Reactions in Electrodeposited Cu/Sn and Cu/Ni/Sn Nanoscale Multilayers for Interconnects

Miniaturization of electronic devices has led to the development of 3D IC packages which require ultra-small-scale interconnections. Such small interconnects can be completely converted into Cu-Sn based intermetallic compounds (IMCs) after reflow. In an effort to improve IMC based interconnects, an attempt is made to add Ni to Cu-Sn-based IMCs. Multilayer interconnects consisting of stacks of Cu/Sn/Cu/Sn/Cu or Cu/Ni/Sn/Ni/Sn/Cu/Ni/Sn/Ni/Cu with Ni = 35 nm, 70 nm, and 150 nm were electrodeposited sequentially using copper pyrophosphate, tin methanesulfonic, and nickel Watts baths, respectively. These multilayer interconnects were investigated under room temperature aging conditions and for solid-liquid reactions, where the samples were subjected to 250 °C reflow for 60 s and also 300 °C for 3600 s. The progress of the reaction in the multilayers was monitored by using X-ray Diffraction, Scanning Electron Microscope, and Energy dispersive X-ray Spectroscopy. FIB-milled samples were also prepared for investigation under room temperature aging conditions. Results show that by inserting a 70 nanometres thick Ni layer between copper and tin, premature reaction between Cu and Sn at room temperature can be avoided. During short reflow, the addition of Ni suppresses formation of Cu3Sn IMC. With increasing Ni thickness, Cu consumption is decreased and Ni starts acting as a barrier layer. On the other hand, during long reflow, two types of IMC were found in the Cu/Ni/Sn samples which are the (Cu,Ni)6Sn5 and (Cu,Ni)3Sn, respectively. Details of the reaction sequence and mechanisms are discussed.


Introduction
As the electronic industry is moving towards miniaturization, three-dimensional (3-D) IC packages are being aggressively pursued. In 3-D packages, IC chips are stacked on top of each other to save space and to improve performance. In miniaturized packages, through-Si-vias (TSVs) and solder micro-bumps are used as interconnects, the size of which ranges from a few micrometres to about 50 micrometres [1,2]. In such small interconnects, Cu metallization and Sn-rich solder alloys react and transform completely into Cu-Sn intermetallic compounds (IMC), e.g., Cu 6 Sn 5 and Cu 3 Sn after reflow. Intermetallic compounds are generally known to be brittle and can adversely affect the reliability of the interconnection particularly in applications involving shock load. Therefore, there is a need to investigate ways to improve the properties of Cu-Sn IMC.
One way to address this could be to add a third element to Cu-Sn based IMCs. Out of different metallic elements studied, Ni is suggested to be a good candidate for addition into solders [3][4][5][6]. Studies have shown that Ni can suppress ε-Cu 3 Sn even when added at a small concentration [3][4][5]. Cu 3 Sn is considered to be more detrimental to reliability [7]. Furthermore, Ni shows marked solubility in η-Cu 6 Sn 5 IMC and can influence its properties. Nogita and Nishimura [6] reported that with the addition of 9% Ni to the (Cu,Ni) 6 Sn 5 IMC, polymorphic phase transformation of the Cu-Sn IMC from the hexagonal structure to the monoclinic structure can be prevented. Cu-Ni-Sn IMC has been reported to have higher hardness and Young's modulus [8][9][10][11].
Addition of Ni to interconnects has been done in the form of nanoparticles in the recent past. It has been shown [3] that Ni nanoparticles of~20 nm size undergoes reactive dissolution during reflow leading to in-situ alloying. It is, therefore, expected that if thin nanometric layers of Ni is introduced in Cu/Sn based interconnects in the form of multilayers, the former can get intermixed with Cu and Sn resulting in Ni alloyed IMC. This work investigates the formation of multilayered interconnects containing layers of Cu, Ni and Sn by electrodeposition. The samples are then subjected to intermixing under different conditions.
Considerable amount of work has been done in the past on the Cu/Sn system, both for solid state reactions and solid-liquid reactions. During solid state reactions, it has been found that Cu 6 Sn 5 IMC forms at room temperature. One study reported that it can form at´2˝C [12]. With the use of Auger electron spectroscopy and W markers, it was found revealed that Cu is the main diffusing species during the formation of η-Cu 6 Sn 5 at room temperature. Chopra and co-workers suggested that at the initial stage, the formation of η-Cu 6 Sn 5 took place in the Sn matrix in the form of precipitates through rapid diffusion of Cu into bulk Sn [13]. However, a recent study by Sobiech and co-workers showed that after five days of room temperature aging, Cu atoms diffuse preferentially along Sn grain boundaries where the η-Cu 6 Sn 5 phase formed first [14]. As has been found in the literature, Cu and Sn can mix readily at room temperature. This can lead to premature formation of IMC even before reflow. This may result in difficulties in obtaining a good interconnection, as Cu and Sn can be consumed completely prior to joint formation, particularly when the interconnections are small as in the case of 3D packages. The addition of the nanometric Ni layers between Cu and Sn as proposed above can help to prevent the premature intermixing before reflow. Ni layers should be thin enough though to undergo complete dissolution during reflow. Little information is available in the literature on the intermixing behaviour in Cu/Ni/Sn multilayered system.
