Microstructure control of unidirectional growth of g-Cu
6
Sn
5
in microbumps on
h1 1 1i oriented and nanotwinned Cu
Han-wen Lin
a, Chia-ling Lu
a, Chien-min Liu
a, Chih Chen
a,⇑, Delphic Chen
b,
Jui-Chao Kuo
b, K.N. Tu
caDepartment of Materials Science & Engineering, National Chiao Tung University, Hsinchu, Taiwan, ROC bDepartment of Materials Science & Engineering, National Cheng Kung University, Tainan, Taiwan, ROC cDepartment of Materials Science and Engineering, University of California at Los Angeles, Los Angeles, CA 90095, USA
Received 22 September 2012; received in revised form 25 March 2013; accepted 26 April 2013 Available online 1 June 2013
Abstract
Anisotropic microstructure is becoming a critical issue in microbumps used in 3-D integrated circuit packaging. We report here an experimental approach for controlling the microstructure of g-Cu6Sn5intermetallic compound in microbumps by usingh1 1 1i oriented
and nanotwinned Cu pads as the under-bump-metallization. By electroplating arrays of large numbers ofh1 1 1i oriented and nanotwin-ned Cu pads and by electroplating the Sn2.3Ag solder on the pads, we form g-Cu6Sn5in the reflow at 260°C for 1 min. The g-Cu6Sn5
showed a highly preferential growth along theh0 0 0 1i direction. As reflow time is extended, the preferred texture of g-Cu6Sn5changed to
{2 1 1 3}. The results indicate that we can control the uniform microstructure of g-Cu6Sn5intermetallic by controlling the microstructure
of the Cu under-bump-metallization.
Ó 2013 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
Keywords: Intermetallic compounds; Soldering; Copper
1. Introduction
Approaching the end of Moore’s law for the large-scale integration of circuits in Si chip technology, a paradigm change from 2-D to 3-D integrated circuits (ICs) is occur-ring in the microelectronics industry. In essence, 3-D IC is a way to extend Moore’s law by bringing packaging tech-nology closer to chip techtech-nology. While Moore’s law per-tains to 2-D chip technology, we ask if there is a similar law for packaging technology; and if so, what is its future? On the basis of the scaling of the density of solder bumps on a chip surface, we find that there are at the least four more generations to go. Thus the future of chip technology can be extended if it is combined with packaging technol-ogy. For example, the diameter of a flip chip solder bump
is about 100 lm today. We expect that it can be reduced to 1 lm, which will increase the density of bumps per unit area by four orders of magnitude. However, at the same time the volume of a solder bump is reduced by six orders of magnitude. This reduction becomes a yield and reliabil-ity issue from the point of view of the microstructure of the solder bump.
We consider a simpler case of reducing the bump diam-eter from 100 to 10 lm and assume a grain size of 10 lm. There is only 1 grain in the 10 lm diameter solder bump, but there are 1000 grains in the 100 lm diameter solder bump. In the latter case, we can assume a randomly ori-ented microstructure in every solder bump, so that all the solder bumps on a Si chip, over several hundred or thou-sands of them, have isotropic physical properties. In the former case, however, with 1 grain in a bump, we cannot do so and we must consider the anisotropic microstructure and properties of solder bumps—especially when we
con-1359-6454/$36.00Ó 2013 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.actamat.2013.04.056
⇑ Corresponding author. Tel.: +886 3 5731814; fax: +886 3 5724727. E-mail address:chih@mail.nctu.edu.tw(C. Chen).
www.elsevier.com/locate/actamat
sider early failure of solder bumps. Thus, microstructure control of small solder bumps becomes a critical issue. This consideration, when applied to 1 lm diameter solder microbumps, will be even more serious.
When a microbump has a very small number of grains, variation of grain orientation or microstructure can result in the microbump having a wide range of anisotropic prop-erties, which is very undesirable from the point of view of yield and reliability. This is because certain grain orienta-tions allow fast diffusion, and can induce early failure due to electromigration, for instance. Since there is a large number of microsolder bumps on a stack of chips in a 3-D IC, how to achieve a uniform microstructure with thou-sands of bumps on a chip is a critical issue for the future of 3-D IC manufacturing.
