344 IEEE TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 26, NO.

3, AUGUST 2013

The Effect of Cu CMP Pad Clean on Defectivity

and Reliability
Leong Lup San, Lin Benfu, Yu Hong, Zhu Yue Qin, Lu Wei, Lydia Helena Wong, and Archana Mishra

Abstract—A type of CMP scratches with an embedded particle

and a comet-like tail has been studied. The nature of the embed-
ded particle was found to be silica which suggested that the source
came from barrier metal polishing step. The inclusion of a pad-
cleaning step during barrier polishing was found to reduce the
defect significantly. This pad cleaning step reduce hydrophobicity
of the polishing pad. Through a series of experiments, it was
found that BTA was the cause of high hydrophobicity build up
on during barrier polish and a defect mechanism was proposed.
In addition, pad cleaning has positive impact on TDDB. The
reason for the the reliability improvement was believed to be
due to formation of divots at the side of the trench which, after
deposition of a capping nitride layer, will increase resistance for
Cu migration during voltage cycle.
Index Terms—Chemical mechanical polish, defects, pad clean,
reliability.

I. Introduction
OPPER Chemical Mechanical Polishing (Cu CMP) using
C a multiplaten approach is a key process in the fabrication
of Cu interconnects. However, being relatively soft and having
high plastic deformation area, copper is easily scratched. There
are many possible root causes for the scratches, as shown in
Fig. 1. Besides the normal arc scratches attributed to embedded
particls on the pad eg diamond abrasives from the conditioner,
there are the normal chatter marks from large particls in the
slurry. We had also previously studied scratches related to
alumina abrasives characterized by deep triangular depressions
[1], [2], with Al elemental signature as shown in Fig. 2. In this
paper, we identified the source and mechanism of a scratch
defect characterized by comet-like signature and uniformly
sized embedded particle about 0.1um in diameter. We also
suggested a method of reducing this defect by means of a Fig. 1. Different CuCMP scratches signatures a) Arc scratches b) Chatter
pad cleaning step. The effect of this improvement on time- marks c) Triangular deep scratches.
dependent dielectric breakdown (TDDB) was also discussed
uration were used, bulk copper removal at platen 1, complete
II. Experiments removal of copper and endpoint at barrier metal at platen 2
Wafers were polished on the Applied Materials 300 mm and barrier removal at platen 3. IC 1010 pads were used for
Reflexion LK polisher with Contour Head TM . Standard config- platen 1 and 2 and Politex pad for platen 3. All wafers were
passivated with 1,2,3 benzotriazole (BTA) at the final step of
Manuscript received February 15, 2013; revised June 12, 2013; accepted
June 16, 2013. Date of publication June 27, 2013; date of current version platen3. Contact angle measurements were measured using
July 31, 2013. a KRUSS EasyDrop® instrument. Topology was measured
L. L. San, L. Benfu, Y. Hong, Z. Y. Qin, and L. Wei are with Global using KLA Tencor HRP 340 TM . Alumina-based copper slurry
Foundries Singapore Pte Ltd., 738406, Singapore.
L. H. Wong and A. Mishra are with Nanyang Technological University, and silica based barrier slurry were used in the experiments.
639798, Singapore (e-mail: [email protected]). Pattern wafers used for the evaluation are of 45 nm node which
Color versions of one or more of the figures in this paper are available has a silicon nitride-based (SiN) hard mask capping a dielectric
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TSM.2013.2271385 material (Dielectric constant K = 3.0).
0894-6507/$31.00 
c 2013 IEEE
SAN et al.: EFFECT OF CU CMP PAD CLEAN ON DEFECTIVITY AND RELIABILITY 345

Fig. 4. SEM (inset) and Auger of the embedded particle.

Fig. 2. SEM (left) and EDX of embedded particle of triangular scratch.

Fig. 5. Particle size distribution of barrier slurry.

Having identified the origin of embedded particle, there

were 2 possible strategies to contain this defect. The first
would involve consumable change – which would be time
consuming and involve re-qualifications of some products.
The second option was for process change, in which several
experiments were attempted.
One of the method used for the second option was to
Fig. 3. Scratch marks with embedded particle and comet tail. insert an additional in-situ acidic Pad Clean (PC) to clean
the Platen 3 pad prior to actual wafer touching down on
the pad. In the original process of record (POR), only a
III. Results
same duration Deionised water (DIW) rinse was performed.
A. Defect Signature and Reduction Comparison of POR and this PC process was done by
The signature of the defect of interest was randomly polishing blanket wafers. The wafers were then scanned by
distributed scratches with a comet-like tail and an embed- KLA Tencor Surfscan® SP1 and defects were sampled and
ded particle as shown in Fig. 3. Using Auger spectroscopy, classified manually after SEM inspection. In this experiment,
the embedded particle was found to be silica in nature as defects classified as “deep scratches” were those with longer
shown in Fig. 4, which suggested it came from platen3 than 0.5um signature while the combined class code “micro
since platen 1 and 2 utilise Alumina slurry. In addition, the scratches and chatter marks” were those below 0.5um. Based
size of the embedded particles is quite uniform – between on the typical size of the defect-of-interest as shown in
0.09 to 0.1um, and corresponded to the median particle Fig. 3, it would certainly be classified under deep scratches
size distribution of the platen3 slurry as shown in Fig. 5, although some may be classified under micro scratches as well.
which suggested that additional filtration on-site will not As shown in Fig. 6, overall defect counts were reduced by
improve the defect without affecting polish rate and unifor- more than three times using PC process, and the longer deep
mity. scratches were almost eradicated. Defects classified as “not
346 IEEE TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 26, NO. 3, AUGUST 2013