In this work, we introduced ultrathin Ni layers of thickness 35 nm to 150 nm between Cu and Sn layers. The motivation is to prevent premature reaction between Cu and Sn prior to reflow. Another motivation is to form, during reflow, Cu-Sn IMCs alloyed with Ni which are expected to lead to better mechanical properties and hence improved reliability of interconnects. Cu/Sn and Cu/Ni/Sn multilayers system were prepared by electrodeposition. Both systems were investigated in terms of IMC formation after 1 day and 24 days of room temperature aging. The effect of Ni in the Cu/Sn system is also studied after both short and long reflow to investigate solid-liquid state reactions. Figures 1 and 2 show FESEM micrographs of FIB milled Cu/Sn and Cu/Ni/Sn samples after room temperature aging for one day and 24 days, respectively. The Cu/Ni/Sn samples used for solid state reactions contain Ni layers with 70 nm thickness. Layers with darker contrast are Sn while that with lighter contrast are Cu (Figure 1a,c). Micrographs at higher magnification (Figure 1b,d) show that the Cu layer contains fine columnar grains with axis perpendicular to the interface. The width of the columnar Cu grain is about 100-200 nm. Each columnar grain contains nanotwins with a twinning plane parallel to the interface. Sn layers (Figure 1b,d) contain more or less equiaxed grains with lateral width of about 1-2 µm. Within the dark grey tin layers, a phase with a lighter grey shade is seen, which corresponds to Cu 6 Sn 5 IMC (Figure 1b).

Solid State Reactions of Cu/Sn and Cu/Ni/Sn Multilayers
After the Cu/Sn sample was aged for one day, Cu 6 Sn 5 is seen as thin columns along the Sn grain boundaries (Figure 1a,b). Nanometric voids can be seen between Cu and Cu 6 Sn 5 IMCs layers. In the Cu/Ni/Sn system, IMC was seen to form at the Ni/Sn interface and also along Sn grain boundaries after one day of room temperature aging (Figure 1c,d). However, the extent of IMC formation in the Cu/Ni/Sn sample was smaller when compared to that of the Cu/Sn sample. After 24 days of room temperature aging, it is observed (Figure 2a) that in the Cu/Sn system, most of the Sn has reacted with Cu to form wider grains of Cu 6 Sn 5 IMC. Some of these grains have merged with neighbouring grains forming blocks of Cu 6 Sn 5 IMC separated from each other by some unreacted Sn. Additionally, a layer of Cu 6 Sn 5 IMC is seen at the Cu/Sn interface (Figure 2a,c). From these observations, it is suggested that Cu 6 Sn 5 IMC grew simultaneously in two directions, depicted by dashed arrow in Figure 2b: (i) it grew along the Sn grain boundary and (ii) grew into bulk Sn parallel to the Cu/Sn interface. Furthermore, voids formed in between Cu and Cu 6 Sn 5 IMC have grown in size (Figure 2b) as compared to those after one day of aging (Figure 1b). In the Cu/Ni/Sn system, it is observed that the extent of formation of IMC is much less. There is still a significant amount of unreacted Sn present in this sample. However, the extent of the growth along the grain boundaries is less in the Cu/Ni/Sn sample as compared with that of the Cu/Sn sample. The diffusion of Cu into Sn through the grain boundary is thought to have been impeded by Ni atoms that may have segregated to the grain boundaries ( Figure 2d). Table 1 shows the number of voids for each range of the void's diameter for both Cu/Sn and Cu/Ni/Sn samples aged at room temperature. The number of voids was calculated from four micrographs for each condition. After one day of room temperature aging, the Cu/Sn sample exhibits an average void number of 12 per micrometre along the interface, while the average void number in the Cu/Ni/Sn sample is 11 per micrometre. After 24 days of aging, the average number of voids in the Cu/Sn sample increased to 13.2 per micrometre. However, the number of voids with diameter smaller than 20 nm is found to be 172. In the Cu/Ni/Sn system, the average number of voids is 14.8 per micrometre with the number of voids with diameter smaller than 20 nm is 202. Though there is not much difference in the number of voids formed, fewer voids with larger diameter (voids with diameter bigger than 20 nm) are seen to have formed in the Cu/Ni/Sn sample ( Figure 2). The number of voids with diameter smaller than 20 nm is more in the Cu/Ni/Sn sample compared with that in Cu/Sn sample after 24 days of room temperature aging.   The nature of the IMC was found out from the reaction of elements present. In Figure 3a-d, the layers with darker contrast are Cu while that with lighter contrast are Sn. Within the light grey Sn layers, a darker grey phase identified as Cu 6 Sn 5 is seen in Cu/Ni/Sn sample ( Figure 3a). As for samples with Ni insertion in between Cu and Sn, no IMC formation is seen. The presence of Ni layers was confirmed with EDX line scans profiles [15]. It may be noted that these samples were stored at room temperature for 5 days prior to taking the micrographs. samples with Ni insertion in between Cu and Sn, no IMC formation is seen. The presence of Ni layers was confirmed with EDX line scans profiles [15]. It may be noted that these samples were stored at room temperature for 5 days prior to taking the micrographs.   Figure 3e). Discontinuous Cu3Sn IMC formed in the middle of the Cu6Sn5 phase due to reaction of the middle layer of copper with its surrounding tin during reflow. As for Figure 3f-h, layers with darker contrast represents Cu while lighter contrast represents the (Cu,Ni)6Sn5 phase. With the addition of Ni, it is seen that (Cu,Ni)6Sn5 IMC was formed regardless of the amount of Ni added. Cu/Ni-35/Sn has been converted completely into (Cu,Ni)6Sn5 ( Figure 3f). However, for Cu/Ni-70/Sn and Cu/Ni-150/Sn samples (Figure 3g,h), unreacted copper (darker contrast) in between (Cu,Ni)6Sn5 (lighter contrast), is still left after reflow. The thickness of the unreacted copper increased with increasing Ni thickness. This suggests that Ni starts acting as a barrier layer between Cu and Sn when its thickness increased to 70 nm.

