High electric current density-induced interfacial reactions in micro ball grid array (lbga) solder joints

Transcription

1 Acta Materialia 54 (2006) High electric current density-induced interfacial reactions in micro ball grid array (lbga) solder joints M.O. Alam a, B.Y. Wu a, Y.C. Chan a, *, K.N. Tu b a Department of Electronic Engineering, City University of Hong Kong, Tat Chee Avenue, Kowloon Tong, Hong Kong b Department of Materials Science and Engineering, UCLA, Los Angeles, CA , USA Received 30 June 2004; received in revised form 22 September 2005; accepted 23 September 2005 Available online 22 November 2005 Abstract The effect of a high electric current density on the interfacial reactions of micro ball grid array solder joints was studied at room temperature and at 150 C. Four types of phenomena were reported. Along with electromigration-induced interfacial intermetallic compound (IMC) formation, dissolution at the Cu under bump metallization (UBM)/bond pad was also noticed. With a detailed investigation, it was found that the narrow and thin metallization at the component side produced Joule heating due to its higher resistance, which in turn was responsible for the rapid dissolution of the Cu UBM/bond pad near to the Cu trace. During an electromigration test of a solder joint, the heat generation due to Joule heating and the heat dissipation from the package should be considered carefully. When the heat dissipation fails to compete with the Joule heating, the solder joint melts and molten solder accelerates the interfacial reactions in the solder joint. The presence of a liquid phase was demonstrated from microstructural evidence of solder joints after different current stressing (ranging from 0.3 to 2 A) as well as an in situ observation. Electromigration-induced liquid state diffusion of Cu was found to be responsible for the higher growth rate of the IMC on the anode side. Ó 2005 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved. Keywords: Joule heating; Heat dissipation; Electromigration; Solder joint; BGA 1. Introduction * Corresponding author. Fax: address: eeycchan@cityu.edu.hk (Y.C. Chan). With the trend towards miniaturization and very-largescale integration of circuits on Si devices, electronic packaging requires higher I/O density, smaller feature size, and better performance [1 3]. A common cause of failure at narrow electric conductors by high electric current densities is electromigration [3 10]. Electromigration is the enhanced diffusion of atoms in the current direction. Because of the higher current density in a fine pitch solder joint, electromigration is a growing concern in electronic packaging that needs to be addressed. Without an electric current, the driving force of intermetallic compound (IMC) formation is the chemical potential difference between the two contact materials. At a high current density above 10 2 A/cm 2, electron flow can play a significant role in IMC formation. Chen and co-workers [11 18] have studied IMC formation in several diffusion couples such as Sn/Ni, Sn/Ag, Sn/Cu, and other solders, etc., with a DC current density of A/cm 2. They observed a directional effect of electric current on the IMC thickness at the interfaces at a current density of A/cm 2. Their samples were sandwich-type bars with a cross-section of 1mm 2. Although the phenomena might not be directly comparable with a real package, they reported that the interactions at the cathode and anode are different due to a polarity effect. While electromigration enhances IMC formation at the anode, it enhances IMC dissolution at the cathode. During current stressing, heat is also generated through Joule heating which may maintain a thermal gradient in the solder joint or may even melt the solder (solder alloy is a low melting point component among the interconnects /$30.00 Ó 2005 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved. doi: /j.actamat

2 614 M.O. Alam et al. / Acta Materialia 54 (2006) used in an electronic package). However, melting of the solder joints during high current stressing has not been reported so far. Moreover, for an area array solder joint such as for a ball grid array (BGA) package, the way that electric current flows through a solder bump is not uniform. The origin of such a non-uniform current distribution is a high to low current density transition at the contact window between the conducting metal line and the solder ball. Thus, current crowding occurs at the contact interface between the solder ball and the metal pad/ under bump metallization (UBM). It was found for a flip chip solder joint that the high current density due to current crowding is about one order of magnitude higher than the average current density in the joint [4]. This current crowding exerts a much greater driving force for electromigration and also generates local Joule heating and/or enhances local dissolution of UBMs [3 5]. For a lbga solder joint, current crowding could be much more intensive. This is because a very narrow Cu trace