[ieee 2010 34th international electronics manufacturing technology conference (iemt) - melaka,...
TRANSCRIPT
34th
International Electronic Manufacturing Technology Conference, 2010
Investigations of the Effects of Blade Type, Dicing Tape, Blade Preparation and Process Parameters
on 55nm Node Low-K Wafer
Koh Wen Shi, Yow K.Y, Rachel Khoo
Freescale Semiconductor Malaysia Sdn. Bhd.
2, Jln. SS8/2, Free Industrial Zone Sungei Way,
47300 P.J. Selangor, Malaysia
Abstract
This paper presents an investigation of the effects of
blade type, dicing tape, blade preparation and the key
process parameters optimization on improving topside ILD
peeling (thicker scribe structures) and chipping for 55nm
low-k wafer. An appropriate dicing blade selection, blade
preparation / conditioning methodology and dicing tape
selection plays an important role in developing a robust saw
process. As such, experimental studies were conducted
under varying Z1 spindle rotation, Z1 cut depth into Si as
well as the blade type property variation as the input factors,
in order to improve the ILD peeling and die chipping. The
settings of machining parameters and blade types were
determined by using the design of experiment (DOE)
techniques and the critical process parameters and materials
were analyzed statistically by using the analysis of variance
(ANOVA). Dicing tape property variations (PO-base or
PVC-base) as well as the blade preparation methodology
posed some influences on the overall dicing quality, such as
die backside chipping, die removal performance, ILD
peeling and die topside chipping. SEM imaging and optical
visual inspection were conducted to validate the impacts of
the ILD peeling / chipping on post-processed low-k wafers.
A thorough quantification and categorization of ILD peeling
and chipping on heavy metallization at the saw scribe
structures were described. As part of the recommendation
for future works, a different approach in dicing technolgy,
namely laser grooving was proposed to eliminate ILD
peeling and chipping. In conclusion, the optimized dicing
recipe for 55nm node low-k wafer suggested by the DOE
model are: (1) a thinner PO-base dicing tape, (2) a dicing
blade with higher diamond concentration and finer grit size,
(3) blade preparation / conditioning done with SiC board and
(4) processing at lower spindle rotation and deeper cut depth
are much preferred. The overall dicing responses and cutting
quality has improved and is better compared to current
production recipe.
1. Introduction
A few years ago, some articles had presented some of the
issues and challenges faced by the wafer dicing process due
to the new emerging low-k / ULK wafer technology. There
is a trend observed in the semiconductor industry which
includes copper and low-k / ULK dielectric film in wafer
fabrication process for better electrical, mechanical and
thermal performance. The aluminium interconnects will be
replaced by copper and the traditional SiO2 will be replaced
by low-k / ULK dielectric materials. The conversion from
traditional oxide materials to advanced materials has
actually made significant improvements on the device
performances and reliability, the die feature can be shrunk
dow. However, the dielectric constant (k is 2.0 or below) is
fairly small for low-k / ULK dielectric film compared to
traditional oxide (k is 4.0). [1] The impact is strong when it
comes to wafer dicing. As the k-value of the dielectric film
is reduced, the mechanical property usually will deteriorate
because of the increase in porous volume introduced into the
materials.
The chemical, thermal, and mechanical behaviours of
low-k / ULK wafer are quite different from traditional
materials. Some problems are being encountered in adapting
this new wafer technology. Low-k / ULK porous wafers
tend to have ILD / metal layers peeling and chipping during
the wafer dicing process because of the brittleness and
fragility of the top layers which consist of oxides, metals and
low-k silicon. The high occurrences of critical dicing defects
could impose challenges for packaging assembly yield,
product performance, quality and reliability issue. Therefore,
a number of assessments were conducted in order to address
and improve the ILD peeling and die chipping issue. Firstly,
dicing tape selection and the effects on silicon backside
chipping, the adhesive strength of the tape to the silicon
during dicing, and the impacts of die removal after UV light
exposure. Secondly, dicing blade selection to understand the
effects of blade properties to the ILD / metal layer peeling
and die topside chipping performances. Thirdly, a study was
done on blade dressing / conditioning and the impacts on
improving ILD peeling and chipping quality. Fourthly, to
optimize the critical dicing process parameters such as
spindle rotation, cut in depth, table speed, and scribe pattern
variations for dicing defects reduction. All evaluations were
performed on 55nm low-k wafers, and expected to establish
a robust wafer dicing process with good and promising
cutting quality. In the next session, we will discuss about the
research methodology and experimentation.