Long Reflow
In order to further understand the growth of Cu-Sn IMC, longer reflow was done for 60 min at 300 °C. Figure 4 shows the FESEM cross-sectional images of Cu/Sn, Cu/Ni-35/Sn, and Cu/Ni-70/Sn after long reflow. Table 2 shows the EDX composition of the IMCs at different spots designated by A-E in Figure 4. In Figure 4a, it is seen that in the Cu/Sn system, all Cu and Sn layers have been transformed into a uniform layer of Cu3Sn. But, with the insertion of 35 nm of Ni, a lighter greycoloured ribbon-like layer of (Cu,Ni)6Sn5 IMC was found to be mixed with (Cu,Ni)3Sn IMC which has a darker grey contrast. An increase of Ni thickness to 70 nm results in an increase in the amount of (Cu,Ni)6Sn5 (lighter contrast). This sample ( Figure 4b) shows a brick-and-mortar type structure with bricks of (Cu,Ni)6Sn5 surrounded by (Cu,Ni)3Sn. Figure 4d-f shows EDX elemental maps of the Cu/Ni-70/Sn sample. Ni, which is indicated by the colour red is observed to segregate more to the light grey colour "brick" region ( Figure 4c,f). EDX results in Table 1 shows that the light grey region  ). Discontinuous Cu3Sn IMC formed in the middle of the Cu 6 Sn 5 phase due to reaction of the middle layer of copper with its surrounding tin during reflow. As for Figure 3f-h, layers with darker contrast represents Cu while lighter contrast represents the (Cu,Ni) 6 Sn 5 phase. With the addition of Ni, it is seen that (Cu,Ni) 6 Sn 5 IMC was formed regardless of the amount of Ni added. Cu/Ni-35/Sn has been converted completely into (Cu,Ni) 6 Sn 5 ( Figure 3f). However, for Cu/Ni-70/Sn and Cu/Ni-150/Sn samples (Figure 3g,h), unreacted copper (darker contrast) in between (Cu,Ni) 6 Sn 5 (lighter contrast), is still left after reflow. The thickness of the unreacted copper increased with increasing Ni thickness. This suggests that Ni starts acting as a barrier layer between Cu and Sn when its thickness increased to 70 nm.

Long Reflow
In order to further understand the growth of Cu-Sn IMC, longer reflow was done for 60 min at 300˝C. Figure 4 shows the FESEM cross-sectional images of Cu/Sn, Cu/Ni-35/Sn, and Cu/Ni-70/Sn after long reflow. Table 2 shows the EDX composition of the IMCs at different spots designated by A-E in Figure 4. In Figure 4a, it is seen that in the Cu/Sn system, all Cu and Sn layers have been transformed into a uniform layer of Cu 3 Sn. But, with the insertion of 35 nm of Ni, a lighter grey-coloured ribbon-like layer of (Cu,Ni) 6 Sn 5 IMC was found to be mixed with (Cu,Ni) 3 Sn IMC which has a darker grey contrast. An increase of Ni thickness to 70 nm results in an increase in the amount of (Cu,Ni) 6 Sn 5 (lighter contrast). This sample (Figure 4b) shows a brick-and-mortar type structure with bricks of (Cu,Ni) 6 Sn 5 surrounded by (Cu,Ni) 3 Sn. Figure 4d-f shows EDX elemental maps of the Cu/Ni-70/Sn sample. Ni, which is indicated by the colour red is observed to segregate more to the light grey colour "brick" region ( Figure 4c,f). EDX results in Table 1 shows that the light grey region is (Cu,Ni) 6 Sn 5 . Thus, Ni is seen to have a tendency to segregate to the (Cu,Ni) 6 Sn 5 phase. Higher solubility limit of Ni in Cu 6 Sn 5 compared with Cu 3 Sn is suggested to be the reason for the segregation [16]. is (Cu,Ni)6Sn5. Thus, Ni is seen to have a tendency to segregate to the (Cu,Ni)6Sn5 phase. Higher solubility limit of Ni in Cu6Sn5 compared with Cu3Sn is suggested to be the reason for the segregation [16].       Figure 5b shows the ratio of the peak intensities for Cu3Sn and Cu6Sn5 as a function of Ni layer thickness. The Cu6Sn5 and Cu3Sn peaks used for the calculation are 30.11° and 37.89°, respectively, which are the most prominent peaks for each phase in the XRD pattern. From Figure 5b, it is seen that the intensity ratio decreases which indicates that the amount of Cu3Sn formed decreases when the Ni thickness increased. With increasing Ni in the Cu/Sn system, more Ni atoms substituted the Cu atoms to form more (Cu,Ni)6Sn5 which are more stable than Cu6Sn5 IMC resulting in less transformation from (Cu,Ni)6Sn5 IMC to (Cu,Ni)3Sn IMC [17,18]. Figure 6 shows zoomed-in section of the XRD patterns of samples shown in Figure 5a recorded 29.5°-31° and 37°-38.5° recorded at a slow scan speed. This range of diffraction angles were selected to investigate possible peak shifts of the most prominent peaks of Cu6Sn5 and Cu3Sn IMC. It is seen in Figure 6a that the peak has shifted to the right when Ni atoms substitute Cu atoms in the (Cu,Ni)6Sn5 IMC. This indicates that the (311) interplanar spacing decreases upon Ni addition. This is in good agreement to previous study where it was found that Ni addition to Cu6Sn5 also decreased the lattice parameter [19]. As Ni atom substitutes Cu sites, the unit cell volume shrinks and the distance between the atoms shortens.