Fig. 8. Trend of normalised defect density (NDD) of polish scratches in 2

Fig. 6. POR vs PC comparison of defects of blanket Cu wafers. products after extended run of PC process.

Fig. 9. Contact angle of blanket wafer.

Fig. 7. POR vs PC comparison of defect pareto of pattern wafers.

molecular mass peaks of 26, 118 and 119 corresponding to
moieties CN, C6H4N3 and C6H5N3 respectively. These peaks
found” were chatter marks or micro scratches masked by the
are to be expected from the breakdown of BTA molecules.
deeper and longer scratch marks. Similar trend was noted on
Spectra of the wafer that undergo PC was found to have
pattern wafers with overall reduction of total defects and most
much less intense signals for these 3 peaks, indicating lower
notably scratches as shown in Fig 7. Since polish scratches
traces of BTA on the surface. Hence, it could be deduced
may be pattern dependent, different devices were tested and
that the high contact angle could be attributed to presence of
all were found to have similar trend but in varying degree of
hydrophobic BTA surface film. In another experiment, contact
reduction. A controlled extended run using PC was initiated
angles of DIW drops were measured on the same 6 × 6 cm
and defect trend chart continued to show lower polish scratches
Politex pad after every step of this sequence, namely: dry
in 2 of the devices as shown in Fig. 8.
(original state), after wetting with DIW, after wetting with
BTA and after wetting with PC.as shown in Fig. 11. At the
B. Mechanism of defect formation end of each step, the pad was dried by a hair dryer and 6
A series of experiments were conducted to explain the contact angles were measured at different locations across
mechanism of scratches observed. BTA has been used exten- the pad. The raw data was analysed using JMP® software
sively for copper corrosion inhibitor [3], [4] including CMP and Student T-test was performed to determine if the mean
process [5] but it was also associated with organic residue contact angle of each condition was significantly different from
defects due to its low solubility[6]. Hence, it was one of the each other. Comparison circle, which was the diagrammatic
potential causative agents of scratches in this investigation. representation of the Student’s T-test in this software, was
In the first experiment, a series of Cu blanket wafers were shown on the right of Fig 11. Circles for means that were
first rinsed with BTA solution and post cleaned without PC. statistically different either do not intersect or intersect slightly.
This was compared to similar Cu blanket wafers cleaned At the other end of the spectrum, nested circles would
with PC and a control (not pre-rinsed with BTA). The result show that the means were not significantly different. Result
summarized in Fig. 9 showed that the control and the wafers of this experiment showed that while contact angle of the dry
cleaned with PC has lower contact angle compared to the rest. pad was relatively higher than after wetting, it was not very
TOFSIMS were performed on the two types of copper blanket different statistically. Interestingly, comparison circle of the
wafers, namely a) pre-rinsed with BTA only and b) Pre-rinsed BTA treated pad was nested within the wet pad, showing that
with BTA followed by PC. As shown in Fig. 10, The wafer that their mean contact angles were statistically the same. Since the
did not undergo PC process was shown to have strong relative pad was not previously used, this could be due to presence of
SAN et al.: EFFECT OF CU CMP PAD CLEAN ON DEFECTIVITY AND RELIABILITY 347

Fig. 12. Drag force fluctuation during platen 3 rinsed with different solu-
tions.
Fig. 10. TOF-SIMS for blanket Cu wafer a) Pre-rinsed with BTA only and
b) Pre-rinsed with BTA followed by PC.