2. Research Methodology and Experimentation
In this session, we will describe the methods and
experiments that were performed to understand and
investigate the effects / impacts of various dicing tapes,
cutting tools (dicing blades), blade dressing methods and
dicing process variables to the ILD peeling and die chipping
performance.
2.1 Experimental Wafer Information
The detailed technical wafer information that was chosen
in this study is shown in Table I. The selected test vehicle in
this project is a pizza mask wafer which consists of multiple
device designs on a single wafer. There is a total of 3
different die devices in each wafer, which needs to be
singulated for ILD peeling and die chipping assessments.
The test vehicle designed and fabricated with 55nm wafer
process technology, the typical cross sectional view of the
metal layer stacks is shown in Fig. 1. The wafer surface was
electroplated with thick metallization (1-3um) on top of the
aluminium bonding pads and test patterns on the saw street.
The metal compounds on the thick metallization contains a
composite of nickel, gold and palladium elements. Dicing
through the thick metal composite tends to result more ILD
peeling and chipping defects due to the hard and brittle
structure in nature. The following dicing assessments were
conducted to develop a robust saw process capability on
dicing through the complicated Cu metals, low-k dielectric
films and thick electroplating metallization.
Table I Test vehicle information
Wafer Technology CMOS 55nm low-k wafer
Die Size Small: 3mmx3mm
Medium: 6mmx7mm
Large: 10mmx7mm
Wafer Diameter 300mm
Wafer Thickness 280um
Saw Street Width 80um
Electroplating Thickness 1-3um
Fig. 1 Cross sectional view of 55nm low-k metallization
2.2 Dicing Tape Selection
Two types of UV dicing tapes were evaluated, which are
A tape and B tape type. Table II shows the comparison
matrix between the 2 dicing tapes. The variations of tape
properties (tape thickness, base film material type and
adhesive strength) could affect the die backside chipping
performance and effectiveness for die pick up process (die
removal) after UV light exposure.
Table II UV dicing tape properties and variations
Wafers are mounted on the UV dicing tapes respectively
(both A and B). The cutting tools, the saw machine and saw
process parameter applied are fixed at constant value
(nominal parameter), as illustrated in Fig. 2. The dicing
responses to-collect are die backside chipping measurement.
As part of the evaluation, the ease of die removal (after UV
light exposure) during die bonding will be assessed, it is to
verify the tape’s adhesive strength to prevent die fly-off
issue.
Fig. 2 The general process flow for wafer mounting and
dicing
2.3 Blade Dressing Condition
To successfully complete a dicing project with good
cutting quality, blade preparation / conditioning is one of the
critical factors (dicing blades, process parameters, and
dicing tapes) one needs to consider. Blade dressing is
important, for the reason, to grind off the excess bonding
material and to expose the diamond particles for cutting. The
traditional tools used to condition / season the dicing blade
are silicon wafer. However, the total completion time to
prepare a blade for wafer cutting usually takes a few hours
(very long). The new method proposed in this experiment is
the use of a dressing board, which is of a silicon carbide-
based (SiC) material. The advantage of using SiC dressing
board is that the total time required to prepare a blade for
wafer cutting is fairly short, which takes less than an hour.
Table III shows the experiment matrix conducted by using
both new and traditional blade dressing method and to
investigate its effects on the cutting quality. The SiC
dressing board was mounted on the dicing tape and a visual
inspection will be done on the dressing board after blade
dressing as shown in Fig. 3. The ILD peeling / delamination
are the main dicing response collected for statistical data
analysis and interpretation.