Cu3Sn possesses two features, orthorhombic unit cell, and a long period superstructure. The most prominent peak for the long period superstructure of Cu3Sn is at 37.56° for the (2 10 0) plane and that for the orthorhombic structure at 37.84° for the (021) plane are indicated by vertical lines. In the recorded patterns (Figure 6b), Cu3Sn in the Cu/Sn sample shows a peak which is close to the (2 10 0) long period superlattice peak. For the Cu3Sn IMC in the Cu/Ni/Sn sample, the peak tends to split into multiple sub-peaks and shift to higher angles. The peak for Cu/Ni-35/Sn sample is closer to that for  Figure 5b shows the ratio of the peak intensities for Cu 3 Sn and Cu 6 Sn 5 as a function of Ni layer thickness. The Cu 6 Sn 5 and Cu 3 Sn peaks used for the calculation are 30.11˝and 37.89˝, respectively, which are the most prominent peaks for each phase in the XRD pattern. From Figure 5b, it is seen that the intensity ratio decreases which indicates that the amount of Cu 3 Sn formed decreases when the Ni thickness increased. With increasing Ni in the Cu/Sn system, more Ni atoms substituted the Cu atoms to form more (Cu,Ni) 6 Sn 5 which are more stable than Cu 6 Sn 5 IMC resulting in less transformation from (Cu,Ni) 6 Sn 5 IMC to (Cu,Ni) 3 Sn IMC [17,18]. Figure 6 shows zoomed-in section of the XRD patterns of samples shown in Figure 5a recorded 29.5˝-31˝and 37˝-38.5˝recorded at a slow scan speed. This range of diffraction angles were selected to investigate possible peak shifts of the most prominent peaks of Cu 6 Sn 5 and Cu 3 Sn IMC. It is seen in Figure 6a that the peak has shifted to the right when Ni atoms substitute Cu atoms in the (Cu,Ni) 6 Sn 5 IMC. This indicates that the (311) interplanar spacing decreases upon Ni addition. This is in good agreement to previous study where it was found that Ni addition to Cu 6 Sn 5 also decreased the lattice parameter [19]. As Ni atom substitutes Cu sites, the unit cell volume shrinks and the distance between the atoms shortens. Cu 3 Sn possesses two features, orthorhombic unit cell, and a long period superstructure. The most prominent peak for the long period superstructure of Cu 3 Sn is at 37.56˝for the (2 10 0) plane and that for the orthorhombic structure at 37.84˝for the (021) plane are indicated by vertical lines. In the recorded patterns (Figure 6b), Cu 3 Sn in the Cu/Sn sample shows a peak which is close to the (2 10 0) long period superlattice peak. For the Cu 3 Sn IMC in the Cu/Ni/Sn sample, the peak tends to split into multiple sub-peaks and shift to higher angles. The peak for Cu/Ni-35/Sn sample is closer to that for Cu/Sn sample and exhibits two sub-peaks. The peaks for Cu/Ni-70/Sn shift closer to (021) Cu 3 Sn orthorhombic structures. These changes suggest that Ni atoms enter into the Cu 3 Sn lattice and exert an influence on the structure.

IMC Formation during Solid State Reactions
During room temperature aging, copper and tin react to form Cu6Sn5 IMC even after one day of aging. In this present work, it is found that Cu6Sn5 IMC forms simultaneously at two places; (i) at the interface between Cu/Sn, possibly through rapid interstitial diffusion of Cu into bulk Sn and (ii) along the Sn grain boundaries.
