PROCESS IMPROVEMENT OF THE SILICON WAFER LASER DICING FOR FRONT SIDE MEANDERING

pigsabusiveΗλεκτρονική - Συσκευές

29 Νοε 2013 (πριν από 3 χρόνια και 6 μήνες)

377 εμφανίσεις

PROCESS IMPROVEMENT OF THE SILICON WAFER LASER DICING FOR FRONT
SIDE MEANDERING


Suwitchaporn Witchakul
, Kasetsart University, Thailand


Poom Tipmonta
, Kasetsart University, Thailand


Saroj Aungsumalin
, Kasetsart University, Thailand


Pornthep
Anussornnitisarn
,
Kasetsart University,

Thailand


ABSTRACT


This study concentrates only silicon wafer laser dicing. It has the major problem which is
meandering. According to the very narrow saw street in silicon wafer,

only 15 μm,

the saw kerf
from laser dicing process exceed into chip circuit that will be the defect.
The objective of the
research study is to find the
optimal

settings
for
silicon wafer laser dicing process

in order to
achieve the target of

3 μm
maximum fluctuation
of meandering.


Keywords:


meandering; laser dicing; design of experiment
; fractional experiment
.


INTRODUCTION


Nowadays, competition in the manufacturing industry is increasing in both domestic and
international markets. Significant factors that make a
manufacturing company competitive in
the worldwide market are quality, delivery and cost. It is generally agree that the quality and
cost represent an essential factor in business’s growth. Therefore, the thought of quality and
cost are highly important.


With ongoing miniaturization of IC’s the ratio of the functional wafer surface gets lower and
lower. The non functional wafer area for the next generation of RFID chips rises up to 30% with
smallest conventional saw lane design available. Therefore new chi
p separation technologies
were evaluated to enable a significant reduction of the saw lane width and thus a significant
increase of the number of dies per wafer.


Currently, the conventional concept to separate chip is blade dicing, which has saw blade to
saw on wafer saw lane. However, the capability of blade dicing is limitation. This method can
support at 50

μm

saw lane width in minimum. And with mechanical sawing on silicon might
lead to generate the chipping or crack on wafer as well.


Therefore, lase
r dicing machine
, the high accuracy and high capability machine used to separate
wafer,
is
applied

to improve quality and
reducing cost
. That’s can support the silicon wafer
which are 15
μm

saw lane width.
The concept of laser dicing is

laser diode will generate the
pulse to destroy silicon structure and breaking the wafer again after laser pulse. With t
his
concept there is no chip

or crack happen compare to blade dicing method. Since this is quite
new concept machine so, sometimes the u
nforeseen problem might be happen.


As mentioned above there are some serious defects happened. Meandering is one
of
the most
critical. The characteristic of defect is look like wavy kerf. That’s still acceptable if the

maximum
fluctuation is not over than

3

μm.


In fact we faced about
4 to 5

μm.


Since the new type of
wafer, the saw lane width was decreased to 15

μm

to gain more and more chip in one wafer.
That’s mean this kind of defect can be destroyed of whole wafer. And those defect silicon
wafers will

be scraped.


As a conclusion, the meandering was a major problem in laser dicing process. The fault tree
analysis is applied to find the potential root cause of meandering. Next, the Design of
Experiment (DOE) is applied to reduce this kind of defect.

Aft
er the experiments are conducted,
the best setting is selected. Next, the confirmation experiment are conducted in order to
confirm the settings that the maximum fluctuation us not over 3

μm
.



BACKGROUND


In the past silicon wafers have been diced by
using blade saw. This method is called blade dicing
method
.
Kumagai et

al. (2007) the
blade
dicing process
cause
s undesirable mechanical
vibrations and stress in the wafer

which make
damage in the wafer consisting of cracks,
chippings, etc.
When the dicing

processing
speed increase
s or wafer thickness decreases, the
defects increase

as well.
However, the semiconductor wafers are becoming thinner and thi
nner
in advanced manufacturing.
Moreover, the processing speed is important
characteristics for the
produc
tivity
. In order to
solve

blade dicing problems,
laser
dicing methods have been
developed

(Hermanns, 2005).
Stealth dicing method is one
kind
of laser dicing methods
(Fukuyo
et al, 2005)
.
T
he laser beam is
focused in
to
the interior of the work wafer as we
ll as
the laser
processing only occurs inside the wafer. The coolant and washing water are not
necessary

at all
for this stealth dicing method since it does not cause any frictional heat and debris. The stealth
dicing method has three advantages as follows

(Kumagai et al., 2007)
.