Controlling the microstructure of solder bumps is non-trivial because of reflow. Reflow means that the solder bump is melted to achieve chip-to-chip or chip-to-substrate joints. When a molten solder bump solidifies, it is very dif-ficult to control the solidified microstructure. We report here that we can achieve a uniform microstructure of an array of a large number of microbumps by usingh1 1 1i ori-ented and nanotwinned Cu as the under-bump-metalliza-tion (UBM) and have obtained a unidirecunder-bump-metalliza-tional growth
of the intermetallic compound (IMC) in all the
microbumps.
Why is IMC growth a concern for microbumps? While microbump volume has been reduced by 3–6 orders of magnitude, the reflow condition, typically 250°C for 1 min for Pb-free solder, has not changed. This means that the IMC fraction in a microbump has increased dramati-cally. In fact, due to the need for several reflows in process-ing 3-D ICs and the need for solid-state agprocess-ing because of reliability concerns, the entire microbump can become an IMC, and there will be no unreacted solder. This indicates that what we have discussed in the previous paragraph on microstructure control actually concerns the control required to achieve a uniform microstructure of IMC in thousands of microbumps. Of all the IMCs encountered in solder joints, the most important is Cu6Sn5because of
the widespread use of Sn-based Pb-free solder and Cu UBM.
Regarding microstructure control of IMCs, the unidi-rectional growth of g- and g0-Cu
6Sn5 has been reported
to occur on single-crystal Cu substrates. Suh et al. found preferential growth behavior of g0-Cu
6Sn5 on (0 0 1)
sin-gle-crystal Cu[1,2]. Zou et al. found a very strong texture of g0-Cu
6Sn5on (0 0 1), (0 1 1), (1 1 1) and (1 2 3)
single-crys-tal Cu substrates[3,4]. On polycrystalline Cu, Kumar et al. have reported the formation of randomly oriented grains of Cu6Sn5 [5]. However, in microelectronic packaging
tech-nology, polycrystalline Cu UBM is typically produced by electroplating Cu metals on Si wafers, resulting in uncon-trolled growth of IMC if the Cu UBM is polycrystalline.
We report here that the next best substrate to single-crystal Cu is h1 1 1i oriented and nanotwinned Cu, on which we have obtained unidirectional growth of IMC with
tilt-type grain boundaries between the IMC grains. We pre-pared the h1 1 1i oriented and nanotwinned Cu UBM by electroplating. Sn2.3Ag was used as solder material. The orientation relationship between the oriented Cu and the oriented Cu6Sn5has been examined and is reported here.
2. Experimental
To prepare the h1 1 1i oriented and nanotwinned Cu pads, we used CuSO4-based electroplating solution with
suitable additives. The current density was 10 or 80 mA cm2, and the rotation speed was set to about 800 rpm. Then, the Sn2.3Ag solder alloy was electroplated on the Cu.
To study IMC formation, the bumped Si die was reflowed at 260°C for 1–5 min to grow Cu6Sn5. Then, after
cooling in the air, the sample was ground by abrasive papers of #1000, #2000, and #4000 followed by polishing with Al2O3powder of 1.0 and 0.3 lm. Finally, colloidal
sil-ica was used to remove the surface layer, which might be damaged during the polishing process. The morphology and the orientation image map in both cross-section view and top view after grinding and polishing were examined. To study microsolder joints, we placed two bumped Si chips face-to-face and reflowed them for several minutes to make an array of microsolder joints between them. Cross-sections of the joined samples were prepared by grinding using #400, #1000, #2500 and #4000 abrasive papers after air cooling, and then polishing with 1 and 0.3 lm Al2O3 and colloidal silica. The focused ion beam
(FIB) technique was adopted for cross-sectional observa-tion. Images of solder joints were taken with a JEOL 7001 field emission scanning electron microscope. Orienta-tion maps, inverse pole figures and pole figures were col-lected with an EDAX electron back-scatter diffraction (EBSD) system. It should be noted that a 3-D coordinate system was used to explain the orientation of grains in EBSD. In this paper, the direction out of the Cu pads was the normal direction (ND). The directions along the surface of the Cu pads were the rolling direction (RD) and the transverse direction (TD). These three directions are perpendicular to each other. X-ray diffraction (XRD) and transmission electron microscopy (TEM) were also adopted to verify the experimental results.