Cu6Sn5 formation at the Cu/Sn interface was not observed in the earlier study done by Sobiech and co-workers [14]. One of the reasons for this might be related to the use of a lower resolution microscopy in that study. The formation of this thin layer of Cu6Sn5 at the entire Cu/Sn interface found in this study might be related to the fine copper microstructure. In this study, the copper layer was electrodeposited from alkaline pyrophosphate bath which is known to yield fine grains [20]. This is confirmed in the present study as the copper layers have grains in the range of 100-200 nm. These tiny grains also contain nano-twins as seen in Figure 2b. These finer microstructures can result in favourable nucleation sites for Cu6Sn5. In previous studies [12,13], deposition of Cu and Sn was done

IMC Formation during Solid State Reactions
During room temperature aging, copper and tin react to form Cu 6 Sn 5 IMC even after one day of aging. In this present work, it is found that Cu 6 Sn 5 IMC forms simultaneously at two places; (i) at the interface between Cu/Sn, possibly through rapid interstitial diffusion of Cu into bulk Sn and (ii) along the Sn grain boundaries. Cu 6 Sn 5 formation at the Cu/Sn interface was not observed in the earlier study done by Sobiech and co-workers [14]. One of the reasons for this might be related to the use of a lower resolution microscopy in that study. The formation of this thin layer of Cu 6 Sn 5 at the entire Cu/Sn interface found in this study might be related to the fine copper microstructure. In this study, the copper layer was electrodeposited from alkaline pyrophosphate bath which is known to yield fine grains [20]. This is confirmed in the present study as the copper layers have grains in the range of 100-200 nm. These tiny grains also contain nano-twins as seen in Figure 2b. These finer microstructures can result in favourable nucleation sites for Cu 6 Sn 5 . In previous studies [12,13], deposition of Cu and Sn was done by thermal evaporation or electron beam evaporation which might yield a different structure and, thus, might provide fewer nucleation points.
Apart from the formation of Cu 6 Sn 5 at the Cu/Sn interface, Cu also diffuses along Sn grain boundaries resulting in a slender Cu 6 Sn 5 morphology at the grain boundaries (Figure 1a). Similar observations involving Cu diffusing along Sn grain boundaries were reported by Sobiech and co-workers [14,21]. After a complete coverage of Cu 6 Sn 5 at Sn grain boundaries and the Cu/Sn interfaces, it grows by a volume diffusion of Cu into Sn in a direction perpendicular to Sn grain boundaries. This is how the Cu 6 Sn 5 IMC grows wider and thicker after 24 days of room temperature aging.
A schematic of the stages in the formation of Cu 6 Sn 5 IMC in the Cu/Sn system during room temperature aging is shown in Figure 7. The stages involved are (i) formation of a thin layer of Cu 6 Sn 5 in between Cu and Sn and along Sn grain boundaries; (ii) growth of IMC along the Sn grain boundaries and (iii) volume diffusion of Cu into Sn. The unique "hourglass" shape of the Cu 6 Sn 5 IMC (Figure 2b) formed at the grain boundaries is due to the diffusion of Cu from two direction as shown in Figure 7e. The corners situated at the junctions between substrate and grain boundaries receive Cu flux from two directions: (i) from Cu substrate side and (ii) from grain boundary side (Figure 7e). As a result, Cu 6 Sn 5 grows faster here leading to the hourglass shape. by thermal evaporation or electron beam evaporation which might yield a different structure and, thus, might provide fewer nucleation points. Apart from the formation of Cu6Sn5 at the Cu/Sn interface, Cu also diffuses along Sn grain boundaries resulting in a slender Cu6Sn5 morphology at the grain boundaries (Figure 1a). Similar observations involving Cu diffusing along Sn grain boundaries were reported by Sobiech and coworkers [14,21]. After a complete coverage of Cu6Sn5 at Sn grain boundaries and the Cu/Sn interfaces, it grows by a volume diffusion of Cu into Sn in a direction perpendicular to Sn grain boundaries. This is how the Cu6Sn5 IMC grows wider and thicker after 24 days of room temperature aging.
A schematic of the stages in the formation of Cu6Sn5 IMC in the Cu/Sn system during room temperature aging is shown in Figure 7. The stages involved are (i) formation of a thin layer of Cu6Sn5 in between Cu and Sn and along Sn grain boundaries; (ii) growth of IMC along the Sn grain boundaries and (iii) volume diffusion of Cu into Sn. The unique "hourglass" shape of the Cu6Sn5 IMC (Figure 2b) formed at the grain boundaries is due to the diffusion of Cu from two direction as shown in Figure 7e. The corners situated at the junctions between substrate and grain boundaries receive Cu flux from two directions: (i) from Cu substrate side and (ii) from grain boundary side (Figure 7e). As a result, Cu6Sn5 grows faster here leading to the hourglass shape. In the case when Ni is introduced between Cu and Sn (Figures 1 and 2), the extent of IMC formation is less. It is seen in Figure 1 that Cu diffusion through the grain boundaries is restricted. Bader et al. has reported on a slower dissolution rate of Ni into Sn solders in comparison to Cu, and thus this can results in less IMC formation in the Ni-containing system [22]. It has been reported that the Cu diffusivity in Sn is 2.5 × 10 −7 cm 2 /s [23] while Ni diffusivity in Sn is 5.4 × 10 −9 cm 2 /s at a temperature above 160 °C [24]. Thus, it is expected that due to the slower diffusivity of Ni, Ni tends to block the diffusion path of Cu at the Sn grain boundaries which leads to less IMC formation.