Table III Experiment matrix for blade preparation / dressing
using Si wafer and SiC dressing board
Fig. 3. (i) Dressing board mounted on dicing tape and (ii)
Cutting lines on the board after completion of blade dressing
Wafer mounted on UV
tape
Wafer dicing
Wafer
Wafer
Dicing tape
Top View
75mm
75mmSide View
1 mm*0.090mm
Dressing board mounted on the
dicing tape
(i) (ii)
(i)
Electroplating
Metallization
Factor + 0 -
Spindle RPM High Medium Lo
Cut depth into Si High Medium Lo
Blade Type (Categorical) Blade Type C Nil Blade Type D
2.4 Dicing Blades Selection and Dicing Process Variables
Choosing a suitable dicing blade is very important. The
key elements that one needs to consider when selecting a
new blade are diamond grit size, diamond grit concentration
and bonding material. Improper dicing blade selection as
well as non-optimized dicing process parameters eventually
will cause poor cutting quality. In order to establish a new
dicing recipe capable of processing CMOS 55nm low-k
wafer, a 3-factor full factorial DOE model was conducted in
order to optimize the topside ILD peeling and chipping
performance. Sawing quality with existing parameters were
not satisfactory and saw process parameter optimization was
needed for cutting through the unfriendly aluminium scribe
pattern (which has a newly added ~3um thick electroplating
metallization). The DOE model consists of 3 variables,
blade type, spindle rotation and cut depth into Si, see Table
IV. A comparison of the properties between blade type C
and D is shown in Table V.
Table IV 3-factor DOE matrix
Table V Blade types and properties comparison
Blade Type Blade Type C Blade Type D
Diamond
Concentration High Low
Diamond Grit Size Larger Smaller
Bonding Material Standard Soft
Blade Exposure 760-890um 640-760um
Blade Thickness 25-30um 25-30um
3. Results and Discussion
There is a total of three sub-session in this part which
will be discussed in detail for all the results collected from
the experiments. First, the effects of different dicing tapes to
the die backside chipping performance. Second, the effects
of different dressing tools to the topside ILD peeling
performance. Third, the effects of blade types and dicing
process parameters on reducing ILD peeling and die
chipping found on cutting through the unfriendly scribe
structures (higher metal density , harder and brittle
metallization impose challenges for wafer dicing).
3.1 Effects of Dicing Tape Variations
Die back and die side chipping performances: Table VI
summarized the die chipping data and performances across
the different dicing tapes (UV) used. Chipping data
calculated in mean value based on sample size, N=50. A few
observations have benn noted based on the data and the
variability chart: first, PO-based tape (harder tape) results
better chipping data compared to PVC-based tape (softer
tape). Second, thinner tape is favourable with smaller die
chipping size. Both tape B and Enhanced-tape B show good
and comparable die chipping size, even with differences in
adhesive strength after UV light exposure. Thicker tape
thickness and softer base film (tape A) tend to produce
bigger chipping size.
Table VI Chipping data (average) across dicing tape
variations
Tape Type Tape
Thickness
Base
Film
Adhesive
Strength
(after UV)
Dieback
Chip
(um)
Dieside
Chip
(um)
Die
Fly
A 110 PVC High
(Strong)
27.5 16.7 No
B 90 PO Low
(Weak)
4.8 6.0 Yes
B
(Enhanced)
90 PO High
(Strong)
10.1 9.0 No
Fig. 4. Variability chart for die side chipping
Die-pick up performances after UV light exposure: The
adhesive strength for tape B is fairly low after UV exposure
compared to tape A and enhanced-tape B. Therefore, die fly
issue is possible to happen on tape B during die bonding (see
Fig. 5). This is not acceptable although the die chipping data
is passable and looks good. This means that we need to have
sufficient adhesive strengh after UV exposure so the die will
not be so easy to fly off before die bonding, see Fig. 6 about
the description of die fly issue for tape B. Therefore, tape B
is not suitable for use due to the weak adhesive strength after
UV and can cause high yield loss due to die fly issue.
Based on the results, enhanced-tape B is more favourable
in this project because of the small die backside chipping
(10um in average) and stronger adhesive strength after UV
resulting to no die fly issue. A harder (PO-based) and
thinner tape is more preferred than a softer and thicker tape
for better die backside chipping performance.
Run
Order Blade Type Spindle RPM Cut Depth into Si
1 D Hi Lo
2 C Lo Lo
3 D Hi Hi
4 C Lo Hi
5 D Lo Hi
6 C Med Med
7 D Lo Lo
8 C Ho Lo
9 D Med Med
10 C Hi Hi
Chipping
As the needle retracts, vacuum is pulling down
the tape and adjacent die. This motion separates the adjacent dies from the tape, causing die throw.