Monoclinic g0-Cu
6Sn5 was found by Larsson et al. in
1994 [6]. Ever since then, the crystal structure of Cu6Sn5
has been controversial. Larsson has reported that the trans-formation temperature of Cu6Sn5 from g0-Cu6Sn5 to
g-Cu6Sn5 is 186°C. However, the reflow of solder joints
involves heating up to 260°C and cooling down to room temperature, and so the actual stable phase after joint for-mation is unclear. Ghosh and Asta reported the kinetics and energy of the g-Cu6Sn5M g0-Cu6Sn5 transformation [7]. Laurila et al. showed that the time required for the g to g0 transformation was insufficient with typical cooling
rates in reflow [8,9]. Therefore, the g- Cu6Sn5 would
time–temperature–transformation diagram of the phase transformation of Cu6Sn5 and concluded that at under
70°C the g-Cu6Sn5in the joint could hardly transform into
monoclinic structure [10]. Nogita et al. have also found that Ni has a stabilizing effect on the structure of g-Cu6Sn5 [11,12]. Schwingenschlo¨gl et al. have further proven the sta-bility of g-Cu6Sn5 by first-principles calculations [13].
Hence, the orientation relationship between g-Cu6Sn5
and Cu is important, as reported here. 3. Results
3.1.h1 1 1i oriented and nanotwinned Cu pad
Nanotwinned Cu was first reported by Lu et al., and possesses the unique combination of properties of excep-tionally strong mechanical strength, good ductility and very little loss of electrical conductivity [14,15]. Neverthe-less, the Cu grains with nanotwins were randomly oriented. In our experiments involving electroplating of Cu, the {1 1 1} twinning plane of the nanotwins is parallel to the Si wafer surface, and thus we haveh1 1 1i oriented nanotw-ins rather than randomly oriented nanotwnanotw-ins. Further-more, we can electroplate the oriented nanotwins in arrays of a large number of Cu pads. The diameter and thickness of each Cu pad are 100 and 20 lm, respectively. To examine the orientation distribution of the nanotw-ins, we cut a flat surface on the Cu pads by FIB. The cut
flat surface 50 lm long and 20 lm wide is shown in Fig. 1a. The square marked by the white dashed line in Fig. 1a indicates the area examined by EBSD (see Fig. 1b) and shows that the colors of Cu grains were all blue or blue purple in the orientation image map. This indi-cates that they all have approximately the same orienta-tions. The inverse pole figure also shows the grain boundaries including twin boundaries on the top surface of the Cu pad. Referring toFig. 1d showing the orientation of Cu, the orientation of all the Cu surface grains on the pads was confirmed to be eitherh1 1 1i or very close to it. Since the twinning plane is also the {1 1 1} plane of face-centered cubic Cu, we have obtained oriented and nano-twinned Cu.
The grain size of electroplated Cu is about 2–5 lm. The {1 1 1} pole figure inFig. 1c proves that the pole at the nor-mal direction of surface is almost parallel toh1 1 1i.Fig. 1c suggests that the {1 1 1} pole of Cu is not exactly normal to the surface. However, the angular difference between the {1 1 1} pole of Cu and the ND of the Cu surface is less than 10°.
In addition, a TEM sample was prepared by FIB. Fig. 2a shows a bright-field image, enabling the grain of Cu to be observed. The diffraction patterns of three adja-cent grains were taken separately and then superimposed on each other inFig. 2b. This shows that these grains all had a pole of {1 1 1} and there were only a few degrees of rotation between different grains. The grain boundaries
Fig. 1. (a) Plan-view ofh1 1 1i unidirectional Cu Pads. (b) ND orientation map overlapped with image quality and grain boundaries. (c) The {1 1 1} pole figure, and (d) the reference figure (color coding).
between them are small-angle tilt-type grain boundaries. Although the Cu is still polycrystalline, theh1 1 1i preferred orientation is clear.