is connected to a relatively larger solder ball. Fig. 1 shows a typical solder ball arrangement on a lbga package. When it is bonded on the printed circuit board (PCB), electric current flows through the Cu trace, of 60 lm 15 lm cross-section, to the solder ball, of 300 lm diameter. A very large current crowding occurs at the interface of the Cu bond pad and the solder ball, where the narrow Cu trace meets. With a larger amount of current (e.g., a 1 A flow through the Cu trace when the current density in the Cu is A/cm 2 ), the temperature rises by Joule heating. As this Cu trace is usually enclosed in a poor thermally conducting dielectric material (such as a solder mask), the temperature may rise to the melting point of the solder alloy. As a result, the solder alloy near the Cu traces may reach the liquid state and thus increase the dissolution of Cu and eventually lead to electrical failure. This work is intended to study interfacial phenomena of the solder joint of a real package under high current stressing. A dummy Fig. 1. Optical micrograph of the general view of a lbga package from the bottom side showing the solder balls attached with the Cu traces on the lbga substrate. lbga package with the same pair of solder joints was used to obtain reproducible results for a practical application. Detailed microstructural characterization was carried out to understand the changes under a range of current stressing. 2. Experimental The samples used in this study were the dummy version of TesseraÕs lbga package (CSP46T75). The area of the package was 5.76 mm 7.87 mm and the height including solder ball was 0.84 mm. The diameter of the solder balls was 0.3 mm. Solder balls were attached to the package with a 6 8 grid and a 0.75 mm pitch (see Fig. 1). The solder bump material was 63Sn 37Pb. The solder bumps were connected with a Cu trace 60 lm wide and 15 lm thick. The Au thickness on the Cu trace was 1 lm. The length of the Cu trace depended on the location of the bumps. Fig. 2(a) shows a schematic of the electrical connection between the solder bumps for the lbga package (component side). The organic FR- 4 substrate was chosen for the test board, since it is one of the most commonly used PCBs. The board size was 110 mm 120 mm with a thickness of 1 mm. The width and thickness of the Cu trace were 100 and 15 lm, respectively. An electroless Ni P layer 5 lm thick was deposited on the Cu trace followed by 1 lm ofau flash. A schematic of the PCB is shown in Fig. 2(b). The outside points are bump test pads with a diameter of 1 mm. The inside 6 8 array of pads with a diameter of 0.3 mm are to receive the solder bumps of the lbga packages. The lbga component was attached to the substrate using flip chip bonding technology by a Karl Suss 9493 Mauren Flip Chip bonder. The bonded samples were then reflowed using a five-zone air convection oven (BTU VIP 70 N) in a compressed N 2 environment. In the temperature profile, the peak temperature was 225 C and the time in the molten state was about 60 s. A completed daisy chain after bonding is shown in Fig. 2(c). Two contiguous joints (nos. 10 and 11) were subjected to different currents ranging from 0.6 to 2 A DC at room temperature and from 0.3 to 1.2 A DC at 150 C. The same pair of solder joints was used for the whole range of current stressing to make the results reproducible. For joint no. 10, the current flowed from the board side to the component side, whereas for joint no. 11, the current flowed from the component side to the board side. The daisy chain allowed one to find which joint failed. The real temperature of the solder interface has not yet been measured. However, the temperature was monitored by attaching a thermocouple to the backside of the component to correlate it with a variation in current stressing. It needs to be mentioned here that unlike other investigations, where a typical test module for an electromigration study was used [4,5,7], solder joints of a real package have been used for this study. Thus, the electrodes and

3 M.O. Alam et al. / Acta Materialia 54 (2006) Fig. 2. The layout of the daisy chain of the lbga used for this study. (a) Component side, (b) PCB side, and (c) completed daisy chain after bonding. Pad 10 on the PCB was connected at the negative terminal of the power source and pad 11 with the positive terminal. the solder joints were not in the same virtual line to examine all the phenomena for a particular condition from the same cross-sectional sample preparation. The solder ball/joints (i.e., nos. 10 and 11) under investigation are marked in Figs. 1 and 2. The scanning electron microscope (SEM) used for this study was a Philips XL 40 FEG equipped for energy dispersive X-ray analysis (EDX). 