Needle pushes tape and die up. Adjacent dies
get push up as well as since the needle is blunt with radius of 350um and does not pierce through the tape. As the pick up tip pulls the die up, the die starts to separate from the tape.
This will not happen to the A and Enhanced-B tape as the after UV exposure, the adhesive strength is still stronger
Tape B Tape B (Enhanced)
Fig. 5 Tape B observed die fly off issue due to weak
adhesive strength after UV exposure
Fig. 6 Illustration of die fly issue during die bonding due to
weak adhesive strength after UV exposure
3.2 Effects of Blade Dressing Tool Variations
Conditioned blade profile inspection: Table VII shows
the resulted blade profile (SEM mircograph) by using
different blade dressing tools, i.e. Si wafer versus SiC board.
The total time required to prepare the blade using SiC board
is shorter than standard Si wafer dressing. The tip of the
board dressed blade looks more rounder than the Si wafer
dressed blade. SEM imaging was performed to inspect the
physical characteristics of the dressed blade profile as well
as the distribution of the exposed diamond grit after
dressing. Si wafer dressed blade (current dressing approach)
resulted to the blunt shape of the exposed diamond grit after
longer hours of dressing time, the distribution / density of
the diamond grit that exposed is high. SiC board dressed
blade results sharper diamond grit and the distribution /
density of the diamond grit that exposed is small. The
desired blade profile is the one that use SiC board for
dressing, even though the observed diamond grit coverage is
actually lower. When the diamond coverage / exposure is
low (new method), it means the SiC board is more capable
than the Si wafer to grind off the excessive bonding
materials and continue to remove the old diamond grits and
regenerate more fresh diamond grit which is embedded
inside the bonding material. This is an effective dressing
method to use SiC and it can enhance the overall cutting
power, unlike the old method which with high diamond
coverage (blunt shape). The blunt and old diamond grits can
not be removed easily and it had blocked the embedded new
diamond grits to expose.
Table VII Conditioned blade profile inspection Blade
Preparation Current Method New Method
Blade Dressing
Tools Si Wafer SiC Board
Time Required A few hours Less than an hour (16x
reduction in time)
Blade Profile
(Cross Sectional
View)
Blade Profile
(SEM Imaging)
Topside ILD peeling performance: Fig. 7 shows the
dicing effect on ILD peeling response by various scribe
patterns and different blade dressing methods. Based on the
optical photos and ANOVA statistical analysis (Fig. 7 and
Fig. 8), both Si wafer dressing and SiC board dressing
results are comparable with no significant difference in ILD
peeling response. From the boxplots and variability chart,
SiC board dressing is just slightly better than Si wafer
dressing in term of the ILD peeling size (mean data), see
Fig. 8. SiC board dressing performed better than the
traditional Si wafer dressing. Therefore, SiC board dressing
is the recommended method for use due to the short blade
preparation time, is lower in cost and produces good ILD
peeling response.
Fig. 7 Optical inspection on topside ILD peeling with
respect to different blade dressing method
Fig. 8 Variability chart for ILD peeling size across different
scribe patterns. SiC board dressing results to smaller ILD
peeling size (mean data) compared to Si wafer dressing.