Under room temperature aging conditions, it is observed that nanometric voids formed even after one day aging. With the increase of aging time to 24 days, the voids at the interface between Cu and Cu6Sn5 become prominent (Figure 2b). It has been reported that the voids can form in the Cu/Sn system due to Kirkendall effects. Kirkendall voids are reported to form at the Cu/Cu3Sn interface during solid state reactions because of the large difference in diffusion fluxes of Cu and Sn [12,25]. In this study, voids formed at the Cu/Cu6Sn5 interface due to the faster Cu diffusion leaving Cu vacancies at the Cu/Cu6Sn5 interface. However, when Ni is introduced into the Cu/Sn system, it is observed that the number of voids is less and the size of the voids is smaller. One reason for this might be the higher Ni diffusion flux in the system which can lead to a lesser difference in diffusion flux of Cu and Ni versus Sn [26]. In the case when Ni is introduced between Cu and Sn (Figures 1 and 2), the extent of IMC formation is less. It is seen in Figure 1 that Cu diffusion through the grain boundaries is restricted. Bader et al. has reported on a slower dissolution rate of Ni into Sn solders in comparison to Cu, and thus this can results in less IMC formation in the Ni-containing system [22]. It has been reported that the Cu diffusivity in Sn is 2.5ˆ10´7 cm 2 /s [23] while Ni diffusivity in Sn is 5.4ˆ10´9 cm 2 /s at a temperature above 160˝C [24]. Thus, it is expected that due to the slower diffusivity of Ni, Ni tends to block the diffusion path of Cu at the Sn grain boundaries which leads to less IMC formation.
Under room temperature aging conditions, it is observed that nanometric voids formed even after one day aging. With the increase of aging time to 24 days, the voids at the interface between Cu and Cu 6 Sn 5 become prominent (Figure 2b). It has been reported that the voids can form in the Cu/Sn system due to Kirkendall effects. Kirkendall voids are reported to form at the Cu/Cu 3 Sn interface during solid state reactions because of the large difference in diffusion fluxes of Cu and Sn [12,25]. In this study, voids formed at the Cu/Cu 6 Sn 5 interface due to the faster Cu diffusion leaving Cu vacancies at the Cu/Cu 6 Sn 5 interface. However, when Ni is introduced into the Cu/Sn system, it is observed that the number of voids is less and the size of the voids is smaller. One reason for this might be the higher Ni diffusion flux in the system which can lead to a lesser difference in diffusion flux of Cu and Ni versus Sn [26].

Dissolution of Ni during Reflow
From Figure 3, it is shown that only (Cu,Ni) 6 Sn 5 IMC was formed after reflow regardless of the thickness of the Ni layer inserted between Cu and Sn layers. This is expected as the amount of Ni inserted into the Cu/Sn system is too low, 2 at %-8 at %, which is calculated from the deposited thickness of the layers. With this amount of Ni in the system, Ni 3 Sn 4 , and Ni 3 Sn 2 IMC do not form [27]. The amount of Ni introduced is only sufficient for Ni to substitute Cu in the (Cu,Ni) 6 Sn 5 IMC. However, when the thickness of Ni increases, the amount of unreacted copper increased (Figure 3f-h). It has been reported earlier [3,28] that some metallic nanoparticles including Ni undergo reactive dissolution during reflow. Thus, at smaller thickness, e.g., 35 nm, most of the Ni is expected to dissolve and react to form IMC. At higher thickness e.g., 70 nm, only a part of Ni layer is thought to have been dissolved. The remaining portion of the Ni layer acts as a barrier film. It has been reported that the diffusivity of Cu into Ni is very slow, 2.2ˆ10´1 8 cm 2 /s at 300˝C [29]. Thus, when the thickness of Ni increased, the diffusion of Cu passing through Ni layers to reach the Sn is slower. This results in more unreacted copper when the thickness of Ni in the system increased.

Effect of Ni Addition in Cu 6 Sn 5 IMC
From the elemental maps in Figure 4, it is observed that Ni preferentially segregates to the (Cu,Ni) 6 Sn 5 phase. EDX results for the (Cu,Ni) 6 Sn 5 formed in the Cu/Ni-70/Sn sample shows that the Ni content in the IMC is 17.85 at %. This is in agreement with results published previously where it was reported that the solubility of Ni in (Cu,Ni) 6 Sn 5 was around 29 at % [4]. The brick-wall morphology structure seen in Figure 4 is suggested to result from the growth of the Cu 3 Sn phase happening along the grain boundaries of the (Cu,Ni) 6 Sn 5 IMC. Schematic in Figure 8 suggests steps in the growth of (Cu,Ni) 3 Sn during long reflow. It will be worthwhile to investigate the effect of the brick-wall morphology of the composite, (Cu,Ni) 6 Sn 5 + (Cu,Ni) 3 Sn on mechanical properties and crack propagation mechanisms.  Figure 3, it is shown that only (Cu,Ni)6Sn5 IMC was formed after reflow regardless of the thickness of the Ni layer inserted between Cu and Sn layers. This is expected as the amount of Ni inserted into the Cu/Sn system is too low, 2 at %-8 at %, which is calculated from the deposited thickness of the layers. With this amount of Ni in the system, Ni3Sn4, and Ni3Sn2 IMC do not form [27]. The amount of Ni introduced is only sufficient for Ni to substitute Cu in the (Cu,Ni)6Sn5 IMC. However, when the thickness of Ni increases, the amount of unreacted copper increased (Figure 3f-h). It has been reported earlier [3,28] that some metallic nanoparticles including Ni undergo reactive dissolution during reflow. Thus, at smaller thickness, e.g., 35 nm, most of the Ni is expected to dissolve and react to form IMC. At higher thickness e.g., 70 nm, only a part of Ni layer is thought to have been dissolved. The remaining portion of the Ni layer acts as a barrier film. It has been reported that the diffusivity of Cu into Ni is very slow, 2.2 × 10 −18 cm 2 /s at 300 °C [29]. Thus, when the thickness of Ni increased, the diffusion of Cu passing through Ni layers to reach the Sn is slower. This results in more unreacted copper when the thickness of Ni in the system increased.

