The mechanism of Pd distribution in the process of FAB formation during Pd-coated Cu wire bonding

Transcription

1 The mechanism of Pd distribution in the process of FAB formation during Pd-coated Cu wire bonding Yahong Du 1,2 Zhi Quan Liu 1,2 Hongjun Ji 3 Mingyu Li 3 Ming Wen 4 Received: 25 April 2018 / Accepted: 16 June 2018 / Published online: 30 June 2018 Springer Science+Business Media, LLC, part of Springer Nature 2018 Abstract The palladium (Pd) distribution over the free air ball (FAB) and its effect on the ball s shear strength were investigated in the Pd-coated copper (Cu) wire bonding process. It was found that for the same FAB/wire diameter ratio of 1.5, the bigger the electrical flame-off (EFO) current was, the larger would be the exposed Cu regions over FAB without Pd distribution. Combining experimental observations, a model of Pd distribution over FAB was first proposed by changing the EFO current and firing time. As the firing time increased, the coated Pd element could be dissolved into the Cu base to form a PdCu alloy at the FAB surface, to protect the bonded ball from corrosion. The Pd-coated Cu wire has higher bonding shear strength than the bare Cu wire. However, the FAB with the largest Pd coverage had the smallest ball shear strength, which revealed that the control of Pd distribution over FAB is very critical for a high-quality bonding interface. 1 Introduction As the primary interconnection technique in microelectronic package, wire bonding is widely accepted because of its flexibility and ease of use compared to other methods. Gold (Au) wire is generally preferred in integrated circuit (IC) package for its excellent ductility and resistance to corrosion and oxidation. However, the price of Au has risen considerably in recent years. On the one hand, copper (Cu) wire and sliver (Ag) wire are much cheaper than Au wire. Moreover, bonding wire like Cu and Ag with higher thermal conductivity and lower electrical resistivity can meet the requirements of a small sized and fine pitched package with more input/ * Zhi Quan Liu zqliu@imr.ac.cn * Ming Wen wen@ipm.com.cn 1 Institute of Metal Research, Chinese Academy of Sciences, Shenyang , China School of Materials Science and Engineering, University of Science and Technology of China, Shenyang , China State Key Laboratory of Advanced Welding and Joining, Harbin Institute of Technology (Shenzhen), Shenzhen , China Kunming Institute of Precious Metals, Kunming , China output (I/O) counts in the semiconductor industry. Therefore, the transition to Cu and Ag wire from Au wire draws much attention in wire bonding technology. However, pure Ag has the problem of sliver migration and corrosion. Thus, alloyed Ag wires with an addition of Au and Pd are being considered [1]. Although the cost of alloyed Ag wire is less than that of Au, it is still higher than that of Pd-coated Cu wire. Ag alloyed wire is not universal for all applications like radio-frequency applications [2]. Pd coating on Cu has shown sufficient potential to replace Au wire, due to its excellent bondability and reliability at a relatively lower cost [3 8]. Kaimori et al. [9] compared Au wire, Cu wire and PdCu wire and reported that the ball shear strength of PdCu wire is the best among the three wires for the same FAB size. Pd coating is free from oxidation and has good adhesion to a Cu wire. It also has a higher tensile strength than a bare Cu wire, when bonded on Al pad. Generally, the thickness of a Pd layer over the Cu core is about 100 nm. The formed free air ball (FAB) of Pd-coated Cu wire is different from that of pure Cu wire, due to the coexistence of Cu and Pd [10]. The FAB forming process of a Pd-coated Cu wire is more complicated than that of a bare Au wire and a bare Cu wire, in which Pd will distribute over the surface of the FAB, for protection. Many studies have reported that a Pd-coated Cu wire shows better performance than a bare Cu wire on Al pads in reliability tests, especially in humidity-related reliability tests [5, 11 13]. The corresponding mechanism is yet to be clarified, since Pd does not participate in an interfacial Vol:.( )