1)

High
-
speed dicing for thinner wafers without any chipping.


2)

No debris contaminants that are caused by blade dicing or laser ablation;


3)

Completely dry process.



METHODOLOGY


In order to search for
the optimal setting of the laser dicing process, the following steps are
performed:


1.

Determine the

potential

factors affected to the fluctuation of meandering
by using fault
tree analysis
.


2.

C
onduct the
screening

experiment
with 3 replications
to eliminate the factors
which is not
affected the fluctuation of meandering.


3.

Conduct the
following

experiment to determine the best settings which
give the smallest
maximum
fluctuation of meandering.

4.

Conduct the c
onfirmation experiment

in order to confir
m the settings that the maximum
fluctuation us not over 3

μm.


RESULTS




By brainstorming with process and manufacturing engineers, four major areas 1) product 2)
machine 3) process and 4) materials, are used as the focused area for fault tree developme
nt.


The fault tree analysis for meandering is shown in figure 1.


Figure 1
:

Fault tree analysis for meandering
.



Figure 1
:

Fault tree analysis of meandering
.


For the product, there is only one product type for experiment and they are from the single
source. For machine, machine was calibrated to make sure that it is ready for running. For
materials, there is only one type of backgrind tape, dicing tape and wafe
r substrates. Therefore,
the potential factors affected to the fluctuation of meandering are from process which are
scanning height (factor A), scan power 1 (factor B), scan power 2 (factor C), beam shape (factor
D) and scanning speed (factor E). These fac
tors are considered at two levels each. The levels for
all factors are shown in table 1.


Table 1
:

Factors and Levels of
5
2
factorial experiment
.


Factors

Level

Low

High

scanning height (A)

10,9

16,3

scan power 1 (B)

0.12

Watt

0.36

Watt

scan power 2 (C)

0.18

Watt

0.54

Watt

beam shape (D)

Ultrathin

Thick

scanning speed (E)

100

mm/s

300

mm/s



The results are shown in table 2
.



Table 2
:

Results of
5
2
factorial experiment

with 3 replications
.


Scanning
Height

(A)

Scan

Power 1

(B)

Scan

Power 2

(C)

Beam

Shape

(D)

Scanning
Speed

(E)

