
sz dy; p
1=2 ¼
4 E p s z 1=2 p H p Ra
(8)
The dependence of the probability of plastic contact on the plasticity index is shown in Fig. 6. For the IC1000 pad, for example, the proportion of plastically deformed asperities is estimated to be about 0.92. Thus almost all the asperities in a contact are expected to deform plastically.
the asperity radius and the asperity spacing, subsurface deformation of the pad would be negligible, and only the pad asperities will deform and flatten [14]. A roller is preferred to a flat plate because it requires much less normal load to initiate plastic flow. Furthermore, by increasing the temperature of the metal plate, or the roller, asperity flattening can be accelerated. Asperities of circular disks, 20 mm in diameter, of the IC1000 pad were flattened by pressing the specimens against a flat stainless steel plate, and by rolling or sliding a stainless steel roller, 4.8 mm in diameter, over the pad. In flattening by a plate, a normal load of 400 N was applied, or at an average pressure of 300 kPa. The hold time was 60 s. In flattening the asperities by rolling and sliding, a normal load of 1 kN/m was applied on the roller, or at an average pressure of 2.3 MPa [14]. The sliding velocity was 5 mm/s. In elevated-temperature processing, the plate and the roller were heated to 185 8C before flattening. Scratching experiments were then conducted on a reciprocating friction apparatus, Fig. 8, using a new CMP pad and five differently asperity-flattened pads. The circular pad disks were slid over 1 mm thick Cu layer. The normal load applied to the specimen was 2 N, which corresponds to an average pressure of 7 kPa (1 psi). The sliding velocity was 7 mm/s. Four experiments were conducted with each fresh pad, using deionized water as a ‘‘lubricant’’. The scratches on the Cu layer were characterized by optical and scanning electron microscopes.
Fig. 6. Proportion of plastically deformed asperities versus 1/c2.
For pad scratching is mainly due to the plastically deformed asperities, scratching can be mitigated, from Eq. (7), by decreasing the value of c. That is, from Eq. (8), by increasing Hp/Ep and Ra/sz. While Hp/Ep is the stronger parameter, in general Hp and Ep are proportional to each other for various polymers and thus cannot be varied independently. Therefore, the most effective way of scratch mitigation seems to be by increasing the value of Ra/sz, i.e., by modifying pad topography. 4. Experimental The simplest way of increasing the value of Ra/sz is by pressing, at high pressure, the pad asperities against a flat metal plate as shown in Fig. 7a. If the pressure is sufficiently high, the radius of the tall asperities will be increased and the height variation reduced, due to flattening, thus increasing the value of Ra/sz. Alternatively, the asperities can also be flattened by rolling or sliding a metal, or ceramic, roller over the pad surface at high enough load, Fig. 7b. If the radius of the roller is much greater than
Fig. 8. Reciprocating friction apparatus.
Additionally, polishing experiments were conducted on a faceup polisher to determine the material removal rates. Two different sets of experimental conditions were used. In the first set, a slurry comprising 5 vol.% of Al2O3 abrasives of average size 300 nm, at a pressure of 13 kPa (2 psi) and a velocity of 0.87 m/s. In the second set a commercial slurry (HS-BT815, Hitachi Chemical Co.) was used, the pressure was 7 kPa (1 psi) and velocity was 0.66 m/s. 5. Results and discussion Table 2 lists the Ra and sz values of an IC1000 pad before and after asperity flattening under different process conditions. Not
Please cite this article in press as: Kim S, et al. Modeling and mitigation of pad scratching in chemical–mechanical polishing. CIRP Annals - Manufacturing Technology (2013), http://dx.doi.org/10.1016/j.cirp.2013.03.069
G Model
CIRP-1022; No. of Pages 4 S. Kim et al. / CIRP Annals - Manufacturing Technology xxx (2013) xxx–xxx
4
Table 2 Radius, standard deviation of asperity heights, and their ratio of IC1000 pads. Also listed are the number of scratches generated by the pads on a Cu layer in sliding experiments. Parameter
Ra (mm) sz (mm) Ra/sz # of scratches
New pad
23.48 3.96 5.93 23
Flattened pad Compression
Rolling
300 kPa
2300 kPa
Sliding
25 8C
185 8C
25 8C
185 8C
25 8C
39.49 3.76 10.59 11
173.41 2.02 85.84 10
57.83 3.62 15.98 5
106.90 3.19 33.51 5
72.33 3.03 23.80 1
surprisingly, the increments in the Ra/sz values of asperityflattened pads using a roller are greater than those of the pad flattened by a flat plate––due to the higher average pressure in rolling. Moreover, the pad flattened by sliding the roller has greater Ra/sz value than that of the pad flattened by just pure rolling, due to interfacial friction. Therefore, asperity flattening could be made more effective by imposing sliding motion. However, it should be noted that sliding at high contact pressures may result in heavy wear of both the pad and the roller. The results of the scratch experiments show, Fig. 9, that the number of scratches indeed decreases as the Ra/sz value increases. While the Ra/sz value can be further increased by raising the process temperature and increasing the duration of loading, surprisingly scratch reduction with pads flattened at high temperatures was found to be less than that using the pads with asperities flattened at room temperature.