Before Saw SiC Board Dressing (New)Si Wafer Dressing (Current)
Scribe Pattern Type 1
Scribe Pattern Type 2
Tip: sharp
Tip: rounder
Peelin
g S
ize (
um
)
10
15
20
25
30
35
F40 D
ressed
Bla
de
Mir
ror
Wfr
Dre
ssed B
lade
F40 D
ressed
Bla
de
Mir
ror
Wfr
Dre
ssed B
lade
F40 D
ressed
Bla
de
Mir
ror
Wfr
Dre
ssed B
lade
F40 D
ressed
Bla
de
Mir
ror
Wfr
Dre
ssed B
lade
Curr
ent
CO
N/M
1 S
PM
Curr
ent
Sta
ck S
PM
Old
CO
N/
M1 S
PM
Old
Sta
ck S
PM
Blade Condition within Scribe Pattern
10
15
20
25
30
35
Peelin
gS
ize (
um
)24.0
8?.6
78665
F40 D
ressed B
lade
Mirro
r W
fr D
ressed B
lade
F40 Dressed Blade
Blade
Condition
Peeling Mode
Mild peeling range from10% -50%
Severe peeling >50%
Comments
No peeling or <10% peelingMode 1
Mode 2
Mode 3
Optical Photo
0
20
40
60
Peelin
g
Mode 3
.5
1.5
?1.2
944
0
20
40
60
80
Chip
pin
g
Mode 2
2.8
21508
?9.2
7598
00.2
50.7
51
Desirabili
ty
0.9
54474
35
40
45
50
55
35
Z1 Rotation
(kRPM)
203O
127J
203O
Blade Type
50
100
150
176.66295
Z1 Cut Depth
into Si (um)
0
0.2
5
0.5
0.7
5 1
Desirability
Lo Hi C Type
DType
Lo Hi
Die Corner Peeling
Die Corner Peeling Mode Distribution vs. Run Order
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11
Run Order
Pe
eli
ng
Mo
de
D
istr
ibu
tio
n
Mode P1
Mode P3.0
Mode P3.5
Control
Sample size: ~240 x
11cells Mode 3.5
Peeling Mode Distribution vs. Run Order
0%
20%
40%
60%
80%
100%
RO
1
RO
2
RO
3
RO
4
RO
5
RO
6
RO
7
RO
8
RO
9
RO
10
RO
11
Run Order
OP
M S
GP
C P
ee
lin
g
Mo
de
Dis
trib
uti
on
Mode P3.5
Mode P3
Mode P2
Control
Die Side Peeling
Die Chipping Mode Distribution vs. Run Order
0%
20%
40%
60%
80%
100%
RO
1
RO
2
RO
3
RO
4
RO
5
RO
6
RO
7
RO
8
RO
9
RO
10
RO
11
Run Order
Die
Ch
ipp
ing
Mo
de
D
istr
ibu
tio
n
Mode C2
Mode C1
Mode C0
Control
RO8 is the best cell
with less mode 3.5 peeling, across all the 10 DOE run orders
RO4 is the best
cell with lesser mode 3.5 peeling
RO4 is the best cell
with lesser mode 2 die chipping
Die Chipping
Sample size: ~240 x
11cells
Sample size: ~240 x
11cells
3.3 Effects of Blade Types and Dicing Process Variables
Die corner, ILD peeling and die chipping performance:
A total of 10 runs were conducted and the input process
factors included in this experiment are dicing blade types,
Z1 spindle rotations and Z1 cut depth into Si (see Table VI).
This is part of our efforts to define an optimize saw recipe
for hard and heavy metallization (via electroplating). The
dicing responses to-collect for all the 10 runs are die corner
peeling, dieside ILD peeling and die topside chipping. Table
VIII shows the general guideline to follow for ILD peeling
charaterization. The distribution of ILD peeling and die
chipping across all the run order is shown in Fig. 9.
According to the bar chart in Fig. 9, RO4 (lower Z1 rotation,
deeper cut depth into Si and using blade type C) turns out to
be the best run in this DOE for ILD peeling and die chipping
performance.
Table VIII General guidelines for ILD peeling
characterization
Fig. 9 ILD peeling and die chipping bar chart by run orders
In order to reduce the ILD peeling and chipping size, the
overall response optimizer recommends to use blade type C,
as well as a lower spindle rotation and deeper cut depth into
Si is preferred / desirable, see Fig. 10. The desirability for
this prediction profiler is fairly reliable and good, which
reads at about 95%.
Fig. 10 ILD peeling and die chipping responses optimizer
Figure 11 shows contour plots for both peeling and
chipping responses. Based on the contour plots, the best
defined saw recipe had been shifted to a new location (i.e at
upper left corner point) from the current production recipe
(i.e. somewhere close to lower right side of the plot). Based
on all the statistical data and analysis, the optimal saw recipe
to reduce the ILD peeling and chipping suggested a lower
Z1 spindle rotation, a deeper Z1 cut depth into Si and the
dicing blade type C (higher diamond concentration, fine
diamond grit size is preferred).
