Electromigration time to failure of SnAgCuNi solder joints

Transcription

1 JOURNAL OF APPLIED PHYSICS 106, Electromigration time to failure of SnAgCuNi solder joints Cemal Basaran, 1,a Shidong Li, 1 Douglas C. Hopkins, 1 and Damien Veychard 2 1 Electronic Packaging Laboratory, State University of New York at Buffalo, Buffalo, New York , USA 2 PTM/CPA Group, STMicroelectronics, Grenoble F-38019, France Received 6 March 2009; accepted 23 May 2009; published online 9 July 2009 Electromigration time to failure and electrical resistivity of 95.5%Sn 1.5%Ag 0.5%Cu 0.03W%Ni microelectronics solder joints have been investigated experimentally. A Black-type electromigration time to failure equation is developed to describe the time to failure versus current density and temperature. It is observed that resistance of a solder joint is not just a function of the temperature but also a function of the current density. The activation energy over the range of C is measured to be ev, and the current density exponent is found to be Itis also shown that the most commonly used Black s electromigration time to failure equation cannot be used for solder joints American Institute of Physics. DOI: / I. INTRODUCTION Insatiable demand for higher functionality and miniaturization of electronics leads to higher density integrated circuit IC devices with much smaller solder joints, which are subjected to much higher current densities. As result, microelectronics and power electronics solder joint failure due to electromigration has been a major concern, recently. According to International Technology Roadmap for Semiconductors in next generation electronics electromigration will be the dominant failure mode. Electromigration is a mass transport process due to momentum exchange between the mobile valence electrons and atoms ions. Due to the scattering phenomenon valence electrons bump into atoms and electron wind force moves the atoms ions in the direction of electron movement. Hence, electromigration driving force moves the mass from the cathode side to anode side. As a result, on the anode side, the mass accumulation causes local compression and eventually mass is squeezed out of material surface to form protrusions called hillocks. 1,2 When the protrusions contact with circuits nearby, short circuit failure occurs. While on the cathode side, mass depletion causes tension and vacancy accumulation. Voids, which nucleate under tension, will grow and coalesce causing increased localized resistivity and increased current crowding with eventual circuit failure. The latter mode is the major failure phenomenon that this study is concerned with. Basaran et al., 1 Abdulhamid et al., 3 Bastawros and Kim, 4 Basaran and Abdulhamid 5 have shown that when temperature gradient is large enough, usually 1000 C/cm, thermomigration can overcome electromigration forces and can be the dominant failure mechanism. Thermomigration is mass transport process due to temperature gradient, which happen in solids only under large temperature gradients. a Electronic mail: cjb@buffalo.edu. A. Black s law In 1967, Black postulated that the lifetime of a metal thin film on a substrate is inversely proportional to the square of the current density and has an Arrhenius relation with activation energy, consistent with grain boundary diffusion. 6 Based on experimental data, he proposed the following median time to failure equation for electromigration failure of thin films; exp t 50 = A E a j n 1 kt, where t 50 is median time to failure defined by the time at which 50% of a large number of identical samples have failed, A is an empirical material constant, j is the current density, n is the current density exponent found to be 2 in Black s experiments, E a is the activation energy consistent with grain boundary diffusion activation energy, k is Boltzmann s constant, and T is the absolute temperature. It has been reported in literature 7 that for solder joints, Black s equation cannot directly be used due to the influence of current crowding and thermal gradient, which does not exist in straight thin films Black tested. Current crowding is nonuniform distribution of current density, similar to stress concentration in solid mechanics. The major purpose of this project is to develop a similar simple electromigration time to failure relationship for SACN solder joints. II. EXPERIMENTAL SETUP The test vehicles, shown in Fig. 1, were manufactured by STMicroelectronics. The lead-free solder joints in the test vehicles have a composition by weight percentage of 95.5% tin, 1.0% 1.4% silver, 0.4% 0.6% copper, and 0.015% 0.03% nickel. The test vehicles is an actual mm 3 ball grid array BGA substrate is soldered to a mm 3 BT printed circuit board substrate. Solder joint dimensions are solder ball diameter is 320 m, standoff height is 127 m, and passivation opening diameter is 250 m. During the experiments four-point method is employed to measure the solder joint resistance. Software was /2009/106 1 /013707/10/$ , American Institute of Physics