Effect of Ni Addition in Cu6Sn5 IMC
From the elemental maps in Figure 4, it is observed that Ni preferentially segregates to the (Cu,Ni)6Sn5 phase. EDX results for the (Cu,Ni)6Sn5 formed in the Cu/Ni-70/Sn sample shows that the Ni content in the IMC is 17.85 at %. This is in agreement with results published previously where it was reported that the solubility of Ni in (Cu,Ni)6Sn5 was around 29 at % [4]. The brick-wall morphology structure seen in Figure 4 is suggested to result from the growth of the Cu3Sn phase happening along the grain boundaries of the (Cu,Ni)6Sn5 IMC. Schematic in Figure 8 suggests steps in the growth of (Cu,Ni)3Sn during long reflow. It will be worthwhile to investigate the effect of the brick-wall morphology of the composite, (Cu,Ni)6Sn5 + (Cu,Ni)3Sn on mechanical properties and crack propagation mechanisms. Peak intensity from XRD patterns ( Figure 6) shows an increase for Cu6Sn5 and a decrease for Cu3Sn IMC as the thickness of Ni in the sample increases from 35 nm to 70 nm. This shows that Ni addition to the (Cu,Ni)6Sn5 IMC stabilises the phase and, thus, there is less transformation to (Cu,Ni)3Sn IMC. This is in good agreement to studies done by Yu [18] and Nogita [19]. Nogita reported that 9 at % of Ni in the (Cu,Ni)6Sn5 IMC is sufficient to stabilise the IMC in the hexagonal structure. Yu et al. reported that when Ni substituted Cu atoms to form (Cu,Ni)6Sn5, volume shrinkage happens and the distance between Cu and Sn atoms is shortened. This resulted in much stronger bonds and, thus, a more stable phase. Peak intensity from XRD patterns ( Figure 6) shows an increase for Cu 6 Sn 5 and a decrease for Cu 3 Sn IMC as the thickness of Ni in the sample increases from 35 nm to 70 nm. This shows that Ni addition to the (Cu,Ni) 6 Sn 5 IMC stabilises the phase and, thus, there is less transformation to (Cu,Ni) 3 Sn IMC. This is in good agreement to studies done by Yu [18] and Nogita [19]. Nogita reported that 9 at % of Ni in the (Cu,Ni) 6 3 Sn in bulk sample prepared by melting and term aging [16,18,30,31]. Table 3 shows the effect of concentration and heating parameters on Ni concentration in (Cu,Ni) 3 Sn IMC. Oberndoff [31] has reported on the formation of (Cu,Ni) 3 Sn after samples containing 40 at % Sn, 35-50 at % Cu and 10-25 at % Ni after annealing at 235˝C for 1100-1700 h. Oberndoff also reported that the solubility of Ni in (Cu,Ni) 3 Sn can be as high as 3 at %. Lin and co-workers [16] also observed the formation of (Cu,Ni) 3 Sn when annealing was done at 240˝C for 1695 h on two groups of samples, with compositions (i) 25 at % Sn, 5-70 at % Cu, 5-70 at % Ni and (ii) 40 at % Sn, 30-50 at % Cu, 10-30 at % Ni. Lin et al. reported that when the concentration of Sn was low (25 at %), the amount of Ni in (Cu,Ni) 3 Sn varied from 5 at % to 70.8 at %. However, when the concentration of Sn was high (40 at %), the amount of Ni in (Cu,Ni) 3 Sn only ranged from 1.5 at % to 4.3 at %. In the present study, EDX results show that the amount of Ni in the (Cu,Ni) 3 Sn is 1.87 at % for the Cu/Ni-35/Sn sample and 4.56 at % for the Cu/Ni-70/Sn sample. This is in good agreement with previous studies as the concentration of Sn used in this study is around 40-50 at %. The formation of (Cu,Ni) 3 Sn IMC found in this study occurs because of the limited Sn supply in the samples used. Past studies used solder balls, solder pastes, or thick layers of Sn solders which can provide an effectively unlimited supply of Sn for the reactions of Cu-Sn IMCs. When the supply of Sn is limited, the amount of Cu 6 Sn 5 IMC formed is limited as well. Since Ni only has a solubility of around 27 at % in (Cu,Ni) 6 Sn 5 , Ni atoms will also substitute Cu in the (Cu,Ni) 3 Sn IMC as the reaction is thermodynamically possible [18]. From the XRD patterns (Y 6), it is seen that peaks for the Cu 3 Sn phase has shifted to higher angles when Ni substitutes Cu. Substitution of Cu by Zn was studied by van Sande et al. [32]. Addition of Zn to Cu 3 Sn was found to reduce the long period in the superstructure. The periodicity was also reduced when Ni was added [33]. Further studies are necessary to understand in detail the effect of Ni addition on the structure of Cu 3 Sn.