Fig. 11 Contour plots for ILD peeling and die chipping
Z1 rotation
Z1 cut depth
5550454035
175
150
125
100
75
50
Blade 203O
Hold Values
>
–
–
–
–
< 10
10 20
20 30
30 40
40 50
50
3.5
mode
Peeling
Contour Plot of Peeling mode 3.5 vs Z1 cut depth, Z1 rotation
DOE Defined
(OPM Fado)
Current Recipe
(Control Fado)
Z1 rotation
Z1 cut depth
5550454035
175
150
125
100
75
50
Blade 203O
Hold Values
>
–
–
–
–
< 10
10 20
20 30
30 40
40 50
50
3.5
mode
Peeling
Contour Plot of Peeling mode 3.5 vs Z1 cut depth, Z1 rotation
DOE Defined
(OPM Fado)
Current Recipe
(Control Fado)
Z1 rotation
Z1 cut depth
5550454035
175
150
125
100
75
50
Blade 203O
Hold Values
>
–
–
–
–
–
< 10
10 20
20 30
30 40
40 50
50 60
60
mode 2
Chippimg
Contour Plot of Chippimg mode 2 vs Z1 cut depth, Z1 rotation
DOE Defined (OPM Fado)
Current Recipe
(Control Fado)
ILD Peeling Contour Plot
Die Chipping Contour Plot
Lo
Lo
Hi
Hi
Hi
Hi
Lo
Lo
34th
International Electronic Manufacturing Technology Conference, 2010
4. Future Work and Recommendations
Based on all the experimentation studies that was
performed, the results show that blade dicing process alone
is not possible to produce zero dicing defects. Dicing
through the unfriendly scribe patterns with a higher metal
density still exhibits metal layer delamination, ILD peeling
and die chipping. Hence forth, on future works and
recommendations, laser grooving could be considered as
another option in order to produce a true dicing defect-free
process.
Unlike traditional blade dicing, laser grooving is a
thermal energy based process; with no direct tool-to-work
piece contact. When a focused laser beam with high energy
density contacts the Si wafer, thermal energy is transferred
and absorbed by Si; heat will transform to the low-k ILD
metal layers and melted into molten and the solids are
vaporized, which can be disposed and removed by the flow
of air pressure. This new advanced dicing process
technology can be an opportunity to resolve the dicing
challenges with more upcoming low-k and ULK wafers.
5. Conclusions
Based on the experiments and result findings, we may
conclude that the dicing blade and its blade dressing
preparation, dicing tape selection, and dicing process
parameters plays an important role in effecting ILD peeling /
delamination, die topside chipping and die backside
chipping. For a better die backside chipping performance, a
PO-base film, thinner tape thickness, a stronger tape with
high adhesion strength are more desirable than a PVC-base
film. A proper blade preparation is required, sufficient blade
dressing / conditioning time is needed in order to grind-off
the excess bonding material and therefore able to expose
diamond grit for cutting (which has been sharpened by the
truing effect). SiC dressing board is more effective than Si
wafer dressing in terms of producing sufficient diamond
coverage being exposed for cutting. The overall cutting
quality is comparable and acceptable for both SiC board
dressing (new method) and Si wafer dressing (traditional
method). SiC board dressing is selected because of the total
time required for the blade preparation which is shorter and
faster than traditional Si wafer dressing. To optimize and
reduce ILD peeling and die chipping on unfriendly scribe
patterns (with heavy metalization), the model suggests to
utilize a dicing blade with higher diamond concentration and
a finer grit size is preferred. Therefore, to obtain good
topside cutting quality, therefore the wafer needs to be
processed at a lower spindle rotation and a deeper cut depth
into Si.
Acknowledgments
The authors would like to thank Freescale
Semiconductor (M) Sdn. Bhd. for their support in this 55nm
low-k wafer dicing process development project. Also my
sincere thanks to all technical and analysis support from
TSO-PAE members in KLM.
References
1. Wang ZhiJie, Wang, S., Wang, J.H., Lee, S., Yao Su
Ying, Han, R., Su, Y.Q., “300mm Low-k Wafer Dicing
Saw Study”, Electronic Packaging Technology, Sept.
2005, pp. 262-268.