Fig. 13. Proposed Mechanism of Polish scratches and effect of PC process.

substrate depending on the topology. However, due to localised

accumulation of BTA on the pad and wafer, an abrasive
Fig. 11. Contact angle of Platen 3 pad vs process sequence. particle may experience increased frictional force which then
embed itself on the softest surface, which would be copper
small chain polyurethane or other hydrophobic by-products in this instance. The chemical PC was effective in removing
that would be expected to be removed during breaking-in the BTA from the interface and hence reduce the frictional
process. The most conclusive observation in this experiment force..
was that contact angle of PC-cleaned pad was statistically
lower than others, as the comparison circle was almost totally C. Reliability Impact
isolated from the rest. This substantiated earlier findings that An additional advantage of the new EC process was that
PC had removed the BTA film on the pad and changed it to a TDDB as shown in Fig. 14, was found to be improved
more hydrophilic state. In the final experiment, motor torque compared to the process without PC. In a typical experiment,
measurements were continuously monitored during blanket break-down time was 6 times longer than POR process. In-
copper wafers polishing step in platen 3 and it was found line SEM inspection and TEM of the new process showed
that drag force was highest during the final rinse step. Three the presence of Cu recess at the trench edge as shown in
liquids were tested for this final rinse, namely BTA (POR), Fig. 15 and AFM verified that the recess in PC process was
DIW and a surfactant. Amongst the three liquid, drag force 25A compared to 15A in POR as shown in Fig. 16. A known
was significantly higher for BTA compared to the other two, as mechanism for TDDB breakdown was the migration of Cu
shown in Fig. 12. This observation is in line with the previous ions between 2 adjacent metal trenches via the interface of
results that BTA induced a high frictional force on wafer dielectric and its SiN hardmask [7]. The edge of the trench is
surface. a particular weak point due to thinner SiN passivation. In the
From these series of experiments, a diagrammatic rep- PC process, since the recesses will be eventually filled by SiN
resentation of the mechanism was proposed as shown in cap, Cu passivation was enhanced and the path for Cu ions to
Fig. 13. In an ideal case, spherical abrasives would roll reach the dielectric/SiN interface will be longer, which may
across wafer surface and remove variable amount of the explain the improved TDDB, as indicated in Fig 17.
348 IEEE TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 26, NO. 3, AUGUST 2013

Fig. 14. TDDB POR vs PC: TBD improve 6 times.

Fig. 16. AFM of trench profile.

Fig. 17. a) POR trench, showing path of Cu ion migration during TDDB
failure b) Trench from PC process.

to build up of BTA during the polish process which in turn

increase the frictional force on wafer/pad interface, possibly
on localized area. An abrasive particle rolling across this
affected area would experience sudden increase in friction and
Fig. 15. TEM of trench profile. lodged itself in the copper, manifesting as a characteristic
comet signature scratch mark. Addition of PC on the pad
was found to reduce BTA and hydrophobicity significantly,
IV. Conclusion which was the explanation of the defect reduction. The new
This paper demonstrated the systematic characterization of process was also found to improve TDDB which could be
a CMP defect and proposed a manufacturable mean of its explained by a change in the trench profile which slow down
control over an extended period in a production environment. the Cu ions migration between adjacent metal lines during high
The root cause was related to hydrophobicity increase due voltage cycles.
SAN et al.: EFFECT OF CU CMP PAD CLEAN ON DEFECTIVITY AND RELIABILITY 349

References [6] T. Cuong, P. Zhang, L. Sun, N. K. Penta, U. R. K. Lagudu, D. Shipp,

and S. V. Babu, “Development of post-CMP cleaners for better defect
[1] T. Y. Teo, W. L. Goh, L. S. Leong, and V. S. K. Lim, “Characteri- performance” ECS Trans. vol. 44, no. 1, pp. 565–571, 2012
zation and reduction of copper chemical–mechanical polishing induced [7] F. Chen, O. Bravo, K. Chanda, P. McLaughlin, T. Sullivan, J. Gill, J.
scratches,” in Proc. SPIE’s Int. Symp. Adv. Microelectron. Manuf., 2003, Lloyd, R. Kontra, and J. Aitken, “A comprehensive study of Low-k
pp. 61–69. SiCOH TDDB phenomena and its reliability lifetime model develop-
[2] T. Y. Teo, W. L. Goh, V. S. K. Lim, L. S. Leong, T. Y. Tse, and L. ment,” in Proc. IEEE 44th Annu. Int. Symp. Reliability Phys., Mar. 2006,
Chan, “Characterization of scratches generated by a multiplaten copper pp. 46–53.
chemical–mechanical polishing process,” J. Vac. Sci. Technol. B, vol.
22, no. 1, pp. 65–69, Jan.–Feb. 2004.
[3] G. Notoya and G. W. Poling, “Topologies of thick Cu-benzotriazole on
copper,” Corrosion-NACE, vol. 32, no. 6, pp. 216–223, 1996. Leong Lup San received the M.Sc. degree in chemistry and M.Sc. degree
[4] G. Notoya and G. W. Poling, Corrosion (Houston), 1976, pp. 21–216. in management of technology from the National University of Singapore,
[5] R. Carpio, J. Farkas, and R. Jairath, “Initial study of copper Singapore. He is a Senior Member of Technical Staff at the Advance Module
CMP slurry chemistries,” Thin Solid Films, vol. 266, pp. 238–244, Technology/CMP Department, GLOBALFOUNDRIES Singapore, Singapore.
1995. He has 15 years experience in CMP and process development, and transfer.