2 Basaran et al. J. Appl. Phys. 106, FIG. 1. Color online Test vehicle and measurement configuration. developed in LABVIEW to coordinate all the testing instrumentations and collect the data every 5 s. Each test vehicle illustrated in Fig. 1 has several solder joints that can be tested to failure independently. During testing solder joints were subjected to dc current only. The testing started with stressing a solder joint to a 8 A for about 12 h at room temperature about 25 C, then current level was stepped up to 9 A, and then to 10 A. Measured resistance time history is shown in Fig. 2. It was not until 10 A that a change in the slope of the electrical resistance time history was observed. It should be pointed out that when a current is applied, due to Joule heating, resistance increases sharply at the beginning but then stabilizes with temperature. Change in resistance discussed in the paper is not the change that happens immediately following the current application but change after stabilization. By stressing the test solder ball with 10 A current loading for 12 h, about 3% irreversible resistance change was observed, Fig. 2, after removal of current and cooling down to room temperature. As a result of these initial trials, room temperature experiments started from 10 A. Based on these experiments we observed that at room temperature, the solder joint will not fail or it will take very long time to fail when it is stressed with a current lower than 10 A. The corresponding nominal current density in the solder joint is about A/cm 2, where the nominal current density is defined as the ratio of applied current with respect to solder mask opening area. This is in agreement with earlier studies reported by Basaran and co-workers. 8 11,5,12,13 It was observed that the ambient temperature has significant effect on failure current density. When the temperature was raised to 60 C, the electrical resistance started to change, poststabilization period, when the applied current is 9 A or higher. In this study, measuring the temperature coefficient of electrical resistance in solder joints was one of the main tasks. However, it was quickly discovered that resistance in solder joints subjected to high current density is also a function of the current density. Three samples, labeled P1, P2, and P3 individually, were placed in a thermal chamber. The electrical resistance was measured in room temperature about 25 C ; then the ambient temperature was increased to 40, 50, and 60 C. At each temperature level, five magnitudes of electrical currents I=0.2, 2, 4, 6, and 8 A, for FIG. 2. Color online Electrical resistance change under current loadings of 8, 9, and 10 A and then unloading to 1 A.

3 Basaran et al. J. Appl. Phys. 106, FIG. 3. Color online Relationship of electrical resistance, current, and measured temperature in Sample P1. sample P3 also I=10 A were applied to measure the dependence of electrical resistivity on ambient temperature and current density. Following the first battery of tests, a second batch of test vehicles was tested under three different ambient temperatures: 25, 60, and 120 C under higher currents than the first batch. In this series of tests applied current ranged from 9 and 12.5 A, until solder joint failed. In these experiments, the failure criterion was a 10% change in resistance of test solder ball after accounting for the influence of both ambient temperature and Joule heating effect. The electrical resistance of a conductor under the high current loading can be expressed by R = R 0 + R T + R J + R EM, 2 where R 0 is the initial resistance defined before, R T is the resistance change due to ambience temperature, R J is the resistance change due to Joule heating, and R EM is resistance change due to electromigration degradation. Both R T and R J reach steady state in a relatively short time. However, the resistance change due to electromigration is a much longer process. Electromigration degradation continuously increases the resistance of the solder joint and finally causes device failure. In this work, the failure criteria threshold R EM is set as 10% of R 0 + R T + R J. This is a reasonable failure criterion because 10% resistance change can lead to serious signal degradation in the device. III. EXPERIMENTAL RESULTS A. Temperature and current dependency of resistance It is difficult to eliminate the influence of Joule heating during the experiments. As a result, a numerical scheme was adopted in order to remove the Joule heating induced resistance change from the total resistance change. Thermocouples were placed on the surface of the tested solder balls to measure the temperature. The relationship between mea- FIG. 4. Color online Relationship of electrical resistance, current, and measured temperature in Sample P2.