2 reaction to form any Pd Al intermetallic compound (IMC). However, the distribution of Pd at the FAB surface will affect the Pd distribution at the bonding interface, which can give us useful information for understanding bonding reliability. Tang et al. [14, 15] investigated Pd distribution in the FAB and reported that the Pd layer is thicker at the wire neck and gradually become thinner towards the equator of FAB. High electrical flame off (EFO) current would obtain a uniform Pd layer around the FAB. In contrast, Li et al. [16] also reported that Pd distribution was not uniform on the FAB after formation, but Hsu et al. [17] found that a higher EFO current leads to a more random Pd distribution on the FAB surface. The relationship between Pd distribution and bonding strength was also not investigated in the reported literatures. In this paper, we aim to investigate the distribution of Pd over the surface of FAB with different EFO current and firing time, and to clarify its effects on the bonding strength. Systematic experiments were conducted to trace how Pd and Cu move during FAB formation. A corresponding model was first proposed for better understanding of Pd-coated Cu wire bonding. 2 Experimental procedure Commercially Pd-coated Cu wire (Nippon Micrometal Corporation) was adopted in this study, and bare Cu wire and Pdalloyed Cu wire (Yantai Yesno Electronic Materials Corporation Ltd.) were used for comparison. The diameter of all the wires was 20 µm. An IconnPSProCu wire bonder (Kulicke and Soffa company) was used to fabricate FAB and bonded samples. The ratio of FAB diameter to wire diameter was fixed as 1.5, which meant that all the formed FABs had a diameter of 30 µm on 20 µm wires. The EFO current used for 20 µm Pd-coated Cu wire was in the range of ma. The EFO current of forming Cu FABs need larger EFO current than Au wire (20 30 ma), because the melting point of Cu is higher than that of Au. However, when the EFO current is enlarged to 100 ma, there are a lot of strawberry balls and golf balls, which can t be accepted. In such a case, when the EFO current is input, the firing time wold be automatically matched by the bonding machine. For Pd-coated Cu wire, the EFO current and its corresponding firing time to form FAB are listed in Table 1. A series of FABs were fabricated to investigate the distribution trend of the Pd element. The presence of Pd on FAB surface could be verified using a chemical etching solution, such as ferric chloride (1 wt% FeCl 3, 40 s), to etch away the exposed Cu regions and leave the Pd-rich region, as Pd does not dissolve in ferric chloride solution Besides FAB samples, practical bonded ball samples of Pd-coated Cu wire with different EFO currents were also fabricated on Al pad using the FAB parameters as in Table 1. These parameters were also optimized to fabricate bonding samples of a bare Cu wire and Pd-alloyed Cu wire (with an addition of 0.3% Pd wt) for comparison. All the bonded ball samples of Pd-coated Cu wire, bare Cu wire and Pd-alloyed Cu wire were aged at 200 C for 24 h in the N 2 atmosphere. The size of Al pad was µm, and the pad pitch was 100 µm. Furthermore, under the Al pad, there was a Ta layer, a low-k layer and a Cu layer respectively on the bonding substrate. Figure 1a shows a wire bonding sample of Pdcoated Cu wire before molding, Fig. 1b is the cross-sectional image of one bonded ball. The initial FAB diameter is about 30 µm when using a 20 µm Pd-coated Cu wire. According to Fig. 1b, the bonded ball has a diameter of 40 µm and a height of 10 µm respectively. Therefore, it can be calculated that about 44.5% of the spherical FAB surface took part in the process to form a bonding interface. The ball shear tests were performed using a Dage 4000 instrument to measure the shear strength of the bond. At least 15 ball bonds were sheared for each sample. Microstructural characterization on FABs and bonding samples were conducted using a Zeiss supra 55 scanning electron microscope (SEM), equipped with an Oxford instrument energy dispersive X-ray (EDX) analyzer. 3 Results and discussions 3.1 The effect of EFO current on Pd distribution Generally, Pd-coated Cu wire with a larger diameter has a thicker Pd layer during fabrication. When the wire diameter Fig. 1 Bonded ball a before molding and b its cross-section morphology Table 1 EFO current and its corresponding firing time to form FAB at a FAB/wire diameter ratio of 1.5 EFO current (ma) Firing time (µs)