3.2. The preferential orientation relationship in solder bump structure
The Sn2.3Ag was electroplated on the h1 1 1i oriented and nanotwinned Cu pads. The reflow temperature was set to be 260°C. After 1 min reflow, the chip was carefully mounted with low-temperature epoxy. The sample was ground and polished for cross-sectional examination. FIB was used to perform a final cut. Since Cu, Cu6Sn5 and
SnAg solder were present, the advantage of adopting FIB is to make a cut precisely at the desired position and to a definitive thickness. Also, by performing the cut with a FIB, a strain-free surface can be revealed for all three phases.
Fig. 3a shows the FIB cross-section of a bump. The void inside the Cu pad was damaged during grinding and those
inside the SnAg were due to flux. The red rectanglar area in Fig. 3a was carefully examined by EBSD. Fig. 3b–d show respectively the orientation image maps of Sn, Cu6Sn5
and Cu. The map shows the orientation of these phases in the direction perpendicular to the silicon chip surface. In other words, the orientation relationships along the direction of growth of these phases are shown in the fig-ures. Once again, the color of the orientation image maps represents the orientation of the phases.
As displayed inFig. 3d, the grain of Cu is columnar and the color of Cu is blue with some interlaced bands on it. Corresponding to the color coding inFig. 3g, the orienta-tion of h1 1 1i Cu was again confirmed. The interlaced bands are confirmed as nanotwins with only 5–20 nm twin spacing. The grain size of intermetallics is about 8 lm in width and 2 lm in height. Each IMC scallop has a uniform but different color, indicating that each scallop is a single grain of Cu6Sn5. Nevertheless these intermetallics are all
shown in red and orange. This suggests that the intermetal-lics have a certain preferred orientation on the h1 1 1i Cu
Fig. 2. (a) TEM image of Cu pads viewed from the top. (b) The diffraction patterns of three adjacent Cu grains.
Fig. 3. (a) The cross-sectional area observed in the bump die structure. The ND orientation maps of (b) Sn, (c) Cu6Sn5and (d) unidirectional Cu. (e) The
{1 1 1} pole figure of Cu and (f) the {0 0 0 1} pole figure of g-Cu6Sn5. (g) The reference figure for Sn, Cu6Sn5and Cu. (For interpretation of the references to
pad. By reference toFig. 3g, the orientation of intermetal-lics along ND is close toh0 0 0 1i. To confirm this result, we examine the pole figure of {0 0 0 1} of intermetallics as well as that of {1 1 1} Cu. InFig. 3e, the {1 1 1} pole figure of Cu showed that the pole was mostly aligned with the ND. Combined with Fig. 3f, the {0 0 0 1} pole of Cu6Sn5 was
found also to center along the ND. As mentioned in Sec-tion 3.1, the {1 1 1} pole of Cu has a small distribuSec-tion around the normal of the Si surface, and hence the pole fig-ure of both Cu and Cu6Sn5showed some leaning behavior.
The amount of leaning is just a few degrees. With these results, the orientation relationship between the nanotwin-ned Cu and g-Cu6Sn5can be clearly identified as: {1 1 1} nt-Cu||{0 0 0 1}Cu6Sn5. This preferred growth relationship is
credible because the intermetallic contacted the Cu directly with little formation of Cu3Sn in between them because the
reflow time was only 1 min.
3.3. Effect of reflow time on the orientation relationship To obtain a general view of the orientation of every intermetallic grain on a single Cu pad, the surface of the intermetallic was revealed by grinding and polishing the top surface. Since the intermetallics in Section 3.2 were too small and hard to be reached precisely when ground by human hands, the sample was further reflowed for 4 min to make the intermetallic grains larger.
Fig. 4. (a) SEM image from the plan view of Cu6Sn5. (b) The ND orientation map of Cu6Sn5, and (c) the ND inverse pole figure of Cu6Sn5. (d) ND
orientation maps of Cu6Sn5IMCs on eight microbumps, and (e) the ND inverse pole figure for the IMCs in (d). Refer to the color coding inFig. 3g. (For
Fig. 4a shows the SEM image of a Cu pad with Sn and Cu6Sn5 on it. The islands on the Cu pad were Cu6Sn5
formed during reflow; between them was the residual Sn phase. With the ND orientation map ofFig. 4b, the colors of intermetallics were purple, orange and yellow. It should be noted that some noise signals are shown on the periph-eral area of the figure. When we revealed the plan-view of the intermetallic by grinding from the top of solder ball, the shape of hemisphere makes the peripheral area bumpy, which will cause some noise around the circumference of the pad.