3. Results Fig. 3 shows a cross-sectional microstructure of a typical lbga solder joint before current stressing. On the component side, Cu Sn IMCs were formed in between the Cu bond pad and the eutectic SnPb solder. Comparing the EDX analysis and the stoichiometry of the typical Cu Sn IMCs, Cu 6 Sn 5 IMC was confirmed. In the backscattered electron (BSE) image of the SEM, a thin relatively darker layer is also visible. Due to the limitation of the resolution of the EDX, the composition could not be determined; however, from a study of the literature, this darker layer is likely to be Cu 3 Sn IMC [19 21]. It is worth mentioning here that this component-to-solder interface experienced two exposures of reflow: one was after the solder paste printing on the component side and the other one was during the bonding of the component to the board. On the Fig. 3. BSE-SEM image of the cross-section of the lbga solder joint just after reflow. board side, a ternary Cu Ni Sn IMC formed in between the solder and bond pad. A thin black layer was also noticeable. From the EDX analysis, 25 at.% P and 75 at.% Ni were found in this black layer. In the previous study of interfacial reactions on electroless Ni layers, it was found that molten solder alloy reacted with the Ni during reflow, and the unreacted P accumulated on the original Ni P layer, forming a crystalline P-rich layer of Ni P compounds which was seen as a deep dark layer under BSE image [22 24]. Table 1 gives mean-time-to-failures of the solder joints with current stressing. A steady-state temperature at the component side was noticed on the samples which survived for a longer time. After current stressing both at room and higher temperatures, four types of phenomena were observed with the variation of current stressing: case 1: instant failure with a burning smell; case 2: failure within h, case 3: survival of the joints even after 500 h; and case 4: failure within h at high temperature Case 1: instant failure with a burning smell Fig. 4 shows cross-sectional SEM images of a typical failed joint which failed within a few seconds of 2 A current stressing at room temperature with a burning smell. Fig. 4(a) shows the whole solder joint (no. 10). Fig. 4(b) shows the magnified view of the reaction interface for the component side (i.e., anode). Localized dissolution of the Cu bond pad was found at the anode side, where the Cu trace was connected to the Cu bond pad. Because of this localized Cu dissolution, the electrical circuit became open. It is believed that such a rapid (10 s) dissolution at the anode side was only possible by a liquid-state reaction. In the following section, a discussion is given as to how Joule heating of the Cu trace is responsible for liquefying the solder alloy. It is clear that the thickness of the interfacial finegrained Cu 6 Sn 5 IMC layer is less than 1 lm, which is less than that of the as-reflowed IMC layer. Bright AuSn 4 IMC is also noticeable which is supposed to be formed after the testing. It is well known that Au can diffuse into the solid Sn to form AuSn 4 IMC at room temperature [25]. At the board side, the IMC layer is similar to that of the as-reflowed sample.

4 616 M.O. Alam et al. / Acta Materialia 54 (2006) Table 1 Test results with the test conditions Test temperature ( C) Current, I (A) Current density in the solder ball, J (A/cm 2 ) Current density in the Cu trace, J (A/cm 2 ) Temperature at the backside of the component ( C) N/A 502 h N/A 21 h N/A 1.5 h N/A 0.5 h h h h min Unstable 10 s Times in italic type indicate the samples that did not fail. Average failure time Fig. 4. Cross-sectional SEM images of a typical failed joint which failed within a few seconds of 2 A current stressing at room temperature (case 1). (a) Whole solder joint (joint no. 10) and (b) magnified view of the dissolved area at the component side Case 2: failure within h Fig. 5(a) shows the cross-section of a joint which failed after 45 min of 1.5 A current stressing at room temperature. Electrical current flowed from the board side to the component side. The magnified view of the upper right corner of the joint is shown in Fig. 5(b). Along with the localized Cu dissolution, elliptical grains more than 15 lm in length are noticeable at the component (anode) side. No such IMC thickening was observed at the board (cathode) side. It was reported elsewhere that electromigration enhances the growth of IMC at the anode and inhibits the growth of IMC at the cathode as compared with the no-current case. Dissolution of the IMC and the metallization at the cathode side is also enhanced by electromigration [3,7]. However, no such dissolution at the anode side was reported previously. Fig. 6(a) shows the two contiguous solder joints which were electrically connected by a daisy chain. In the lefthand solder joint (no. 10), electrons flowed from the board side to the component side and in the right-hand solder joint (no. 11), the direction was the opposite. The circuit failed after 3 h of 1.2 A current stressing. A localized dissolution of the Cu pad at the component side is Fig. 5. Cross-sectional SEM images of a typical failed joint which failed after 45 min of 1.5 A current stressing at room temperature (case 2). (a) Whole solder joint (joint no. 10). (b) Upper right corner to depict the Cu dissolution.