REP I

REP II

REP III

Average

10,9

0.12

0.18

Ultrathin

100

3.42

3.11

3.43

3.32

16,3

0.12

0.18

Ultrathin

100

3.54

3.03

3.5

3.36

10,9

0.36

0.18

Ultrathin

100

2.67

3.43

2.83

2.98

16,3

0.36

0.18

Ultrathin

100

3.02

3.23

2.8

3.02

10,9

0.12

0.54

Ultrathin

100

2.45

3.76

2.4

2.87

16,3

0.12

0.54

Ultrathin

100

2.98

3.01

2.34

2.78

10,9

0.36

0.54

Ultrathin

100

2.04

1.54

1.45

1.68

16,3

0.36

0.54

Ultrathin

100

1.52

2.2

1.75

1.82

10,9

0.12

0.18

Thick

100

6.2

4.4

5.67

5.42

16,3

0.12

0.18

Thick

100

6.34

5.87

5.34

5.85

10,9

0.36

0.18

Thick

100

4.65

5.67

4.63

4.98

16,3

0.36

0.18

Thick

100

5.52

5.21

5.22

5.32

10,9

0.12

0.54

Thick

100

5

4.53

5.05

4.86

16,3

0.12

0.54

Thick

100

4.67

4.56

4.63

4.62

10,9

0.36

0.54

Thick

100

4.01

4.12

4.24

4.12

16,3

0.36

0.54

Thick

100

4.03

3.96

4.17

4.05

10,9

0.12

0.18

Ultrathin

300

3.21

3.21

3.59

3.34

16,3

0.12

0.18

Ultrathin

300

3.48

3.49

3.21

3.39

10,9

0.36

0.18

Ultrathin

300

2.66

2.93

2.95

2.85

16,3

0.36

0.18

Ultrathin

300

3.25

2.81

2.94

3

10,9

0.12

0.54

Ultrathin

300

2.67

2.99

2.89

2.85

16,3

0.12

0.54

Ultrathin

300

2.74

2.85

2.64

2.74

10,9

0.36

0.54

Ultrathin

300

1.94

1.85

1.99

1.93

16,3

0.36

0.54

Ultrathin

300

2.35

2.02

2

2.12

10,9

0.12

0.18

Thick

300

5.39

5.69

4.61

5.23

16,3

0.12

0.18

Thick

300

5.17

5.39

5.17

5.24

10,9

0.36

0.18

Thick

300

5.04

5.08

4.92

5.01

16,3

0.36

0.18

Thick

300

4.73

4.78

5.49

4.98

10,9

0.12

0.54

Thick

300

5.19

5

5.18

5.12

16,3

0.12

0.54

Thick

300

4.61

5.02

5.29

4.97

10,9

0.36

0.54

Thick

300

4.02

4.51

3.84

4.12

16,3

0.36

0.54

Thick

300

3.93

3.53

4.58

4.01


The Analysis of Variance is applied to analyze the results as shown in table 3
.


Table 3
:

ANOVA table of
5
2
factorial experiment

with 3 replications
.


Source of
Variation

Degree of
Freedom

Sequential

Sum Square

Adjusted

Sum Square

Mean

Square

F

P
-
value

A

1

0.0360

0.0360

0.0360

0.31

0.578

B

1

9.2877

9.2877

9.2877

80.22

0.000

C

1

14.9468

14.9468

14.9468

129.10

0.000

D

1

107.8232

107.8232

107.8232

931.29

0.000

E

1

0.0011

0.0011

0.0011

0.01

0.924

A*B

1

0.0504

0.0504

0.0504

0.44

0.511

A*C

1

0.2035

0.2035

0.2035

1.76

0.189

A*D

1

0.0054

0.0054

0.0054

0.05

0.830

A*E

1

0.0273

0.0273

0.0273

0.24

0.628

B*C

1

1.4652

1.4652

1.4652

12.66

0.001

B*D

1

0.0294

0.0294

0.0294

0.25

0.616

B*E

1

0.0063

0.0063

0.0063

0.05

0.816

C*D

1

0.0077

0.0077

0.0077

0.07

0.797

C*E

1

0.4760

0.4760

0.4760

4.11

0.046

D*E

1

0.0782

0.0782

0.0782

0.68

0.414

Error

80

9.2622

0.1158




Total

95

143.7066






R
-
Sq(adj) = 92.35%






2
1
2
1
2
1
2
1
5
4
3
5
4
3
5
4
3
5
4
3
A
B
C
D
E
1
2
A
1
2
B
1
2
C
1
2
D
I
n
t
e
r
a
c
t
i
o
n

P
l
o
t

f
o
r

R
e
s
p
o
n
s
e
F
i
t
t
e
d

M
e
a
n
s

Figure 2: Interaction plot for response
5
2
factorial experiment.


According to table 3, the factors which affect to fluctuation of meandering is factor B,C, D

and
interaction of factor B and C by using 0.01 significance level.

For the following experiment,
factor B and C is conducted
with four levels. However, factor D is conducted with two levels.

According to figure 2, interaction plot of factor D with factor

A, B, C and E at level 1 (ultrathin)
give the smaller responses than level 2 (thick). Therefore, the following experiment is
conducted with
two levels which are

ultrathin and thin for factor D.
Next the
2 1
4.2

factorial
experiment with 5 replications is conducted. The levels
for each factor

are shown in table 4.


Table 4
:

Factors and Levels of
2 1
4.2

factorial experiment with 5 replications
.


Factors

Level

1

2

3

4

scan power 1 (B)

0.12

0.24

0.36

0.
48

scan power 2 (C)

0.18

0.36

0.54

0.
72

beam shape (D)

Ultrathin

Thin




For the factors which are not
affected

to the fluctuation of meandering, they are fixed at the
appropriate value. Scanning height

(factor A)

is fixed at
the middle value
of

10,9 and 16,3

which
is 13,6. Scanning speed (factor E) is fixed at 300
mm/s
because this is the maximum value of
scanning speed which gives the highest productivity.