3 13776 is smaller, the thickness of the surface Pd layer decreases. After wire bonding, the amount of Pd taking part in the protection of bonded ball also became smaller. Therefore, there was a higher risk for thin wires to have a larger exposed Cu region on the bonded balls [17]. Figure 2a shows the crosssection image of the 20 µm Pd-coated Cu wire used in this study. It has a concentric Cu core and a uniform Pd surface layer, as verified by EDX mapping in Fig. 2b, c respectively. Different EFO currents ( ma) were used to fire FABs as shown in Fig. 3. Since the FAB diameter is fixed as 30 µm, the input of each EFO current results in an optimized firing time automatically. It was found that when the EFO current is below 40 or above 90 ma, the formed FABs were not symmetrical and their surfaces were not smooth, which will not be discussed in this work (shown in Fig. 3a, h). After FAB formation, part of the ball surface could be covered by the Pd layer, while others exposed Cu directly. Since 1 wt% FeCl 3 solution can corrode Cu but not Pd, the distribution of Pd over the FABs surface could be detected by this corrosion method (1 wt% FeCl 3 for 40 s). The dotoutlined area on FAB in Fig. 3 is the exposed Cu area (corroded area) without a Pd surface protective layer. When the EFO current was 40 ma (Fig. 3b), the exposed Cu region was concentrated on the spherical crown of the FAB, which was far away from the ball neck. The area of the exposed Cu region seems symmetrical, and occupies about 10% of the surface area of the ball. It means that both the 10% exposed Cu region and surrounding Pd region would take part in the following bonding process to form an interface. This is because about 44.5% of the FAB surface is involved in this process, as estimated from the bonded sample of Fig. 1b. When the EFO current turned to 50 ma, as shown in Fig. 3c, the exposed Cu region became bigger and asymmetrical, locating at the bottom and side of the ball. The area of the exposed Cu region was calculated at about 15%, which was much lesser than 44.5%. As the EFO current continued to increase to 60 ma, the area of the exposed Cu region increased to about 20% of the FAB spherical surface (see Fig. 3d). In the case of 70 ma in Fig. 3e, the area of the exposed Cu region was calculated for about 25%, which is also less than 44.5%. However, the boundary of exposed Cu region was extended to the equator of the FAB. At a higher EFO current of 80 and 90 ma, larger exposed Cu regions were generated as shown in Fig. 3f, g, which is decentered and larger with a boundary close to the equator of the ball. Although the area of the exposed Cu region at 80 and 90 ma was close to 44.5%, after wire bonding there was still the exposed Cu region on the bonded ball. It can be concluded from Fig. 3 that higher the EFO current, lesser the Pd coverage over the FAB surface. Both the Fig. 2 Cross-section image of a 20 µm Pd-coated Cu wire, b Cu core mapping and c coated Pd mapping Fig. 3 Distribution of Pd over the FABs of the Pd-coated Cu wire with different EFO current a 30 ma, b 40 ma, c 50 ma, d 60 ma, e 70 ma, f 80 ma, g 90 ma and h 100 ma after corroded by 1 wt% FeCl 3 for 40 s. The actual line is the symmetry axis of the FAB, while the dotted line indicates the borderline of the Cu exposed area

4 EFO firing time and the EFO current need to be considered in wire bonding [18]. During FAB firing, the molten metal will solidify to form FABs under the effect of surface tension. At a temperature of T, the surface tension coefficient of Cu and Pd can be described in Eqs. 1 and 2, respectively [19] as: σ(cu) = (T 1085) mn m σ(pd) = (T 1552) mn m (2) When Pd and Cu are both in molten state, the surface tension coefficient of Pd is bigger than that of Cu based on calculations, which means that Pd is easier to shrink, to reduce the surface area of itself than that of Cu at melting state. Under the same volume condition, the surface area of the Pd-covered ball is smaller than that of the original Cu wire. So, at the point of FAB formation, Pd will cover Cu on the surface. Furthermore, the density of Pd is greater than the density of Cu. Therefore, during the process of solidification, Pd will move down to solidify, as time increases. Thus, a longer EFO firing time will result in a larger Pd coverage on the FAB surface. According to Table 1, a higher EFO current has a shorter firing time. Therefore, the coverage of Pd on FAB surface decreases as the EFO current increases, which was verified by our experiments as shown in Fig. 3. Coverage of Pd surface layer can protect the ball from oxidation in wire bonding. When the Pd surface region is (1) > 55.5%, as shown in Fig. 3b e, the Pd layer can cover all bonded ball surface for protection, and the Pd layer at the bottom of FAB ball will also participate in the interfacial bonding reaction. However, when its coverage is decentered with a boundary close to the equator of the ball, as shown in Fig. 3f, the exposed Cu region will be left after wire bonding, which is not good for service reliability. 3.2 The effect of firing time on Pd distribution In order to observe the FAB forming process, various firing times were used in experiments at an interval of 20 µs with the 80 ma EFO current, and the morphology of formed FAB after corrosion is shown in Fig. 4. When using the ball wedge bonding type, there is a slope at the end of the Pd-coated Cu wire before FAB formation. This can be seen in Fig. 4a. With 20 µs of firing time, the end of the Cu core was corroded, and the surrounding coated Pd layer remained. This firing time is too short to form FAB as yet. When the firing time was 60 µs, as shown in Fig. 4b, a small FAB was formed with a diameter, the same as that of the wire. Also, the Cu core was corroded and the coated Pd layer was left. At a firing time of 160 and 180 µs, spherical FAB balls were formed. But the Pd surface layer only existed at the upper neck area, while the asymmetrically exposed Cu region was corroded at the bottom of the FAB (see Fig. 4c, d). When the firing time increased to 260 and 300 µs, the Fig. 4 Distribution of Pd over FABs of Pd-coated Cu wire (after corrosion) with EFO current of 80 ma and at firing time of a 20 µs, b 60 µs, c 160 µs, d 180 µs, e 260 µs and f 300 µs. All the scale bars represent a length of 10 µm

5 13778 corroded morphology changed much, as shown in Fig. 4e, f. No continuous smooth Pd layer can be observed any more as those in Figs. 3 and 4c, a new chapped morphology covered all FAB surfaces. It is believed that the sunken cracks of this chapped structure should be the exposed Cu region, while the raised lands are the Pd-rich area. That is to say, with such long firing times, the surface Pd layer has melted with Cu core to form PdCu alloy on the surface of FAB. The distribution of Pd can be verified by elemental mapping using energy dispersive spectrometer (EDS) as shown in Fig. 5. At a firing time of 160 µs (Fig. 5a), the Pd element can be detected only at the upper neck area of the FAB but not the bottom of the ball (Fig. 5c), which is consistent with Fig. 4c, forming a continuous smooth Pd surface layer. However, in the case of 260 µs in Fig. 5d, Pd distributes all over the ball area, together with Cu, as shown in Fig. 5e, f. Moreover, before corrosion the as-formed FAB has a smooth surface (see Fig. 5d), while it turns into a chapped surface after corrosion, as in Fig. 4e. Combining the co-existence of Pd and Cu elements at the surface, the formation of PdCu alloy under a long firing time (> 260 µs) is reasonable. 3.3 The model of FAB formation Before the first bond is formed in the wire bonding process, the free end of the wire is divided into three parts: a FAB, a heat-affected zone (HAZ), and as-drawn wire due to the effect of EFO [20]. FAB formation is achieved by ionization of the air gap by the EFO process. At the end of discharge, the molten spherical ball starts to cool and then solidifies to form a FAB [21]. The majority of wire bonds in the electrical package are formed with ball wedge bonds. After the first ball bonding with FAB, the second one will bond the opposite end of the wire loop to the metal of the substrate. Following the second bond formation, the bonding tool (capillary) will break the wire tail and rise up further to the ball formation height. The Pd at the tail of bond wire is asymmetrical, which is due to the shape of the capillary and the second bond formation. Furthermore, the asymmetrical Pd will affect the Pd distribution during the FAB formation, which has been verified by experimental observations (see Fig. 4) and discussed in former sections. The schematic illustration of Pd distribution over FAB was shown in Fig. 6. The light color represents the coated Pd layer on the surface, and the grey color inside the wire is Cu or PdCu alloy (with few diffused Pd). Because the alloyed Pd in Cu is in solid solution state and the content of Pd is very small, the same color of bare Cu is used for PdCu alloy in Fig. 6. Before the wire tail is transformed into FAB, the coated Pd layer and Cu core were asymmetrical as shown in Fig. 6a (also see