5 M.O. Alam et al. / Acta Materialia 54 (2006) Fig. 6. (a) Two contiguous solder joints (left-side solder is joint no. 10 and right-side solder is joint no. 11) which were electrically daisy-chain connected and failed after 3 h of 1.2 A current stressing (case 2). (b) Magnified view of the substrate side of the right-side joint (anode). seen clearly for both solder joints. The dissolution at the cathode side in the right-hand solder joint is understandable but the dissolution at the anode side would only be possible if the temperature of that portion had been higher than the melting point of the solder alloy. One significant difference is that a large number of elongated Cu 6 Sn 5 IMC grains grew vertically on the board side (anode) for the right-hand solder joint, see Fig. 6(b). There was no Ni detected in the elongated grains, although the substrate metallization was electroless Ni P. Only 2 3 at.% Ni was found in the IMC layer near to the board metallization. From this finding, it is clear that a large amount of Cu diffused from the component side to the substrate side Case 3: survival of the joints even after 500 h With a lower current stressing, the lbga solder joints survived for a longer time. At currents below 0.9 A, no failure was noticed even after 500 h of room temperature testing. For a high-temperature test, no failure at 0.3 A current stressing was found. Cross-sectional studies of the solder joints reveal that the IMC growth rate was different and the composition of the IMC was also different depending on the polarity and metallization. Fig. 7 shows high-magnification BSE images of the interfaces after 512 h of 0.9 A current stressing at room temperature. The arrows marked by e on the image indicate the direction of current flow. Anode and cathode sides of the left-hand solder joint (no. 10) are shown in Fig. 7(a) and (b), respectively. Anode and cathode sides of the right-hand solder joint (no. 11) are shown in Fig. 7(c) and (d), respectively. For the component side where metallization is only Cu, both Cu 6 Sn 5 and Cu 3 Sn IMCs were detected for both solder joints. However, a difference in thickness is clear depending on the polarity: at the anode side, Cu 6 Sn 5 and Cu 3 Sn IMCs were thicker than that at the cathode side (see Fig. 7(b) and (d)). For the board side, where electroless Ni P was used, thicker (Cu,Ni) 6 Sn 5 IMC is noticeable for the anode side than that on the cathode side (see Fig. 7(a) and (c)). A difference in composition of the (Cu,Ni) 6 Sn 5 IMC at the solder board interface depending on the polarity is also interesting. From EDX analysis, around 9 at.% Ni at the cathode side was found, whereas for the anode side it was only around 2 at.%. Table 2 summarizes the difference between solder joints 10 and 11 resulting from the high electric current density in the solder ball ( A/ cm 2 ). However, it needs to be mentioned here that the exact temperatures of the solder joints have not yet been measured during the current stressing Case 4: failure within h at high temperature Fig. 8 shows the cross-section of the contiguous solder joints (10 and 11) where failure was noticed after 21 h of 0.6 A current stressing (current density in the solder ball was A/cm 2 ) at 150 C. Table 3 shows the thickness and the compositions of the interfacial IMCs for both the board side and component side of these solder joints. Although the current was lower than in the cases mentioned above, high temperature accelerated electrical currentinduced interfacial reactions. A localized dissolution of

6 618 M.O. Alam et al. / Acta Materialia 54 (2006) Fig. 7. High-magnification BSE images of the interfaces that were stressed by 0.9 A current for 512 h at room temperature (case 3), (a) board side and (b) component side of solder joint 10 where electrical current was flowing from the board side to the component side. (c) Substrate side and (d) component side of solder joint 11 where current was in the opposite direction. Table 2 Typical thickness and composition of the IMCs formed at the interface for case 3 IMC thickness and composition the Cu pad at the cathode side of the solder joint, where current entered in to the solder joint, was also noticed. Unlike Figs. 4 6, here the Cu dissolution was found only on the cathode side. 4. Discussion Left solder, joint no. 10 Right solder, joint no. 11 Substrate side/cathode Component side/anode Substrate side/anode Thickness (lm) at.% Sn at.% Ni at.% Cu at.