The results are shown in table 5
.



Table 5
:

Results of
2 1
4.2
factorial experiment with 5 replications
.


Scanning
Height

(A)

Scan
Power 1

(B)

Scan
Power 2
(C)

Beam
Shape

(D)

Scanning
Speed

(E)

REP

I

REP II

REP
III

REP
IV

REP V

Average

13,6

0.12

0.18

Ultrathin

300

3.42

3.56

2.54

4.02

2.78

3.26

13,6

0.12

0.18

Thin

300

3.96

3.54

3.63

3.42

2.98

3.51

13,6

0.12

0.36

Ultrathin

300

2.87

2.89

3.25

3.02

2.87

2.98

13,6

0.12

0.36

Thin

300

2.86

2.84

2.57

2.89

3.05

2.84

13,6

0.12

0.54

Ultrathin

300

2.59

2.65

2.54

2.36

2.58

2.54

13,6

0.12

0.54

Thin

300

2.56

2.56

2.87

2.69

2.61

2.66

13,6

0.12

0.72

Ultrathin

300

3.56

3.47

3.84

3.15

3.21

3.45

13,6

0.12

0.72

Thin

300

3.45

3.56

3.06

3.52

3.02

3.32

13,6

0.24

0.18

Ultrathin

300

2.89

2.56

2.87

3.02

2.56

2.78

13,6

0.24

0.18

Thin

300

3.21

3.25

3.31

3.36

3.35

3.30

13,6

0.24

0.36

Ultrathin

300

1.98

2.03

2.56

2.21

1.87

2.13

13,6

0.24

0.36

Thin

300

2.13

2.19

2.13

2.20

1.98

2.13

13,6

0.24

0.54

Ultrathin

300

1.54

1.52

1.23

1.54

1.26

1.42

13,6

0.24

0.54

Thin

300

2.09

1.99

2.23

2.27

2.38

2.19

13,6

0.24

0.72

Ultrathin

300

2.25

1.54

1.25

1.54

1.32

1.58

13,6

0.24

0.72

Thin

300

2.68

2.54

2.62

2.61

2.54

2.60

13,6

0.36

0.18

Ultrathin

300

2.54

2.25

2.31

1.26

2.25

2.12

13,6

0.36

0.18

Thin

300

2.62

2.56

2.41

2.81

2.23

2.53

13,6

0.36

0.36

Ultrathin

300

2.30

2.24

2.25

2.31

2.25

2.27

13,6

0.36

0.36

Thin

300

2.32

2.26

2.19

2.31

2.35

2.29

13,6

0.36

0.54

Ultrathin

300

1.68

1.84

1.98

1.84

1.20

1.71

13,6

0.36

0.54

Thin

300

2.01

2.06

2.21

2.25

2.21

2.15

13,6

0.36

0.72

Ultrathin

300

2.54

2.38

2.56

2.41

2.56

2.49

13,6

0.36

0.72

Thin

300

1.53

1.85

2.15

2.12

2.03

1.94

13,6

0.48

0.18

Ultrathin

300

2.87

2.89

2.56

3.01

2.54

2.77

13,6

0.48

0.18

Thin

300

2.56

2.25

2.74

2.89

2.45

2.58

13,6

0.48

0.36

Ultrathin

300

2.56

2.84

2.63

2.85

2.41

2.66

13,6

0.48

0.36

Thin

300

2.56

2.65

2.58

2.23

2.55

2.51

13,6

0.48

0.54

Ultrathin

300

2.57

2.56

2.54

2.56

2.58

2.56

13,6

0.48

0.54

Thin

300

2.47

2.85

3.02

2.56

2.86

2.75

13,6

0.48

0.72

Ultrathin

300

3.25

3.25

3.03

3.06

2.98

3.11

13,6

0.48

0.72

Thin

300

3.25

3.46

3.54

3.56

3.24

3.41


The Analysis of Variance is applied to analyze the results as shown in table 6
.




Table 6
:

ANOVA table of
2 1
4.2
factorial experiment with 5 replications
.


Source of
Variation

Degree of
Freedom

Sequential
Sum Square

Adjusted
Sum Square

Adjusted
Mean
Square

F

P
-
value

B

3

21.6533

21.6533

7.2178

100.25

0.000

C

3

8.8777

8.8777

2.9592

41.10

0.000

D

1

1.2691

1.2691

1.2691

17.63

0.000

B*C

9

8.0395

8.0395

0.8933

12.41

0.000

B*D

3

2.1260

2.1260

0.7087

9.84

0.000

C*D

3

1.0527

1.0527

0.3509

4.87

0.003

Error

137

9.8638

9.8638

0.0720



Total

159

52.8820

52.8820





R
-
Sq(adj) =78.35%




4
3
2
1
2
1
3
.
0
2
.
5
2
.
0
3
.
0
2
.
5
2
.
0
B
C
D
1
2
3
4
B
1
2
3
4
C
I
n
t
e
r
a
c
t
i
o
n

P
l
o
t

f
o
r

R
e
s
p
o
n
s
e
F
i
t
t
e
d

M
e
a
n
s

Figure 3:

Interaction plot for response of
2 1
4.2
factorial experiment.



According to table 6, the factors which affect to fluctuation of meandering is factor B, C, D, BC
interaction, BD interaction and CD interaction by using 0.01 significance level. According to
figure 3, the best settings which give the smallest maximum fluc
tuation of meandering are level
2 (0.24

Watt
) of factor B, level 3 (0.54

Watt
) of factor C and level 1 (ultrathin) of factor D.


Next,
the confirmation experiment is

conducted
in order to confirm the settings that the
maximum fluctuation us not
exceed

3

μm.
The results are shown in table 7.


Table 7
:

The results of the confirmation experiment with 40 replications
.


Wafer no.

Response

Wafer no.

Response

Wafer no.

Response

Wafer no.

Response

1

0.93

11

0.77

21

2.79

31

1.66

2

2.40

12

0.90

22

2.37

32

2.48

3

3.38

13

1.31

23

2.48

33

0.58

4

1.57

14

0.93

24

1.65

34

0.84

5

1.74

15

2.61

25

0.45

35

1.40

6

2.17

16

2.26

26

2.31

36

1.21

7

1.76

17

0.41

27

0.74

37

0.91

8

1.29

18

2.61

28

1.39

38

0.94

9

0.80

19

2.02

29

1.80

39

0.83

10

1.76

20

0.27

30

0.98

40

2.64


According to table 7, there is only one
slice which is shown the maximum fluctuation of
meandering

exceeds the target
. This can be

acceptable.


CONCLUSIONS


The factors affected to the fluctuation of meandering are scan power 1, scan power 2 and beam
shape.
The best setting

of scan power 1 is 0.24 Watt, scan power 2 is 0.54 Watt and beam
shape is ultrathin. The benefit after implement
ed

the new settings is sho
wn in table 8


Table 8
:

The benefit
after implemented the new settings
.


Process

Average
loading

%
reject

wafer
amount

cost/wafer

(k
.euro)

Reject cost

Saving

Past parameter

520

4.62

24

14

336

45.83%

New

parameter

520

2.5

13

14

182


After implemented
new parameters to shop floor, specimen silicon wafers from 40 slices. There
is 1 from 40 slices which shown defects from meandering. That is 2.5 %. Typically, average
loading is 520 wafers per month, average rejected wafers per month are 24 slices that is
4.62%.
Comparing to defect rate of the new parameter, the numbers of rejected wafers are 2.5% or 13
wafer slices per month by using the same average loading per month. Wafer cost is totally
14,000 euro per slice. Therefore, it can be save cost up to 45.83%
.



REFERENCES


1.

Fukuyo, F., Fukumitsu, K. and Uchiyama, N. (2005) ‘The
s
tealth
d
icing
t
echnology and
t
heir
a
pplication
’,
Proceedings of 6
th

L
aser Precision Microfabrication
2005
, pp.

1
-
6.

2.

Hermanns, C. (2005) ‘Laser separation of flat glass’,
Proceedings of

the 63th the Laser
Materials Processing Conference
, pp. 105
-
109.

3.

Kumagai M.
, Uchiyama N., Ohmura, E., Sugiura, R., Atsumi, K. and Fukumitsu, K.

(2007)

‘Advanced dicing technology for semiconductor wafer
-
stealth dicing’, IEEE Transactions on
Semiconductor Manufacturing, 20
(
3
)
, pp. 259
-
265.