These samples differ in a number of respects from those reflowed for only 1 min. First, more Cu6Sn5grains appear
yellow inFig. 4b. This indicates that the preferred texture of intermetallics has changed from {0 0 0 1} to {2 1 1 0}. The ND inverse pole figure (see Fig. 4c) shows that the major texture of the intermetallic has been changed to {2 1 1 3}. In this study, dozens of microbumps with {1 1 1} nanotwinned Cu pads were examined by EBSD. Fig. 4d shows the ND orientation maps for the Cu6Sn5
IMCs on eight microbumps. All of them had similar pre-ferred texture. By merging the Cu6Sn5 orientation data
from all these microbumps, we can construct the ND inverse pole figure (see Fig. 4e), which is obtained from the EBSD data in Fig. 4d. The ND preferred texture in Fig. 4d is almost identical to that inFig. 4b, indicating that the Cu6Sn5on every bump has the same preferred growth
behavior in the ND. There are reports in the literature which show that the morphology of Cu6Sn5would change
after reflow at higher temperatures or for longer times[1,2]. Another intermetallic, Cu3Sn, starts to grow at the
inter-face between Cu and Cu6Sn5. The formation of Cu3Sn
would break the preferential growth relationship between Cu6Sn5 and Cu because the Cu6Sn5 would then grow on
Cu3Sn, instead of on Cu.
3.4. The preferential orientation of IMC in a solder joint The preferential relationship between Cu6Sn5and Cu is
important in real packaging solder joints. The solder joints were fabricated by joining two bump-on-dies together. We flipped one chip with the Sn2.3Ag solder and the oriented and nanotwinned Cu pads on another chip, and these were then reflowed at 260°C for 3 min in total to make the microsolder joints. The sample was then ground and pol-ished, and finally cut by FIB.
Fig. 5a shows the ND orientation maps of the joints. The Cu pad is blue, because it has an oriented microstruc-ture from the bottom to the top. By reference toFig. 3g it can be confirmed that the entire column was allh1 1 1i ori-ented and contained nanotwins in each grain. There were some grains showing colors other than blue at the bottom. This might be a transition region from the seed layer to the nanotwinned Cu. The scallop-type Cu6Sn5appeared in red,
orange and yellow.Fig. 5b shows the two colored layers of Cu6Sn5 separately. The middle part in Fig. 5b is the
remaining Sn2.3Ag. As noted in the previous section, the orientation of Cu along the ND wash1 1 1i and that of Cu
6-Sn5was close to h0 0 0 1i.
When we reflowed the joint for 4 min, some of the inter-metallic from the top and from the bottom contacted each other and seemed to merge together as shown inFig. 6a. It should be noted that an unequal growth rate of Cu6Sn5on
each side of Cu pad was observed. This is mainly because of the unequal Cu flux during reflow. This phenomenon has been reported in our previous study [16]. In addition, the orientation map of Cu6Sn5in Fig. 6a shows that the
two intermetallic grains from the top and the bottom became one single grain as soon as they come into contact with each other. This seems to be a unique property of Cu
6-Sn5, which has a very fast grain growth rate or ripening
Fig. 5. The ND orientation maps of (a) the solder joints reflowed at 260°C for 3 min, and (b) the two colored layers of Cu6Sn5. Refer to the color coding
rate. We note that the intermetallic grains were small after 3 min of reflow. Hence, they needed to grow first and then they can contact with each other. The reaction when they are in contact with each other must be very fast since the transition can be finished in an additional minute, from 3 to 4 min, at 260°C.