% Au Component side/cathode It is known that the resistance (R) of a conductor is related to the resistivity (q) of the material, the length (l), and the cross-sectional area (q) of the conductor by the following equation: R ¼ ql q. ð1þ From this equation, it is seen that the smaller the cross-sectional area, the higher the resistance. In our daisy-chained lbga package, the cross-sectional area of the Cu traces connected to the solder balls on the component side had the smallest cross-sectional area and thus gave the highest resistance. The resistance of a Cu trace 650 lm long with a cross-section of 60 lm 15 lm is about 120 lx. By comparison, the resistance of a spherical lbga solder of 300 lm diameter is about 5.6 lx (for simplification a cube-shaped piece of 300 lm length was considered). The Joule heating produced from a conductor is proportional to I 2 R. Thus, heat generation in the Cu trace would be higher than in the solder itself, although the same current is flowing through the Cu trace and the solder joints. Moreover, the Cu trace is enclosed by insulating materials. The Cu trace on the board side, used for this study, is wider and thicker without any insulation, and can thus carry current at a lower resistance as well being free to radiate and conduct heat to the surroundings. It is expected for this setup that the temperature of the interface of the Cu to the solder on the board side would be lower than that on the component side. During the current stressing, when the Joule heating exceeds the heat dissipation, the temperature of the Cu trace on the component side may exceed the melting point of the solder alloys. A temperature of 200 C does not produce any change of phase in the Cu conductor; however, at this temperature the eutectic Sn Pb solder alloy melts. Table 1 shows the test conditions for this study, where the current density of the Cu trace on the component side as well as the current density of the solder joint have also been compared. The temperature at the solder interface has not been measured in this study; however, temperatures at the backside of the component were measured

7 M.O. Alam et al. / Acta Materialia 54 (2006) Fig. 8. Two contiguous solder joints (left-side solder is joint no. 10 and right-side solder is joint no. 11) which were electrically daisy-chain connected and failed after 21 h of 0.6 A current stressing at 150 C. Table 3 Typical thickness and composition of the IMCs formed at the interface for case 4 IMC thickness and composition Left solder, joint no. 10 Right solder, joint no. 11 Substrate side/cathode Component side/anode Substrate side/anode Thickness (lm) at.% Sn at.% Ni at.% Cu at.% Au Component side/cathode Fig. 9. Stereomicrographs to show cyclic melting and solidification phenomena of a solder joint under 1.2 A current stressing. to compare the variation of the interface temperature with the increase of applied current. A special setup was prepared to observe the melting phenomena at the solder interfaces. Fig. 9 shows an in situ melting phenomenon under high current stressing. The right-hand solder ball was in electrical connection to see/ compare with the left solder joint which was not under high current stressing. Fusion of that solder ball was observed under a stereomicroscope through the rapid change in the color and curvature of the solder ball (see Fig. 9(b)). Later, the solder ball was seen to resolidify. Fig. 9(c) shows the solidified solder which depicts metallic luster by solid-state reflection. Liquification of the solder ball was seen again just like the case shown in Fig. 9(b). This happened cyclically. Waves of phase transition were seen to move to and fro from one side to the other. The liquid phase was seen to nucleate where the Cu trace on the component side met with the solder joint (which can be termed a hot spot of the solder joint) and extended down from the component side to the board side. Solidification again started to proceed from the board side. Partial liquification was found only at the component side when the current was lower than 1.2 A. Solder bumps were very unstable in both the molten and solid state. From these observations, it is also clear that the component side was at a higher temperature than the board side. It was found that after a certain period, depending on the current stressing, the Cu trace near the hot spot of the solder joint (where the time in the liquid state of the solder alloy was longest) became dissolved away which led to an electrical failure. Microstructural evidence of the localized Cu dissolution shown in Figs. 4 6 also validates these in situ observations. For liquification, the solder ball absorbed the latent heat of fusion. The source of the latent heat for the fusion of the solder ball was the Cu trace on the component side, which rose to a higher temperature due to the Joule heating. During melting of the solder ball, the temperature of the Cu trace would decrease substantially. Simultaneously, heat dissipation from the solder ball was also responsible for reducing the temperature of the joint. In particular, the Cu circuitry on the substrate side helped to extract heat as a heat sink. Thus, because of the overall decrease of temperature, the solder ball started to solidify from the substrate side. Meanwhile, the temperature of the Cu trace on the component side rose again because of the cumulative Joule heating. The heat from the Cu trace could raise the temperature of the system to again melt the solder ball starting from the hot spot, i.e., where the Cu trace met with the solder joint. An interesting finding from this study is that the solder ball solidified and liquefied repeatedly when 1.2 and 1.5 A

