Modelling of Free Air Ball for Copper Wire Bonding Jonathan Tan, Boon Hoe Toh and Hong Meng Ho
Kulicke & Soffa Pte. Ltd.
6, Serangoon North Ave 5, #03-16, Singapore 554910
Email:
Abstract
Copper wire ball bonding has gained considerable attention due to its economic advantage, strong resistance to sweeping and superior electrical performance. In order to have a good first bond, consistent free air-ball formation for copper bonding are even more crucial than they are in the gold wire process.
To create a Free Air Ball (FAB), the wire bonder us an Electronic Flame-Off (EFO) unit, where high voltage is connected. During operation, the EFO gap is breached by a high current, creating a high voltage spark, which melt the tail of the copper wire in a glow discharge to form a spherical ball. Unlike gold wire, copper wire oxidizes readily. Hence, during the formation of the FAB, the copper wire must be enclod in an inert gas environment in order to prevent oxidisation of the FAB.
In this study, an empirical methodology was developed to model a consistent FAB for copper wire diameters ranging from 0.8mil to 2.0mil. Cherry pit bonds were created to study the FAB. Numerous tests were run on an automatic ball bonder in which current and time were varied and the resulting ball size measured. The data points were then ud as inputs for the empirical model.
This methodology us the EFO current and time as measurable energy inputs and the FAB size as the measurable energy output. It is simple to u and has the advantage of avoiding complex computation of phenomena analysis, which involves the pha change during FAB melting and formation. It is also able to predict copper wire FAB to provide consistent FAB size.
1. Introduction
Wire bonding has been the most popular interconnection method in miconductor device packaging. Recently, due to the increasing demand for enhancing the reliability of the devices and saving materials costs, attention has been focud on wire bonding using high strength and conductive copper wire.
The wire bonding cycle commences with electrical breakdown of the air gap between the wire and wand followed by the discharge which heats and melts the tail of wire. Surface tension caus the m
elted part to roll up into a ball, in which this ball would then be presd and formed into first bond by the capillary.
In wire bonding, constant ball and wedge dimensions are important for the connection quality. Slight variations in the deformed ball size of the first bond cau yield loss The key for a reliable ultra fine-pitch copper wire process is the formation of a round, reproducible FAB. To prevent the FAB from oxidising, an inert atmosphere around the tail-EFO wand area during the flame-off pha is created. The shape and size of FAB formation deeply affects the quality of copper wire bonding as it affects the bondability of the first bond as well as the loop height due to its Heat Affected Zone (HAZ).
In order to have a consistent FAB formation, preci control of the current/voltage of the EFO has a direct impact on the formation of the FAB (melted ball before actual deformation during the bond process). The EFO generates the spark that melts the end of the copper wire to form the tiny ball.
The phenomena of the FAB formation had been investigated by various rearchers. Cohen et al had studied the melting and solidification of thin wires FAB experimentally and numerically [1]. Qin et al showed that the HAZ length can be decread by using higher current and minimising EFO time [2]. Chen et al studies indicated that EFO current and time are the most significant factors in the
FAB ball formation [3]. The studies were focus on gold wire FAB. For copper wire FAB formation, there is an additional factors of inert cover gas. Onuki et. al had deduced that copper wire FAB formed in a shield gas heated above 175°C is softer than ball formed in a room temperature shield gas [4].
For the stable bonding of copper wire, it is important to characterize the wire with various diameters during the bonding process. To investigate this relationship, experiments were carried out on the various sizes of copper wire diameters and FAB. The wire size chon for the test range from 0.8 mil to 2.0mil.
2. Experimental
2.1 Test Setup
The modeling of the FAB was employed using the following procedure: First, numerous tests were run on an automatic ball bonder with various diameters of copper wire in which current and time were varied and the resulting ball size measured. The variation of the EFO current levels is ±0.5mA, the variation of the EFO fire time is ±0.002ms according to the specification of the EFO generation system.
A special bond program was ud to obtain sticks of FA
B which were as shown in Figure 1 using the cherry pits bond feature of K&S wire bonder. For every t of data, 40 FABs were generated. The FAB size were then measured using the NEXIV VMR 3020 vision system. The FAB were then inspected using JEOL scanning microscope JSM-5410 for their sphericities and tho that were not were excluded to ensure that the measurement taken by the NEVIX automatic vision measuring system was reprentative.
Figure 2 showed the SEM picture of a good accepted FAB and Figure 3 showed a bad FAB that was rejected in the data compilation. The parameters investigated were as shown in Table 1.
Figure 1. SEM pictures of cherry pits bonds for FAB study.
Figure 2. SEM pictures of good FAB.
Figure 3. SEM pictures of good FAB.
新概念第二册Table 1. Range of experimental parameters investigated. Parameters Range Investigated
Copper wire diameter 0.8mil, 1.0mil, 1.3mil, 1.5mil & 2.0mil Type of cover gas Forming gas (5%H2+95% N2) and Purified Nitrogen Cover gas flow rate 0.5 scfh to 1.5 scfh EFO gap tting 10 mil to 30 mil Wand geometry 20 mil and 32mil tip diameter EFO current 20mA to 150mA
EFO fire time 0.6ms to 4.5ms
托福听力
In general, all parameters investigated in Table 1 affect the dimension of the FAB. However, the main focus of this
paper is on modeling the dimension of the FAB, therefore this will be the main stream of discussion. 竞争者英文
For example, even though the type of cover gas had a lesr impact on the dimension of the FAB except for 2.0mil copper wire, it did affect the surface finish of the FAB significantly.
After the initial screening, only parameters of EFO current and fire time were ud to generate the characteristics curves for various diameters copper wires FAB as shown in Figures 4 to 8.
Figure 4. Relations of FAB size for EFO current and time for 0.8mil copper wire with forming gas at flow rate of 1.0
scfh.
Figure 5. Relations of FAB size for EFO current and time for 1.0mil copper wire with forming gas at flow rate of 1.0
scfh .
Figure 6. Relations of FAB size for EFO current and time
for 1.3mil copper wire with forming gas at flow rate of 1.0
scfh.
Figure 7. Relations of FAB size for EFO current and time for 1.5mil copper wire with forming gas at flow rate of 1.0
scfh.
Figure 8. Relations of FAB size for EFO current and time for 2.0mil copper wire with forming gas at flow rate of 1.0
hold什么意思scfh.
The configuration of the t up for various wire diameters were shown in Table 2.
Table 2. Configuration of the t up for various Copper wire
diameters.
Cu wire diameter Cover gas Flow Rate
(SCFH)
EFO
geometry
0.8 mil 5/95 H2/N20.8 – 1.0 scfh 20 mil
1.0 mil 5/95 H2/N20.8 – 1.0 scfh 20 mil
1.3 mil 5/95 H2/N20.8 – 1.0 scfh 20 mil普林斯顿大学公开课
1.5 mil 5/95 H2/N20.8 – 1.0 scfh 32 mil
2.0 mil 5/95 H2/N20.8 – 1.0 scfh 32 mil
2.2 Empirical Modelling
The formation of a FAB from the EFO box can be regarded as a system having both energies input and output. The input for the FAB is from the EFO box and the controlling parameters on the EFO system are constant current and firing time.
A plasma discharge is then generated from the spark gap and a heat flux is generated as the plasma stuck the tail of the wire. This heat flux then formed the energy input to melt the wire from its tail. Surface tension caus the melted part to roll up into a ball. Eventually, the measurable output is the size of FAB.
Bad upon the above analysis, it is reasonable to assume that FAB is a function of constant current output and EFO firing time and this can be written as
FAB=f(I,t)
The basis of empirical ari from the concept of Energy output is a result of Energy Input. Referring to Figure 8, example from relation of EFO current and time for 2.0mil copper wire FAB, it is reasonable to assume that with the Energy Output (reprented by the same FAB size), the Energy input by various current characteristics curves are the same. For example, from Figure 8, the energy level to produce a FAB size of 100µm are the same, ie, energy level of (90mA+2.2ms) is the same as that of (120mA+1.5ms), so are (110mA+1.7ms) and (100mA+1.87ms).
Hence, it is necessary to transform various characteristics curve into a common characteristics curve so that the energy output can be read-off as a common energy input which compris of various characteristics curves.
To carry out this transformation process, first examined the EFO-FAB as a system that consists of energy output and input. The input variables were current and EFO fire time and the measurable output is FAB size. Within the system, the ideal energy input is
E in = V.I.t = R.I2.t
The resistance of the system at the gap is negligible once the breakdown voltage occurs and can be regarded as a constant, therefore,
FAB (D) = f (I n.t) where 1<n<2
The x-axis is then transformed from the variable EFO fire time into energy input variable (I n.t). The ideal energy efficiency should be (I2.t). With n=1.3, the characteristics curves of a range of current ttings had been transformed into a single common characteristics curve. It is now (I1.3.t), which can be regarded as the efficient energy from the EFO box in the formation of the FAB ball.英译汉在线翻译器
Once a common curve is obtained with various current tting, curves fitting technique is then ud to obtain a trend equation, e Figure 9.
Therefore both energy input and output can then be co-related by a common characteristics curve described by the equation of the form as in shown in Equation (1). The fitted equation took the form of the natural logarithm, hence by plotting the x-axis using the natural logarithm scale, the characteristics curves is able to describe the efficiency of the EFO-FAB system. This will be discusd in more details at a later ction.
The description of this characteristics curve is currently in another domain. To convert the characteristics curves back to the Time domain, an inver transformation is then performed. Now, a t of characteristic curves with different current ttings, described by a common equation, were th
en made available as shown in Figure 10 for the copper wire size of 0.8mil.
Figure 9 Energy Domain Characteristics curve for 2.0mil Copper wire FAB. 3. Results and Discussions
Using this above approach, a single characteristics equation is able to describe the relationship between EFO tting parameters, namely current and EFO fire time, and the desired FAB for a specific type of wire size.
To simplify the equation, curve fitting technique is employed and so that EFO fire time can be predicted using equation (1)
n
C D C D C C D C D C FAB I
e
t w w w w 6
5243
221+++++=
(1)
where the cofficients are shown in Table 3.
Table 3. Cofficients of Empirical Models for Copper wires
and gold wires.
Samller to Medium Cu wire Larger Cu wire Larger Au
wire
t (ms) Predicted EFO firing time in ms
FAB (µm) Desired FAB size in µm
物理化学学报
fectI (mA) EFO current tting in mA
D w (mil) 0.8, 1.0, 1.3 1.5, 2.0 1.5, 2.0 C 1 +141.63 +211.78 +359.69 C 2 -85.99 -420.11 -1007.10 C 3 +53.65 +317.62 +917.17 C 4 +23.44 +36.04 +49.39 C 5 -16.75 -78.20 -138.30 C 6 +27.52 +76.38 +149.09 n 1.3 1.3
1.3 It is noted that equation (1) only applied to copper wire diameter of 0.8, 1.0, 1.3mil in the smaller to medium diamter range and 1.5,
2.0mil in the larger diameter range. After transforming to the Time domain, the characterisitcs equation or the empirical model compris of a ries of curves describing the relationship between the desired FAB size, EFO current and corresponding EFO time. Figures 10 to 14 showed the EFO-FAB characteristics curves for various wire diameters obtained by this empirical
method with configuration as stated in Table 2. In the figures, the desired FAB size were plotted on the x-axis and the calculated EFO time plotted as the y-axis. Using the graphs, the predicted EFO time could be read-off directly once the desired FAB size and EFO current were t. Equation (1) could be implemented in software coding. To verify the empirical models, the experimental data points in Figures 4 to 8 were re-plotted on Figures 10 to 14. Figure 10 showed that the model describes the 0.8mil copper wire FAB acurately in the medium current range
(30mA to 50mA). The experimental data points at 20mAcompare的用法
agree well starting from a FAB size of 42µm. Figures 11
and 12 showed that the models described both 1.0 and
1.3mil copper wire FAB very well. Figure 13, however, showed that the experimental data points only agree well with the tting at higher current tting, in the range of 90mA to 120mA. Experimental obrvations also showed
that the 40 FAB data points at the lowest current, 40mA, had larger standard deviation when compared to tho at the higher current tting.
Figure 11. Empirical Model for 1.0 mil Copper wire FAB. Figure 14 showed that the experimental data points of the 2.0mil copper wire agree very well with the model at the higher current range tting of 90mA to 120mA. In general, the desired EFO current tting for configuration of Table 2 tting were summarid in Table 4. In the later ction, a detailed discussion will be highlighted with additional experimental data points to
suggest that EFO current at 120mA might not be sufficient for 2.0mil copper wire.
Figure 12. Empirical Model for 1.3 mil Copper wire FAB.
Figure 13. Empirical Model for 1.5 mil Copper wire FAB.
Figure 14. Empirical Model for 2.0 mil Copper wire FAB.
Table 4. Desired EFO current tting range for various copper wire diameters in forming gas.
Cu wire diameter Desired EFO current
range
Corresponding EFO
fire time range
0.8mil 30mA to 50mA 0.2ms to 1.4ms
1.0mil 40mA to 60mA 0.4ms to 1.6ms
1.3mil 50mA to 70mA 0.5ms to
2.5ms
1.5mil 90mA to 120mA 0.5ms to
2.0ms
2.0mil 90mA to 120mA 1.0ms to 2.8ms
Likewi, the same methodolgy have also been employed to generate empirical model to describe the FAB size of various copper wire diameters with purified nitrogen as cover gas and different geometries of EFO wand tip diameter. The results of the will be reported elwhere. It is noted that the type of cover gas and EFO wand geometry also affected the formation of the FAB significantly. The will be discusd in the later ction.
Investigation on Efficiency of Energy Transfer
wrongIt is obrved that the characteristics curves at the energy domain can be ud to describe the energy efficiency of the EFO-FAB system.
Figure 15 showed the characteristics curves for two EFO wands with different geometries, one with tip diameter of 20mil and the other 32mil. It can be en that the characteristics curves for the larger diameter wand are shifted to the left when compared with the smaller diameter wand. Using the analogy that FAB size is a product of EFO current and time, it is obrved that the smaller wand required longer time to produce the same FAB size as compared to the larger wand. This additional time means that the smaller wand required more energy to produce the same FAB size, hence indic
ating that the larger diameter has a higher energy efficiency in the EFO-FAB system.
Figure 15. Comparison of different EFO wand geometry.
By plotting the two different wand diameter’s characteristics curves in the energy input domain as shown in Figure 9, it is obrved that the larger wand’s characteristics curve was above the smaller wand’s characteristics curve. This obrvation confirms that the highest characteristics curve in the plot has a higher energy efficiency in the EFO-FAB system.
A similar t of experiments were also carried out with purified nitrogen as cover gas for various copper wire diameter. The results of the tests will be reported elwhere. In short, it is obrved that forming gas has a higher energy efficiency in the EFO-FA
B system.
As in the ca of the larger EFO wand in the 2.0mil Copper wire, Figure 9 showed that the characteristics curve of pre-mixed forming gas (5% hydrogen in 95% nitrogen) as cover gas lies well above that of the purified nitrogen cover gas. This clearly indicates that forming gas has a higher energy efficiency compared to purified nitrogen.
In our context, the definition of energy efficiency is to obtain the same FAB size using the minimum combination of EFO current and time. In particular, the shorter the EFO