The reflow time was further increased for an additional minute for another joint, giving a total reflow time of 5 min. The results are shown inFig. 6b. Most of the inter-metallics were merged together. The orientations of these intermetallics were still close to h0 0 0 1i in the ND. The magnified image inFig. 6c shows tiny white hexagonal unit cells that represent the exact orientation of each grain of g-Cu6Sn5. As illustrated by this figure, the {0 0 0 1} planes of
the hexagons were almost parallel to the Cu pads whether they were at the top chip or at the bottom chip. All the c-axes of these hexagonal unit cells lie along the direction perpendicular to the Cu pads. When two grains start to merge together, their orientations seem to merge too. 4. Discussion
4.1. Coherence between {0 0 0 1} g-Cu6Sn5and {1 1 1} Cu
There have been many reports about the preferential growth of g0-Cu
6Sn5 on Cu. The crystal structure of g0
-Cu6Sn5 is monoclinic (C2/c, a = 11.022 A˚ , b = 7.282 A˚,
c = 9.827 A˚ , b = 98.84°). However, in Section 1 we have noted that the g-Cu6Sn5would be the phase present in
sol-der joints. In this study, we mainly focused on the preferen-tial growth behavior between g-Cu6Sn5 (P63/mmc,
a = 4.2032 A˚ , c = 5.1107 A˚) and the h1 11i oriented Cu. The crystal structure of Cu is face-centered cubic with a = b = c = 3.615 A˚ .
Our experimental results in Section 3.2 indicate that the {0 0 0 1} plane of g-Cu6Sn5 is parallel to the {1 1 1}
plane of Cu during the early stages of reflow. After superimposing the Cu atoms of these two planes, we could hardly found any two directions that are of lower lattice mismatch through the whole plane of contact.
Thus it is difficult to explain this coherent relationship in terms of lattice matching. Nevertheless, in this study, although the Cu pads were made to have a h1 1 1i pre-ferred orientation and additionally the surfaces of the Cu pads were {1 1 1} planes, these pads still possessed a polycrystalline textured surface. Fig. 2b has already pro-ven that there were rotation or tilt-type behaviors between adjacent Cu grains. Hence, the arrangement of Cu atoms on the surface is not complete identical to sin-gle-crystal Cu. Moreover, the grain size of Cu6Sn5 was
larger than that of columnar Cu grains, indicating that the g-Cu6Sn5 must grow on several grains of Cu.
There-fore it is not possible to have a low lattice mismatch across the entire interface.
The soldering reaction between Cu and Sn forms Cu
6-Sn5. It is obvious that the bond energy between Cu and
Sn must be lower than that between Cu and Cu[11]. We should also discuss the coherence between the Sn in the g-Cu6Sn5and the Cu in theh1 1 1i oriented Cu. However,
it is also hard to achieve coherence since Sn atoms are more difficult to located in the hexagonal lattice of Cu6Sn5. The
Sn atoms in the hexagonal Cu6Sn5lie on the plane close to
{2 1 1 3}. Hence, we can now only conclude that because the {0 0 0 1} plane of g-Cu6Sn5 and the {1 1 1} plane of
Cu are the most densely packed atomic planes in both lat-tices, there is a high density of Cu–Sn bonds across the interface which lowers the interfacial energy.
4.2. The change in orientation of the intermetallic with reflow time
Inverse pole figures were constructed to show the orien-tation of g-Cu6Sn5with time inFig. 7. At the early stage of
reflow, up to 3 min, the g-Cu6Sn5s were mainlyh0 0 0 1i
ori-entated in Fig. 7a and b. When the reflow time was extended to 4 min, another orientation was found to play a more important role, as shown inFig. 7c. And finally, the h0 0 0 1i orientated Cu6Sn5 disappears on the joint
reflowed for 5 min (Fig. 7d). The texture of g-Cu6Sn5kept
at 260°C for 5 min lies between {0 0 0 1} and {2 1 1 0}. By
Fig. 6. The ND orientation maps of g-Cu6Sn5after the solder joints were reflowed at 260°C for (a) 4 min and (b) 5 min; (c) magnified image of (b) with
grain orientations indicated by the tiny hexagonal unit cells. Refer to the color coding inFig. 3g. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
using the software of the analysis system, we can define this new texture as {2 1 1 3}.
Fig. 7clearly shows the changes in texture with time. We can read the relative counts of {0 0 0 1} and {2 1 1 3} at each stage of reflow and make a statistical analyses. The results are shown inTable 1andFig. 8. We note that the relative counts are defined as the intensity of specific pole divided by the intensity of any random orientations (in units of “times random”).