8 620 M.O. Alam et al. / Acta Materialia 54 (2006) currents were applied (similar to case 2). An increase in the current gave more heat generation and thus the duration of the liquid-state solder in the solder joint also increased especially in the region of the solder that was near to the Cu trace on the component side (i.e., hot spot) would remain in a liquid state until the circuit failed. However, with a further increase of current flow, when the supply of heat was high enough to increase the temperature of the liquid solder above its melting point, the circuit failed with a very rapid dissolution of the Cu trace and the polymeric material near the Cu trace burnt. Careful observation of the cross-sectional microstructures of the failed samples also led to the conclusion that the appearance of the liquid state in the solder joint is in fact responsible for the catastrophic failure mentioned in case 1 and case 2. However, for case 1, the temperature of the molten solder increased until the circuit failed. For case 2, the liquid temperature remained near the eutectic temperature because of the solid-to-liquid cyclic transition. At 1.2 A current stressing, the total duration in the solid state condition was longer than that in the liquid state condition and thus nearly 3 h was needed to dissolve the Cu bond pad near the Cu trace. At 1.5 A current stressing, the duration in the liquid state was longer than that in the solid state and thus the Cu dissolved within 1 h. It has been reported elsewhere that during electromigration/high current stressing experiments, the dissolution of either the base metallization or IMC occurs only at the cathode side, where electrons are leaving [3,7,26]. In the present study, along with the dissolution of the Cu pad at the cathode side, dissolution at the anode side has also been found (see Figs. 4 6), although these experiments were conducted at room temperature. It is believed that this dissolution was due to the liquid-state dissolution of Cu connected to the solder joint, which was at the higher temperature during high current stressing. After 3 h of 1.2 A current stressing during the room temperature experiments, a large number of Cu 6 Sn 5 IMC grains formed at the anode (see Fig. 6(b)) on the board side. When the current was in the opposite direction (i.e., at the cathode side) only a 2 3 lm thick IMC layer was noticed on the similar metallization. Other investigators have reported that electromigration accelerates IMC formation at the anode side [9 18]. However, this IMC formation should be accomplished by atom diffusion. For the case considered here, if the solder is thought to remain in the solid state with 1.2 A current stressing for 3 h, a solid-state diffusion coefficient of Cu in solder of D 10 8 cm 2 /s is needed to calculate the time required to move Cu atoms from the component bond pad side to the board side. The stand-off height of the package after bonding is 0.2 mm, i.e., Cu atoms need to diffuse through a distance of 0.2 mm. Using the relationship of x Dt, the time required for solid-state diffusion of Cu in the solder layer of 0.2 mm is around 11 h. Thus, the time required to form a 20 lm thick IMC layer should be much larger than 11 h. However, a 20 lm thick IMC layer was found to form within 3 h. This would only be possible if solder was in a molten condition. Using a liquid-state diffusion coefficient of Cu in solder, D 10 5 cm 2 /s, 40 s is required to diffuse atoms from the solder-component side to the solder board interface. From the Cu Sn binary phase diagram, it is conceivable that below the eutectic temperature a large amount of Cu 6 Sn 5 IMC could stay in an equilibrium condition with the Sn matrix (as they form an eutectic couple). However, from a consideration of liquid-state reaction kinetics mentioned above, an alternative explanation for the rapid dissolution of copper may be offered. The high currentinduced Joule heating which is responsible for raising the temperature up to the melting point of the solder alloy at the component-to-solder region (more specifically, the portion of the solder joint where the Cu trace meets the Cu bond pad) causes rapid dissolution of the Cu UBM by the molten solder. From this explanation, it is also believed that rapid dissolution of Cu UBM reported by Hu et al. [26] was in fact caused by the appearance of a liquid phase in their solder joints. A large quantity of irregular-shaped IMC formation in the bulk solder within 90 min of current stressing also demonstrates Joule heating induced melting phenomena, especially when Cu trace is replaced by the