the experimental results in Fig. 4a). The Pd and Cu metals have a different melting point. FAB starts to form after the Pd-coated Cu wire end is heated and subsequently melted by the low-energy plasma discharge. The molten wire turns into a semi-spherical ball under the effect of surface tension. As the melting point of Pd is higher than that of Cu, the Pd turned to solid before Cu and the firing time was not long enough for Pd and Cu for an inter-diffusion. Thus, there was a Pd-rich layer over the Cu core, as shown in Fig. 6b, in which there exists primary exposed Cu regions when the wire diameter was the same as the ball diameter. As the firing time increased, the FAB turned bigger with more molten materials. There was a Pdrich layer over the most places of the FAB. The location of the exposed Cu regions still existed, as shown in Fig. 6c. The higher the temperature is, the smaller is the surface tension, according to Eqs. 1 and 2. When FAB continued to enlarge, the effect of surface tension became weaker. Pd moved down due to a higher density than Cu. There was enough time for Cu and Pd to inter-diffuse, which brought a thick PdCu alloy layer over the FAB (see Fig. 6d). As firing time increases, the inter-diffusion of Pd into Cu will be enhanced to form the PdCu alloy on the surface of FAB. Using this model, the distribution of Pd on the surface of FAB can be described Fig. 5 Elemental mapping of Pd-coated Cu wire before corrosion fabricated with a EFO current of 80 ma at firing time of a 160 µs with b Cu mapping and c Pd mapping, and d 260 µs with e Cu mapping and f Pd mapping

6 13779 Fig. 6 Schematic illustration of Pd distribution over FAB in the ball wedge wire bonding process clearly. This is also verified by the experimental investigations as explained in the former section (see Fig. 4). 3.4 The effect of Pd on bonding strength The effect of Pd distribution on the shear strength of bonding samples was further investigated. A series of bonded samples were prepared using different EFO currents, as shown in Table 1 and Fig. 3. Ball shear tests were conducted on these samples as compared to the bare Cu wire sample and the Pd-alloyed Cu wire sample. The measured shear strengths are shown in Fig. 7, which is presented using the JMP software to evaluate their relativity. The red boxplots are the quantiles of each set of data from one kind of sample, which are in sequential order, the maximum, the third quartile, the median, the first quartile and the minimum from top to bottom. The green diamonds represent the means for one-way anova, displaying the lower 95% and the upper 95% percentiles. On the right side of Fig. 7, the center of the Tukey Kramer circle represents the average value. The diameter of the circle represents the 95% confidence interval. If the two circles intersect and the confidence intervals overlap each other, it means that the average values are not significantly different for the two represented samples. On Fig. 7 JMP analysis of ball shear strength for bare Cu wire bonding, Pd-alloyed Cu wire bonding, and Pd-coated Cu wire bonding at different EFO currents

7 13780 Table 2 The average value of shear strength for different wire bonding samples Bonding samples Pd-alloyed Cu wire Bare Cu wire Pd-coated Cu wire 40 ma 50 ma 60 ma 70 ma 80 ma 90 ma Average ball shear strength (g) the contrary, if the two circles do not intersect, it indicates that their average values are significantly different. According to Fig. 7, the ball shear strength of the bare Cu wire sample is in the range of g, which is lower than those of both Pd-alloyed and Pd-coated Cu wire samples. Also, the circle representing the bare Cu wire does not intersect any other circles on the right side of Fig. 7. This means that the average ball shear strength of a bare Cu wire sample is significantly different from that of Pd alloyed and Pd-coated Cu wire samples. The average ball shear strengths (the circle center in Fig. 7) of different bonding samples are summarized in Table 2. According to Fig. 7 and Table 2, it is clear that a Pd addition can increase the ball shear strength of bare Cu wire bonding, although its value (10 15 g) is still higher than the industry standard (8 g for 30 µm FABs). The average shear strength of Pd-alloyed Cu wire bonding (19.7 g) is marginally lower than that of Pd-coated Cu wire bonding sample ( g), fabricated at EFO current of ma. This indicates that coated Pd on