Table 1shows the relative counts of each pole for vari-ous reflow durations. For 1 min reflow, the relative counts of {0 0 0 1} are 13.13 and that of {2 1 1 3} are 5.77. As the reflow time is extended, the intensity of {0 0 0 1} decreased down to only 0.17 and that of {2 1 1 3} increased up to 10.24. In Fig. 8, the trends of orientation change are shown. It is very clear that the intensity of {0 0 0 1} decreases rapidly after reflow for 3 min. And the texture of {2 1 1 3} increased slowly at the early stage of reflow and rapidly between 4 and 5 min of reflow.
The reason why the preferred orientation of g-Cu6Sn5
on {1 1 1} Cu might change is due to the formation of Cu
3-Sn. During the early stages of reflow, the g-Cu6Sn5forms
directly on Cu, making it a lower energy state by stacking its {0 0 0 1} plane on {1 1 1} Cu. However, as the time of reflow is extended, the g-Cu6Sn5grows larger and blocks
the diffusion path of Cu. At the same time, Cu3Sn forms
at the interface between Cu6Sn5and the Cu. Subsequently,
the formation of g-Cu6Sn5occurs not on the Cu but on the
Cu3Sn instead.
The mechanism of this transformation, including how the Cu6Sn5 is coherent with Cu3Sn and how the Cu6Sn5
changes its orientation in solids, is still unclear. Some pre-vious studies have shown that the morphology of Cu6Sn5
did change with reflow time. The formation of Cu3Sn
would break the coherence relationship between g-Cu6Sn5
and single-crystal Cu. None of this earlier research has shown the orientation of Cu6Sn5after the surface
morphol-ogy had disappeared. In this study, we showed that the preferential growth of scallop-type Cu6Sn5continued even
after the formation of Cu3Sn. However, we believe that
subsequent adequate thermal annealing would change the orientation of these intermetallics again. This is a much more complicated reaction since it involves not only the crystallization behavior of Cu6Sn5on Cu3Sn, but also the
phase transformation for Cu6Sn5 from g to g0. Further
study is required to establish the entire mechanism. How-ever, the {1 1 1} Cu is able to control the microstructure of Cu6Sn5during the fabrication stage and the early stages
of use. Therefore, the properties of each microbump can be controlled.
4.3. Effect ofh1 1 1i oriented Cu pads on the orientation of Cu6Sn5
In electroplating, factors including solution base, cur-rent (including densities and alternating or direct curcur-rent), temperature and additives can affect the quality of electro-plated metals. To examine the effect of Cu pads on the growth of preferred orientation of g-Cu6Sn5, other chips
with Cu pads electroplated with various current densities were prepared. It should be noted that other electroplating parameters were fixed and the pattern of Cu pads are sim-ilar on all the chips.
Fig. 9 shows the results for the Cu pads electroplated with various current densities. In Fig. 9a, the ion image was taken after the making a final cut by FIB. The Cu pad was electroplated with a lower current density at 1 ASD. The grain structure of Cu resembles that of typical electroplated Cu. There are some twin boundaries in it, though most of these are microtwins, with a few being nanotwins. The columnar grain appeared sometimes by chance. XRD analysis shows these pads still had large
Table 1
Relative counts of (0 0 0 1) and (2 1 1 3) at each stage of reflowing. 0001 2113 1 min 13.13 5.77 3 min 4.75 7.15 4 min 1.60 6.47 5 min 0.17 10.24
Fig. 8. The trend of orientation change of g-Cu6Sn5.
Fig. 7. The ND inverse pole figure of g-Cu6Sn5after the solder joints were
counts of about 100,000 about the {1 1 1} peak of Cu, as shown in Fig. 9c. In our experiments, we have adopted a chip with Cu pads electroplated at a higher current density of 8 ASD. As shown in Fig. 9b, the Cu grains were all columnar throughout the entire pad. Each grain contains a high density of nanotwins. The diameter of each column was 2–5 lm. Fig. 9c shows the XRD results of these Cu pads; there were more than 300,000 counts of the {1 1 1} peak of Cu. In our other electroplating process, we have adopted several settings of much higher and lower current densities. The best result we had was similar to that shown in Fig. 9b. Therefore, the optimum current density of 8 ASD was adopted to produce ah1 1 1i oriented and nano-twinned Cu.
After the Cu pads were produced, the Sn2.3Ag was then electroplated on them. These chips were reflowed at 260°C for up to 5 min to grow the intermetallic. Fig. 10a and b
show respectively the orientation image map and the inverse pole figures of the Cu6Sn5on the Cu pads
electro-plated at 1 ASD. Fig. 10c and d show respectively the results for the Cu6Sn5 on the Cu pads electroplated at 8
ASD. InFig. 10a, the colors of Cu6Sn5are more divergent.
There are now several grains blue and green, and the color divergence is greater than that inFig. 10c. In addition, the inverse pole figures inFig. 10b and d show that the g-Cu
6-Sn5was more concentrated on the oriented Cu pads of
bet-ter quality. This means that the Cu6Sn5 grains had no
strong preferred orientations on the Cu pad electroplated by 1 ASD. Again, after being reflowed at 260°C for 5 min, the texture of g-Cu6Sn5 was centralized at
{2 1 1 3}. Conversely, the orientations of Cu6Sn5 were
more random on the Cu pad electroplated with lower cur-rent densities.
5. Conclusions
h1 1 1i oriented and nanotwinned Cu has been produced by electroplating. The Cu grains are columnar with a diam-eter of about 2–5 lm. After electroplating Sn2.3Ag on the oriented Cu pads and followed by reflow at 260°C, the g-Cu6Sn5showed a preferential growth of {0 0 0 1} texture at
the early state of growth. As the time of reflow is extended, the preferred texture of Cu6Sn5changes to {2 1 1 3}. Since
theh1 1 1i oriented and nanotwinned Cu was polycrystal-line, the coherence between the g-Cu6Sn5and the Cu most
likely is achieved via a high density of Cu–Sn bonds across the interface between them. Electroplating parameters affect the quality of oriented Cu and therefore would affect the preferential growth of Cu6Sn5. Being able to control the
electroplating of h1 1 1i oriented and nanotwinned Cu means that we can control the orientation and in turn the microstructure of the intermetallic in microsolder joints. Acknowledgement
The authors gratefully acknowledge the financial sup-port of the National Science Council of the Republic of China (Grant No. NSC 101-2628-E-009-017-MY3).
Fig. 10. (a) The orientation image map and (b) the inverse pole figure of g-Cu6Sn5on 1 ASD Cu pads. (c) The orientation image map and (d) the
inverse pole figure of g-Cu6Sn5on 8 ASD Cu pads.
Fig. 9. The cross-sectional ion image of Cu pads electroplated with a current density of (a) 1 ASD and (b) 8 ASD. (c) The XRD results of Cu pads electroplated with current densities of 1 and 8 ASD.
References
[1]Suh JO, Tu KN, Tamura N. Appl Phys Lett 2007;91:051907. [2]Suh JO, Tu KN, Tamura N. J Appl Phys 2007;102:063511. [3]Zou HF, Yang HJ, Zhang ZF. Acta Mater 2008;56:2649. [4]Zou HF, Yang HJ, Zhang ZF. J Appl Phys 2009;106:113512. [5]Kumar V, Fang ZZ, Liang J, Dariavach N. Metall Mater Trans A
2006;37:2505.
[6]Larsson AK, Stenberg L, Lidin S. Acta Cryst 1994;B50:636. [7]Ghosh G, Asta M. J Mater Res 2005;20:3102.
[8]Laurila T, Vuorinen V, Paulasto-Kro¨ckel M. Mater Sci Eng R 2010;68:1.
[9]Laurila T, Vuorinen V, Kivilahti JK. Mater Sci Eng R 2005;49:1. [10]Nogita K, Gourlay CM, McDonald SD, Wu YQ, Read J, Gu QF.
Scripta Mater 2011;65:922.
[11]Nogita K. Intermetallics 2010;18:145.
[12]Nogita K, Nishimura T. Scripta Mater 2008;59:191.
[13]Schwingenschlo¨gl U, Paola CD, Nogita K, Gourlay CM. Appl Phys Lett 2010;96:061908.
[14]Lu L, Shen Y, Chen X, Qian L, Lu K. Science 2004;304:422. [15]Lu L, Chen X, Huang X, Lu K. Science 2009;323:607. [16]Kuo MY, Lin CK, Chen C, Tu KN. Intermetallics 2012;29:155.