solder composition (and thereby the resistance is increased that is again responsible for further increase in joule heating). During an electromigration test of a thin Al strip on a Si substrate, Joule heating is not a problem because the heat dissipates quickly. However, for the lbga solder joints, when they are connected with a metallization with a smaller cross-section in a package, heat dissipation would be less than the contribution from Joule heating, and thus there is a chance to rise the temperature. As solder is the low melting point component in the interconnection, it is very sensitive to temperature. Rapid dissolution of the Cu UBM in this study demonstrates the occurrence of the melting phenomena of a solder joint. Even when melting does not occur, the temperature might be raised to an unpredictable range depending on the ease of heat dissipation of a particular arrangement. It would be worthwhile to measure the real-time temperature of a solder interface to report and/or understand the electromigration phenomena in a solder joint. Until a real temperature distribution in the solder interface and in the bulk of the solder bump has been obtained, it is impossible to generalize the electromigration behavior in a solder joint. 5. Conclusions Systematic experimental work by varying the applied current load to find the influence of high current stressing on the damage mechanism of lbga solder revealed four cases: case 1: instant failure with a burning smell; case 2: failure within h; case 3: survival of the joints even after 500 h; case 4: failure within h at high temperature. A detailed microstructural investigation was carried

9 M.O. Alam et al. / Acta Materialia 54 (2006) out to understand the interfacial reaction phenomena of the component metallization to the solder and the solderto-substrate metallization. It has been found that the appearance of a liquid solder phase during high current stressing causes rapid dissolution of the Cu bond pad. The appearance of the liquid phase has been confirmed through a critical examination of the microstructural features in terms of reaction kinetics as well as by in situ observations. Joule heating due to the high current density in the conducting lines of the component has been found to fuse the solder ball. Once the solder is fused, it accelerates the reaction kinetics through faster diffusion. It has been found that a repeated solid-to-liquid-phase transformation occurs by the absorption of the latent heat of fusion of the solder ball plus heat dissipation and the time delay to resupply cumulative Joule heating. When the Cu trace connecting the bond pad dissolves away, the circuit fails. There is a competition between the heat dissipation and the heat generation from Joule heating. Instant failure has been found when the heat generation dominates over heat dissipation and thus the temperature of the system increases rapidly and leads to catastrophic failure (i.e., case 1). When the temperature remains at around the eutectic temperature of the solder, circuits fail in h (i.e., case 2). Electromigration-induced interfacial reactions were observed for case 3 and case 4. Acknowledgments This research was supported by NSFC/RGC Joint Research Scheme N_CityU/103/03 (CityU internal ref ) and RGC funded CERG project CityU 1106/ 04E (CityU internal ref ). One of the authors (K.N.T.) acknowledge the support by NSF contract #DMR and SRC contract #NJ-774. References [1] Puttlitz K, Totta PA. Area array interconnection handbook. Massachusetts: Kluwer Academic; [2] Tu KN, Zeng K. Mater Sci Eng Rep 2001;R34:1 58. [3] Tu KN. J Appl Phys 2003;94: [4] Yeh ECC, Choi WJ, Tu KN, Elenius P, Haluk B. Appl Phys Lett 2002;80: [5] Ye H, Basaran C, Hopkins D. Appl Phys Lett 2003;82: [6] Conrad H. Mater Sci Eng A 2000;287: [7] Choi WJ, Yeh ECC, Tu KN. J Appl Phys 2003;94: [8] Liu CY, Chen C, Liao CN, Tu KN. Appl Phys Lett 1999;75: [9] Garay JE, Anselmi-Tamburini U, Munir ZA. Acta Mater 2003;51: [10] Bertolino N, Garay J, Anselmi-Tamburini U. Scripta Mater 2001;44: [11] Chen CM, Chen SW. J Electron Mater 1999;27: [12] Chen SW, Chen CM, Liu W-C. J Electron Mater 1998;27: [13] Du MY, Chen CM, Chen SW. Mater Chem Phys 2003;82: [14] Chen SW, Chen CM. JOM-J Min Met Mater Soc 2003;55:62 7. [15] Chen CM, Chen SW. Acta Mater 2002;50: [16] Chen CM, Chen SW. J Appl Phys 2001;90: [17] Chen CM, Chen SW. J Electron Mater 2000;29: [18] Chen CM, Chen SW. J Mater Res 2003;18: [19] Lucas JP, Rhee H, Guo F, Subramanian KN. J Electron Mater 2003;32: [20] Gagliano RA, Fine ME. J Electron Mater 2003;32: [21] Nah JW, Paik KW, Suh JO, Tu KN. J Appl Phys 2003;94: [22] Alam MO, Chan YC, Tu KN. J Appl Phys 2003;94: [23] Alam MO, Chan YC, Hung KC. Microelectron Reliab 2002;42: [24] Alam MO, Chan YC, Hung KC. J Electron Mater 2002;31: [25] Nakahara S, Mccoy RJ, Buene L, Vandenberg MJ. Thin Solid Films 1981;84: [26] Hu YC, Lin YH, Kao CR, Tu KN. J Mater Res 2003;18: