Chapter 4 Results and Discussion
4.1 Measurement Results
4.1.1 Nanoindentation results of blank films
Table 4.1 showed the mechanical properties of materials commonly used in the multi-layered, patterned BOAC structures as illustrated in Figures 3.2 through 3.4. All blanket films except ultra low-k were measured by nanoindentation at NCTU. The data for ultra low-k were kindly provided by UMC. These basic properties would be used for quantitative analysis and discussion in the later chapter.
Table 4.1 The modulus and hardness of blank films in BOAC structures
Materials Modulus rdness
(GPa)
Std. Dev. (GPa) (GPa) Std. Dev. (GPa) Ha
Al 69.2 3.0 0.25 0.03
Cu 137.2 6.1 1.31 0.19
FSG 64.0 2.8 6.68 0.45
Low-k 15.4 1.0 2.24 0.17
oxide 79.5 3.4 7.80 0.53
Si 182.8 3.2 12.56 0.39
Ultra low-k 9 n/a n/a n/a
4.1.2 BOAC s
4.1.2.1 Identification of indentations’ positions
asurement, indentation marks, which should be M. Figure 4.1 showed SEM top-v
tructures
To ensure the accuracy of the me
placed in the center of the pad, were verified by SE
iew graph of 90nm BOAC parts after indentation test, indicating that indentation marks were truly placed in the center of the pad. The same procedure was employed for the other BOAC samples to ensure proper indentation tests with correct marking.
Figure 4.1 SEM top-view graph of indentation makes and their locations
4.1.2.2 Nanoindentation results of BOAC structures
chnology (90LKBOAC A and 90
igure 4.2 showed the nanoindentation results of these 2 BOAC structures.
First we took 2 BOAC structures fabricated for 90 nm te
LKNormal) as an example to demonstrate how we use nanoindentation to quantify the mechanical strength of BOAC structures. These 2 structures had the same Al pad, but different substrate structures under Al pad, whose cross-sectional diagrams have been shown in Figure 3.2 and 3.4. The 90LKNormal had full array of dummified trench/via in Cu/ILD. Therefore, the 90LKBOAC A only had sparse via in Cu/ILD.
F
0 500 1000 1500 2000
40
Figure 4.2 Nanoindentation results of 90LKBOAC A and 90LKNormal
In chapter 3, we had already explained how to distinguish the different substrate effects for a given film/substrate system. would increase with increasing indenter depth and Es when the Es (m was larger than Ef
(modulus of film). From Figure 4.2, the modulus ( ) of 90LKNormal was
ation
obviously larger than that of 90LKBOAC A, indicating that the modulus of composite
rneath Al pad. Therefore, this thesis intends to set up a methodology to differentiate if the BOAC structures can pass the wire bonding or wafer testing by quantitative analysis of the mechanical strength of BOAC structures in the next section.
substrate Ecs of 90LKNormal was stronger than that of 90LKBOAC A.
This method was also used for all BOAC structures. Figure 4.3 showed the nanoindentation results of all BOAC structures. The mechanical strength of all BOAC substrates Ecs in decreasing order was 65LKNormal > 65ULKNormal > 90LKNormal
> 65LKBOAC > 90LKBOACA > 65ULKBOAC > 90LKBOACB.
Based on these results, we can establish that nanoindentation test was easy and quick to compare the mechanical strength of various BOAC structures unde
0 500 1000 1500 2000
40
Figure 4.3 Nanoindentation result C structures
4.1.2.3 King’s model Fitting
King’s model, which has been described in the previous chapter, will be used to quan
was the raw data from nanoindentation measurem Figure 4.4,
cs
f able
odulus of the composite substrate in 90LKNormal was 109 GPa fitting. From Figure 4.4, the range of fitting was only valid between 0 and 1200 nm because King’s model fitting was limited to the Al film thickness, which was 1.2 μm in our BOAC structures.
This method was also applied to all BOAC structures with fitting results shown in Figure 4.5 ~ 4.11. Table 4.2 listed the Ecs of all BOAC samples. In the later section, we will discuss the learning and issues of such fitting method and the relationship between the Ecs and BOAC structures.
tify the mechanical strength of BOAC structures. We took a BOAC structure, for example, 90LKNormal to illustrate the fitting step of
E
fnanoindentation .ation nanoindent
E
f ent as shown inwhile the fitting curve related to substrate effect up to 1.2 μm was represented by
E
Kingf 's−fitting. First Ef of aluminum was measured to complete thefitting s King
E
f ' − fitting and calculate the Ecs, which was the modulus of composite BOAC substrate in a two-layered Al/substrate model. We changed E and used “try and error” to fit the most matched curve. The E of Al was 69.2 GPa as listed in T 4-1. Therefore, the mfrom
E
Kingf 'sfittingf
rious BOAC structures Samples Ecs (Al film/composite substrate), GPa
Figure 4.4 Curve fitting of E using King’s model for displacement below 1200 nm
Table 4.2 Modulus of composite substrate in va
90LKBOAC A 75.5
0 500 1000 1500 2000
20
0 500 1000 1500 2000
Figure 4.5 King’s model fitting result for 90LKBOAC A
2000
0 500 1000 1500
40
0 500 1000 1500 2000
Figure 4.7 King’s model fitting result for 90LKNormal A
0 500 1000 1500 2000
50
Figure 4.8 King’s model fitting result for 65LKBOAC
0 500 1000 1500 2000
Normal Pad 65LK 125GPa
Figure 4.9 King’s model fitting result for 65LKNormal
0 500 1000 1500 2000
20
Figure 4.10 King’s model fitting result for 65ULKBOAC
-200 0 200 400 600 800 1000 1200 1400 1600 1800 2000 20
30 40 50 60 70 80 90 100 110 120 130 140
modulus(GPa)
displacement(nm)
119GPa
Normal Pad 65ULK
Figure 4.11 King’s model fitting result for 65ULKNormal
4.2 Analysis and Discussion
4.2.1 Parameters affecting BOAC mechanical strength
In the previous sections, we have reported the measurement and fitting for 7 different BOAC samples, which included cases for studying the effects of technology nodes, mechanical strength of low-k materials, bond pad structures and copper density of metal lines. The Ecs, the modulus of composite BOAC substrate in a two-layered Al/substrate model, will be further analyzed and discuss how these parameters can affect the mechanical strength of BOAC structures.
4.2.1.1 Bond pad structures
From Table 3.2, we took 90LKBOAC A vs. 90LKNormal to analyze how the bond
pads affect .12. From
able 4.2, the Ecs of 90LKBOAC A and 90LKNormal were 75.5 GPa and 109 GPa, spectively. The difference of Ecs between these two bond pad structures was 33.5 Pa. This could be attributed to the full array of dummified, stacked trench/via inforcement in M1-6 in normal pad. In contrast, BOAC contained typical M1-6 nes and via with less than 6% of die area, randomly distributed throughout the die as
own in Figure 4.13. Similar result was observed for normal pad vs. BOAC structure 65 nm node and design as shown in Figure 4.14. However, normal pad was much ronger than BOAC pad in the case of 65 nm parts. This magnified the need for nderstanding the mechanical strength of structure underneath the Al pad in the wire onding process. Although the BOAC structure did not possess mechanical strength s strong as conventional normal pad, it still possessed the basic mechanical strength resist the impact force from wire boning or wafer probe as judged from their assing in bondability test.
ed the mechanical strength of BOAC as illustrated in Figure 4 T
0 500 1000 1500 2000
20
ion results of 90LKNormal vs. 90LKBOAC A
duluGPa
Figure 4.12 Nanoindentat
Figure 4.13 Structures of 90LKBOAC A and 90LKNormal Copper
20% copper density
Al (1.2um) Oxide (1.2um)
Si substrate
90% copper density
90LKBOAC A 90LKNormal
0 500 1000 1500 2000 40
50 60 70 80 90 100 110 120 130 140 150 160
modulus(GPa)
displacement(nm)
65LKNormal 65LKBOAC
Figure 4.14 Nanoindentation results of 65LKNormal vs. 65LKBOAC A
4.2.1.2 Copper density
Next we investigated how the copper densities affected the mechanical strength of BOAC using 90LKBOAC A vs. 90LKBOAC B listed in Table 3.2 and shown in Figure 4.15. From Table 4.2, the Ecs of 90LKBOAC A and 90LKBOAC B were 75.5 GPa and 70 GPa, respectively. The difference in Ecs was 5.5 GPa since these two structures were almost the same except top copper densities as shown in Figure 4.16.
The top copper densities for 90LKBOAC A and 90LKBOAC B were 90 % (block) and 10 % (ring). The difference of Ecs was small although the difference in copper density was large. It indicated that the oxide layer over the top metal layer absorbed
the main im have little
influence on the mechanical strength of BOAC structures.
pact force from indenter, so the top metal’s density seemed to
0 500 1000 1500 2000 30
40 50 60 70 80 90 100 110 120 130 140
modulus(GPa)
displacement(nm)
90LKBOAC A 90LKBOAC B
Figure 4.15 Nanoindentation results of 90LKBOAC A vs. 90LKBOAC B
We then studied how the backend dimensional scaling affected the mechanical strength of BOAC structures using 90LKBOAC A vs. 65LKBOAC listed in Table 3.2.
Copper Block Copper Ring
Figure 4.16 The cross-sectional view of M7 copper densities in BOAC structure
4.2.1.3 Technology nodes
Si substrate Al (1.2um)
Oxide (1.2um)
Si substrate
Copper (M7) Block (90%)
Al (1.2um) Oxide (1.2um)
Copper (M7) Ring (10%)
The thicknesses of passivation, Al pad and top oxide layer remained the same, while ILD/metal thickness in M1-6 scaled from 90 nm node to 65 nm node with increasing AR in M1-6. From Table 4.2, the Ecs of 90LKBOAC A and 65LKBOAC were 75.5 GPa and 95 GPa, respectively. The difference was 19.5 GPa as shown in Figure 4.17.
The increased Ecs in 65 nm BOAC structire was attributed to the larger Cu fraction in M1-6 layer as illustrated in Figure 4.18, where the Cu fractions in 90 nm and in 65 nm were 44 % and 58 %, respectively. It indicated that the scaling effect had large influence on the mechanical strength of BOAC structures and the higher Cu fraction in the Cu/Low-k layer, the stronger BOAC structure would be.
0 500 1000 1500 2000
40 60 80 100 120 140 160
modulus(GPa)
displacement(nm)
65nm_BOAC 90nm_BOAC
Figure 4.17 Nanoindentation results of 65LKBOAC vs. 90LKBOAC
Figure 4.18 Cu fractions in the Cu/Low-k layers of 90 nm and 65 nm structur
4.2.1.4 Low-k materials
We also addressed how the low-k materials affected the mechanical strength of BOAC structures using 65LKBOAC and 65ULKBOAC listed in Table 3.2 and shown in Figure 4.19. These two structures were almost the same except the low-k materials in ILD layer M1-5 as shown in Figure 4.20. From Table 4.2, the Ecs of 65LKBOAC and 65ULKBOAC were 95 GPa and 71 GPa, respectively. The difference in Ecs (24 GPa) was large, which cannot be sorely attributed to the decrease in ILD’s mechanical strength from low-k (15 GPa) in 65 nm to ultra-low-k (9 GPa) in 65 nm parts. Weaker interfacial adhesion in ULK/metal due to the existence of porosity [22, 37] may be the culprit of the much lower Ecs. To prove this hypothesis, we used FIB to examine the
cross- ures
4.21~4.24. For the low-k sample (LK), there was no crack found under the indentation and the ILD seemed flat except top copper layer. However, in ultra low-k sample (ULK), there were many cracks observed under the indentation and the ILD bended under the impact force. These observations indicated that the ultra low-k
section of BOAC structures after nanoindentation test as shown in Fig
researcher [37] had shown that the weak adhesion would lower the modulus during nanoindentation. In this study, low modulus and weak adhesion were believed to be the main reasons affecting the mechanical strength of BOAC structures.
0 500 1000 1500 2000
20
Figure 4.19 Nanoindentation results of 65LKBOAC vs. 65ULKBOAC
Figure 4.20 Structures of different low-k materials in 65LKBOAC and 65ULKBOAC
Figure 4.21 FIB cross-sectional view of 65LKBOAC (8,000 X)
No cracks happened
.22 FIB cross
Figure 4 -sectional view of 65LKBOAC (25,000 X)
Figure 4.23 FIB cross-sectional view of 65ULKBOAC (8,000 X)
Lots of cracks
Figure 4.24 FIB cross-sectional view of 65ULKBOAC (25,000 X)
So far
rength of various BOAC structures in a soft film/ hard substrate system. However, the accuracy of these quantified values warranted verification. As a result, BOAC structures in a hard film/soft substrate system will be further analyzed to check if the same mechanical strength of BOAC structures can be obtained. In addition, the applicability and issues of these quantified values will be addressed.
, there is no problem using the quantified values to analyze the mechanical st
4.2.2 H
was the same as that with the original BOAC structures Al film on the top.
Curve fitting was also used to characterize the mechanical strength of various
f for the oxide film was measured to able 4.1. We also changed Ecs and used “try and error” to fit ost matched curve. The efficient fitting range was from beginning to the lowest because the fitting curve didn’t rise up in the end of
cs (Modulus of composite BOAC substrate) for various BOAC structures rve fitting as illustrated in Figures Table 4.3.
ard film/soft substrate system
When the Al pad was etched from BOAC structures, the stacking of BOAC structures became oxide film/composite substrate structure, which was a hard film/soft substrate system. The nanoindentation results, Efnanoindentationof BOAC structure with oxide as the top layer shown in Figure 4.25, decreased with increasing indenter depth, indicating that the moduli of composite BOAC substrates were smaller than the modulus of oxide film. The mechanical strength of composite BOAC substrates in decreasing order
with
BOAC structures with oxide layer on the top. E be 79.5 GPa as listed in T
with oxide on the top can be obtained by cu 4.26-4.32 and summarized in
0 500 1000 1500 2000
30
Figure 4.25 Nanoindentation results of BOAC structures with oxide film on the top
Table 4.3 Modulus of composite BOAC substrate with oxide film on the top in a strate model
two-layered oxide/composite sub
Samples Ecs (oxide film/composite substrate), GPa
90LKBOAC A 45
90LKBOAC B 41
90LKNormal 63 65LKBOAC 54 65LKNormal 80 65ULKBOAC 42 65ULKNormal 74
0 500 1000 1500 2000
40 60 80 100 120 180 200
140 160
displacement(nm)
90LKBOAC A_oxide 45GPa
Figure 4.26 King’s model fitting result for 90LKBOAC A with oxide film as the top layer
0 500 1000 1500 2000 40
60 80 100 120 140 160 180 200 220
modulus(GPa)
displacement(nm)
BOAC 90 B_Etch 41GPa
Figure 4.27 King’s model fitting result for 90LKBOAC B with oxide film as the top layer
0 500 1000 1500 2000
60 80 100 120 140 160 180 200 220
modulus(GPa)
displacment(nm)
Normal Pad 90_Etch 63GPa
Figure 4.28 King’s model fitting result for 90LKNormal with oxide film as the top layer
260 280
0 500 1000 1500 2000
BOAC 65LK_Etch
Figu s model fitting result for 65LKBOAC with oxide film as the top layer
re 4.29 King’
0 500 1000 1500 2000
60
Normal Pad 65LK_Etch 80GPa
Figure 4.30 King’s model fitting result for 65LKNormal with oxide film as the top layer
0 500 1000 1500 2000 40
60 80 100 120 140 160 180 200 220
modulus(GPa)
displacement(nm)
BOAC 65ULK_Etch 42GPa
Figure 4.31 King’s model fitting result for 65ULKBOAC with oxide film as the top layer
0 500 1000 1500 2000
40 60 80 100 120 140 160 180 200
modulus(GPa)
displacement(nm)
Normal Pad 65ULK_Etch 74GPa
Figure 4.32 King’s model fitting result for 65ULKNormal with oxide film as the top layer
Judging from the curve fitting in Figures 4.26-4.32, it was found that significant deviations from were observed for displacement greater than
700-800 nm. In general, of BOAC structure with oxide as the top layer, rose up at the disp 700-800 nm due to (1) copper/low-k patterned layer underneath oxide layer and (2) the strain-hardening effect. Next we examined the data between 65 nm parts and 90 nm parts as shown in Figure
4.33. of 65 nm samples rose up at shorter displacement compared to 90 nm e (1) the pitch of each metal layer in Cu/low-k for 65 nm is
(2) the thickne 65 nm samples were thinner than that of 90 nm; thus its affected region touched the hard oxide la er earlier. These two reasons contributed the of 65 nm parts to rise up earlier than that of 90 nm at the displacement near 700-800 nm.
ation
smaller (0.16 μm vs. 0.32 μm in 90 nm parts) providing more of strain hardening and sses of the
y
ation nanoindent
E
f0 500 1000 1500 2000
30
Figure 4.33 Nanoindentation results of BOAC 90A vs. BOAC 65LK with oxide
4.2.3 The difference in Ecs between Al film/composite substrate and oxide film/composite substrates
We combined Table 4.2 and Table 4.3 into Table 4.4 for comparison.
Table 4.4 Modulus of composite BOAC substrate (Al vs. oxide)
From Table 4.4, the Ecs of oxide film/composite substrate were lower than those of Al film/composite substrates for the same BOAC design. The reasons for such discrepancy may be generalized as follow.
1. The indentation affected region within the composite substrates was not the same
For Al film/composite substrate system, the indentation affected region within composite substrates included oxide layer and Cu/low-k layers to a less degree as illustrated in Figure 4.34. In contrast, for oxide film/composite substrate, the indentation affected region within composite substrates included primarily Cu/low-k
Ecs (Modulus of composite BOAC substrate), GPa Samples
Al film/composite substrate Oxide film/composite substrate
90LKBOAC A 75.5 45
90LKBOAC B 70 41
90LKNormal 109 63
65LKBOAC 95 54
65LKNormal 125 80
65ULKBOAC 71 42
65ULKNormal 119 74
layers and to a less degree of Si substrate. Since the modulus of oxide and low-k were 79.5 GPa and 15.4 GPa, respectively, it could be easily understood that the Ecs would be larger in the Al film/composite substrate system than that in the oxide film/composite substrate system.
Figure 4.34 The affected regions within the composite substrates
2. The impact force was different in two different film/substrate systems
When the indenter traveled through the film, it needed higher impact force in the oxide film/composite substrate system than that in the Al film/composite substrate system because the oxide film was harder than the Al film. Moreover, high impact force would damage the structures under the impact of indentation; thereby the modulus would be underestimated. As a result, the oxide film/composite substrate
Al (1.2 μm) Oxide (1.2 μm)
Si substrate
Copper/low-k
would be damaged by impact force more easily than Al film/composite substrate, (3 μm)
(0.9 μm) Copper /FSG Oxide (1.2 μm)
Si substrate
Affected
Al system Oxide system
(0.9 μm)
region within composite Copper /FSG
Copper/low-k (3 μm)
However, there are two conflicting observations in these two film/substrate sy
1. There was large discrepancy of the calculated Ecs of 65LKBOAC between the
Al/su GPa as
listed in Table 4.4. Figure 4.35 showed the relation between these two
film/substrate system odul a
and the Ecs was 95 GPa, indicating that the modulus of the substrate ide layer should be an 79.5 GPa, or even 95 GPa. But in the ate system, the modulus of the composite substrate under oxide layer, GPa, which was not le.
l/substrate system, Ef t increased with inc g indenter the indenter penetra ide layer at the displacement of ≧
shown in Figure 4.3. It was thought that composite substrate under , the should decreased stems.
bstrate system and the oxide/substrate system, namely 95 GPa vs. 54
s. In the Al/substrate, the m us of oxide was 79.5 GP composite
under the ox larger th
oxide/substr
Ecs, was 54 reasonab
2. In the A nanoindenation reasin
depth after ted into the ox
1200 nm as
oxide layer was weaker than the oxide layer Efnanoindentation
after the indenter penetrated into the oxide layer. Instead, Ef
increased after the displacement ≧ 1200 nm as shown in Figure 4.3, which was not reasonable.
Figure 4.35 Comparison of Ecs between different film/substrate systems The reason for such discrepancies was the contact-area effect induced by the difference in the hardness of materials in the BOAC stack. The contact-area effect will be further discussed in the following section.
38]
w
hardness and modulus would be affected by the contact area (A). For Al
film/com cs )
ated. For the oxide film/composite substrate, the Ecs would be
effect and m
4.2.4 The contact area effect
The hardness would affect the modulus during nanoindentation. Researchers [36, found that the hardness would affect the contact area (A) during nanoindentation.
Soft films on hard substrates tended to pile-up when indented, while hard films on soft substrates tended to sink-in. Hence the true contact depth was underestimated in the case of a soft film on a hard substrate and overestimated in the case of a hard film on a soft substrate, when compared to the calculated contact depth using the Oliver–
Pharr method. Since contact area was a function of contact depth, contact area (A) as underestimated in the case of a soft film on a hard substrate and overestimated in the case of a hard film on a soft substrate. From Equation 3.1 and Equation 3.2, the
posite substrate, the E would be overestimated while the contact area (A was underestim
underestimated while the contact area (A) was overestimated. The contact area (A) ethod to eliminate the contact area factor will be discussed in the following section.
4.2.4.1 Eliminating the contact area factor
A method [36, 38] based on P/S2 was employed to eliminate the contact area by combining Equation 3.1 and Equation 3.2 into Equation 4.1.
2
as directly proportional to hardness H and inversely proportional to the
exp
hig ples were lower than those of 90 nm
ement depth,
were higher than those of 90 nm e
previous section. The P/S2
1100 nm depth. This was caused by the changing Ecs as the affected region moving from stiff oxide layer down to more of weak low-k layers with the increasing indenter dep
diff
continually changing substrate with increasing indenter depth through different layer in the stack. For BOAC structures using 90 nm technology such as 90LKBOAC A and 90LKBOA B which had the same layout in the copper/low-k layers, the increasing slopes between 200 nm to 1100 nm depth were almost the same. This
continually changing substrate with increasing indenter depth through different layer in the stack. For BOAC structures using 90 nm technology such as 90LKBOAC A and 90LKBOA B which had the same layout in the copper/low-k layers, the increasing slopes between 200 nm to 1100 nm depth were almost the same. This