Materials and Methods
In this study, the multilayer interconnect systems were prepared on commercially pure copper substrates by electrodeposition. Sequential deposition was carried out to prepare both Cu/Sn and Cu/Ni/Sn multilayer interconnects. Cu/Sn multilayers were prepared in the sequence of Cu/Sn/Cu/ Sn/Cu while Cu/Ni/Sn multilayers were prepared in the sequence of Cu/Ni/Sn/Ni/Cu/Ni/Sn/ Ni/Cu. For the sake of simplicity, Cu/Sn/Cu/Sn/Cu and Cu/Ni/Sn/Ni/Cu/Ni/Sn/Ni/Cu will be designated by Cu/Sn and Cu/Ni/Sn, respectively. Copper deposition in this study, was performed using the alkaline pyrophosphate copper bath (CuSO 4¨5 H 2 O, 30 g/L; Na 4 PO 2 O 7 , 120 g/L; NH 3 , 1 mL/L). The bath pH was adjusted to 8.5 by using sulphuric acid for the pyrophosphate copper bath. Watts bath (NiSO 4 , 150 g/L; NiCl 2 , 60 g/L; H 3 BO 3 , 37.5 g/L) was used for the deposition of nickel. Methanesulfonic acid (MSA)-based bath (SnSO 4 , 30 g/L; gelatin, 1 g/L; hydroquinone, 5 g/L; MSA, 120 mL/L) was used for the deposition of tin [15,34]. Stirring condition was set at 80 rpm for all baths. Deposition current density was 10 mA/cm 2 for copper deposition and 20 mA/cm 2 for tin and nickel deposition. The deposition time required to achieve required thickness was estimated using Faraday's law. Copper was plated to achieve a thickness of 1000 nm while tin was plated to a thickness of 2000 nm. Nickel was plated to a thickness of 35 nm, 70 nm, and 150 nm to study the effect of varying Ni concentration in the Cu/Sn system.
The multi-layered samples were subjected to different processing conditions to study their intermixing and phase transformation characteristics: (i) room temperature aging; (ii) short reflow and (iii) long reflow. Room temperature aging studies were done at 25˝C. Short reflow was done in a FT-02 convection reflow oven at 250˝C while long reflow was done in a tube furnace in 2% H 2 /N 2 atmosphere at 300˝C. FEI Quanta 450 (Houston, TX, USA) Field-Emission Scanning Electron Microscope (FESEM) was used to examine the cross-section of the electrodeposits. The number and size of the voids formed in the Cu/Sn and Cu/Ni/Sn samples were calculated from four micrographs for each condition. Energy dispersive X-ray spectroscopy (EDX, Oxford Instruments, Oxfordshire, UK) was used to determine the composition of the multilayer system. FIB milling was done in a FEI Helios Nanolab 650 Dual Beam FESEM. The phases of the multilayer sample after long reflow were investigated by X-ray Diffraction (PANalytical, Almelo, The Netherlands) using a PanAnalytical diffractometer with Cu Kα radiation which has a wavelength of 1.5418 Å. The step size used was 0.26˝and the scan step time was 2.11 s. For slower scans, a step size of 0.01˝and scan step time of 8.80 s were used. Peaks shown in the XRD pattern were identified by using the Powder Diffraction File card (JCPDS).

Conclusions
1. In the electrodeposited Cu/Sn system, Cu 6 Sn 5 IMC is seen to grow even after one day of room temperature aging, while less Cu-Sn IMC is found in Cu/Ni/Sn system under the same conditions. This finding is important to prevent premature intermixing between the multilayers prior to reflowing during manufacturing process. 2. During solid state reaction in the Cu/Sn system, the growth of the Cu 6 Sn 5 first started through rapid reaction at the Cu/Sn interface followed by grain boundary diffusion of Cu atoms into the Sn grain boundaries. With the addition of Ni in between Cu and Sn, Ni atoms are suggested to block the diffusion path of Cu atoms into Sn grain boundaries and slow down IMC formation. 3. During liquid solid reaction, at 70 nm thickness Ni dissolution is incomplete in the Cu/Ni/Sn multilayers. Thus, to achieve homogeneous IMC layer during short reflow, Ni thickness less than 70 nm is recommended. The addition of Ni suppresses formation of Cu 3 Sn IMC, regardless of the thickness of the Ni layers. 4. After 60 min of long reflow, Cu/Sn multilayers have been transformed totally into Cu 3 Sn. In the Cu/Ni/Sn system, Ni atoms take part in the formation of (Cu,Ni) 6 Sn 5 and, thus, stabilizes the IMC and retards transformation into (Cu,Ni) 3 Sn. Formation of (Cu,Ni) 3 Sn is suspected to be due to the limited Sn supply in the system. 5. The mechanical properties of the unique "brick-wall" morphology of the [(Cu,Ni) 6 Sn 5 + (Cu,Ni) 3 Sn] composite formed in the Cu/Ni-70/Sn samples after long reflow would be interesting, as it may influence fracture propagation.