a Cu wire surface is more effective than alloyed Pd inside a Cu wire during wire bonding. However, the Pd-coated Cu wire sample fabricated at 40 ma EFO current has the smallest shear strength among all Pd-coated Cu wire samples. According to Fig. 3, the FAB fabricated at 40 ma has the largest Pd coverage on the surface. This surface Pd layer will get in touch with the Al substrate first, during wire bonding. Since Pd does not take part in the interfacial reaction between Cu and Al, it may suppress the formation of Cu Al IMCs, thereby reducing the interfacial strength. Combining the Pd distribution as illustrated in Fig. 3, and the shear strength illustrated in Fig. 7, it is concluded that an appropriate coverage of Pd at the FAB surface is critical for a better bonding reliability, which need to be further investigated from the view point of interfacial reaction. 4 Conclusions Unlike bare Au and Cu wires, the FAB formation in a Pdcoated Cu wire has the problem of Pd distribution over the FAB due to the different melting points and densities in the two materials. By changing the firing current and time, a model of Pd distribution over FAB is first proposed for the ball wedge wire bonding process. Due to the shape of the capillary and the wedge type of the second bond, the Pd at the tail of the bond wire was asymmetrical, which can affect the Pd distribution over the FAB. At the same ball/ wire diameter ratio, the bigger the EFO current, the larger the exposed Cu regions over FAB surface. Under the same EFO current, the Pd distribution varies with EFO firing time. At a short firing time (< 180 µs at EFO current of 80 ma), a Pd-rich layer is formed on the FAB surface. When the firing time was increased to 260 µs, PdCu alloy was formed over the entire surface of the FAB due to inter-diffusion, which acts as a protective shield layer against corrosion attack by halogen ions in molding compounds. Pd addition can increase the ball shear strength of bare Cu wire bonding by > 50%, from 12 to over 20 g. An appropriate coverage of Pd at the FAB surface is critical for a better bonding strength, noticing that the FAB with the largest Pd coverage has the smallest ball shear strength. In conclusion, the control of Pd distribution over FAB is very critical for a high quality bonding interface. Acknowledgements This work was financially supported by the National Key R&D Program of China (Grant No. 2017YFB ), the National Natural Science Foundation of China (Grant No ), and the Fund of the State Key Laboratory of Advanced Technologies for Comprehensive Utilization of Platinum Metals (Grant No. SKL-SPM ). References 1. R. Guo, L. Gao, D. Mao, Microelectron. Reliab. 54, 2550 (2014) 2. Z.W. Zhong, Microelectron. Reliab. 51, 4 (2011) 3. P. Chauhan, Z.W. Zhong, M. Pecht, J. Electron. Mater. 42, 2415 (2013) 4. T. Uno, K. Kimura, T. Yamada, in Microelectronics and Packaging Conference, T. Uno, Microelectron. Reliab. 51, 148 (2011) 6. H. Xu, C. Liu, V.V. Silberschmidt, S.S. Pramana, T.J. White, Z. Chen, V.L. Acoff, Acta Mater. 59, 5661 (2011) 7. B.K. Appelt, A. Tseng, C.H. Chen, Microelectron. Reliab. 51, 13 (2011) 8. C.S. Goh, W.L.E. Chong, T.K. Lee et al., Crystals 3, 391 (2013) 9. S. Kaimori, T. Nonaka, A. Mizoguchi, IEEE Trans. Adv. Packag. 29, (2006) 10. G.S. Oehrlein, G.J. Scilla, S. Jeng, Appl. Phys. Lett. 52, 907 (1988) 11. B. Zhang, K. Qian, T. Wang et al., Semicond. Tech. 35, 662 (2010) 12. E.M. Schindel-Bidinelli, Int. J. Adhes. Adhes. 16, 33 (1996) 13. C.J. Vath, M. Gunasekaran, R. Malliah, in Electronics Packaging Technology Conference, EPTC 09, 11th, 2009, pp L.J. Tang, H.M. Ho, Y.J. Zhang, Y.M. Lee, C.W. Lee, in Electronics Packaging Technology Conference (EPTC), 12th, 2010, pp

8 15. L.J. Tang, H.M. Ho, W. Koh, Y.J. Zhang, K.S. Goh, C.S. Huang, Y.T. Yu, in Electronic Components and Technology Conference (ECTC), IEEE 61st, 2011, pp J. Li, X. Zhang, L. Liu et al., J. Microelectromech. Syst. 22, 560 (2013) 17. H.C. Hsu, L.M. Chu, W.Y. Chang et al., J. Mater. Sci.-Mater. Electron. 24, 3594 (2013) 18. W.H. Song, M. Mayer, Y. Zhou et al., Microelectron. Reliab. 52, 2744 (2012) P. Haonan, W. Jiaji, Y. Hongkun, Semicond. Tech. 38, 623 (2013) 20. W. Koh, T.K. Lee, H.S. Ng, K.S. Goh, H.M. Ho, in Electronic Packaging Technology and High-Density Packaging (ICEPT- HDP), 2011, pp C.J. Hang, W.H. Song, I. Lum, M. Mayer, Y. Zhou, C.Q. Wang, J.T. Moon, J. Persic, Microelectron. Eng. 86, 2094 (2009)