It is apparent from Fig. 10 that the material removal rate can be increased and the propensity for scratching decreased by controlling the topography of the CMP pads, i.e., by flattening the asperities. One way of controlling pad topography is by casting the pads in a mold with a micro-dimpled surface [15]. Then, the height, radius and spacing of the asperities, too, can be independently controlled. Unfortunately, however, since the pads used in CMP are rather large, 0.7 m in diameter, manufacture of large molds with high-precision surfaces may not be possible. Furthermore, since pad topography continuously changes during CMP, due both to conditioning and wear, mitigation of pad scratching by molding the pads with optimal initial geometry may not eliminate scratching. Only novel methods of maintaining high Ra/sz value during polishing, such as in situ asperity flattening, are expected to mitigate scratching. 6. Conclusions In this work, pad scratching and its mitigation in CMP have been investigated, and the following conclusions are drawn. 1. Contact mechanics models predict that the number of scratches produced in CMP can be reduced by decreasing the proportion of plastically deformed asperities for they are the primary source of pad scratching. 2. The ratio of asperity radius to the standard deviation of asperity heights, Ra/sz, is identified as the key parameter that determines the proportion of asperities in plastic contact. 3. Scratching experiments have shown that pad scratching is mitigated that by flattening the asperities. 4. Polishing experiments have shown that material removal rate is enhanced by flattening the pad asperities.
Acknowledgments The present study was funded by the Samsung Electronics Company, Ltd. References
Fig. 9. Normalized number of scratches versus 1/c2. The points represent the average values and the bars the standard error.
The results of polishing experiments are shown in Fig. 10. Material removal rates were higher with pads flattened by sliding rollers. It seems, then, that an added benefit of flattening the asperities is that the contact area and hence the material removal rate will be increased. As asperities are flattened to have more elastic contacts, the real area of contact for a given nominal pressure will be increased compared with that of the new pads. Therefore, the material removal rate with flattened pads is expected to be greater than that with new pads.
Fig. 10. Material removal rate using new and asperity-flattened IC1000 pads. The asperities were flattened by sliding a roller over the pad.
[1] Singer P (1998) Tantalum, Copper and Damascene: The Future of Interconnects. Semiconductor International 21(6):90–98. [2] Saka N, Lai JY, Chun J-H, Suh NP (2001) Mechanisms of the Chemical Mechanical Polishing (CMP) Process in Integrated Circuit Fabrication. Annals of the CIRP 50(1):233–238. [3] Gambino J (2009) Copper Interconnect Technology For the 32 nm Node and Beyond. IEEE Custom Integrated Circuits Conference (CICC), San Jose, CA, 141–148. [4] Chandrasekaran N, Ramarajan S, Lee W, Sabde M, Meikle S (2004) Effects of CMP Process Conditions on Defect Generation in Low-k Materials. Journal of Electrochemical Society 151(12):G882–G889. [5] Teo TY, Goh WL, Lim VSK, Leong LS, Tse TY, Chan L (2004) Characterization of Scratches Generated by a Mutiplaten Copper Chemical–Mechanical Polishing Process. Journal of Vacuum Science and Technology 22(1):65–69. [6] Chandra A, Karra P, Bastawros AF, Biswas R, Sherman PJ, Armini S, Lucca DA (2008) Prediction of Scratch Generation in Chemical Mechanical Planarization. Annals of the CIRP 57(1):559–562. [7] Armini S, Whelan CM, Moinpour M, Maex K (2009) Copper CMP with Composite Polymer Core-Silica Slurry Abrasives: A Defectivity Study. Journal of the Electrochemical Society 156(1):H18–H26. [8] Saka N, Eusner T, Chun J-H (2008) Nano-Scale Scratching in Chemical– Mechanical Polishing. Annals of the CIRP 57(1):341–344. [9] Saka N, Eusner T, Chun J-H (2010) Scratching by Pad Asperities in Chemical– Mechanical Polishing. Annals of the CIRP 59(1):329–332. [10] Eusner T, Saka N, Chun J-H (2011) Breaking-in a Pad for Scratch-Free, Cu Chemical–Mechanical Polishing. Journal of the Electrochemical Society 158(4):H379–H389. [11] Greenwood JA, Williamson JBP (1966) Contact of Nominally Flat Surfaces. Proceedings of the Royal Society of London A 243:190–205. [12] Elmufdi CL, Muldowney GP (2006) A Novel Optical Technique to Measure PadWafer Contact Area in Chemical Mechanical Planarization. Materials Research Society Proceedings 914: 0914-F12-06. [13] Johnson KL (1985) Contact Mechanics, Cambridge University Press, Cambridge, UK. [14] Rajendrakumar PK, Biswas SK (1997) Deformation Due to Contact Between a Two-Dimensional Rough Surface and a Smooth Cylinder. Tribology Letters 3:297–301. [15] Lee S, Kim H, Dornfeld D (2005) Development of a CMP Pad with Controlled Micro Features for Improved Performance. IEEE International Symposium on Semiconductor Manufacturing (ISSM) 173–176.
Please cite this article in press as: Kim S, et al. Modeling and mitigation of pad scratching in chemical–mechanical polishing. CIRP Annals - Manufacturing Technology (2013), http://dx.doi.org/10.1016/j.cirp.2013.03.069