4 Basaran et al. J. Appl. Phys. 106, FIG. 5. Color online Relationship of electrical resistance, current, and measured temperature in Sample P3. sured solder joint temperature and electrical resistance for test vehicles P1, P2, and P3 are plotted in Figs. 3 5, respectively. In these figures, we observe that when the applied current is constant, the electrical resistance change is linearly related to the actual measured temperature, which can be expressed by the following relation: R = R ref + T, where R ref is the electrical resistance at a reference temperature which is 0 C in this study, T is temperature change, and stands for the temperature coefficient of resistance. However, as can be seen in Figs. 3 5, once the applied current is increased, both R ref and change, which means that electrical resistivity of SACN solder alloy is both temperature and current dependent. In order to better describe the electrical resistivity of SACN solder joints, a model considering both temperature dependence and current dependence is proposed below. By using the method of least squares, each set of experimental data can be fitted to the following type of polynomial equation with the curve fitting coefficients listed in Table I. For sample P3 three extra data points were obtained with I =10 A, R = R 0 1+ T + I 2, where R 0 is the initial electrical resistance, which is the resistance at T=0 C obtained by extrapolation, is the temperature coefficient of resistance, is the current coefficient TABLE I. Coefficients used for least-square curve fitting 3 4 of resistance, T is the ambient temperature in celsius, and I is the direct current applied for measurement. Using the coefficients given in Table I the experimental data were fitted to Eq. 4 for each sample. However, combining the data from all three samples into one regression curve yields smaller R-square value due to apparent differences in initial manufacturing defects, which are very common in solder joints. On the other hand, if we take R 0 measured from each sample second column in Table I separately, however, use the average values of and from three samples, using Eq. 4 leads to R-square coefficient of for all 63 data points. Therefore, the relationship between electrical resistivity and ambient temperature and electrical current can be expressed by R = R T I 2. 5 IV. FINITE ELEMENT SIMULATIONS A finite element analysis was conducted, using ABAQUS finite element analysis FEA code, to simulate the electrical and thermal responses of the test vehicles. For Joule heating analysis three-dimensional 3D DC3D20E element for the solder joints of interest, 3D element DC3D8E was used for other solder balls and the other components in the test vehicles, and 3D element DC3D6E was used for patches which are used to mesh the irregular parts. The finite element mesh is shown in Figs At ambient room temperature T=25 C, when 10 A current is applied to the solder joint in position 1, the temperature contour in the test vehicle is shown in Fig. 9. FEA Sample R 0 m Number of data points R-square P P P Average FIG. 6. Color online Finite element mesh-global appearance.

5 Basaran et al. J. Appl. Phys. 106, FIG. 7. Color online Finite element mesh-bga. simulations indicate that the maximum temperature occurs at the tapered narrowing section of the copper trace, labeled T1 in Fig. 9, where temperature is 90.8 C. During FEA self-heat convection and heat radiation on the exposure surface were both taken into consideration. The simulation results are very close to what is measured from the samples, for example, for solder joint S15P1 measured temperature is 89.6 C versus FEA computed temperature of C. The difference between simulation and measured value is small. In Fig. 10 it can be seen that the temperature inside the test solder ball varies from 81 to 85 C, which shows that the thermomigration effect can be neglected, 8,3,11 Simulations also indicate that if we place the thermocouple at location T1, the temperatures of the tested solder joints are overestimated by a few degrees of celsius. Therefore, it is reasonable for us to assume that the measured temperature at T1 is the service temperature of the tested solder joint. Current crowding in solder joints subjected to high current densities is well known 14,15,3 and it can be observed in Fig. 11, which shows the current density contour in solder joint P1. The nominal current density in the solder joint for a current of 10 A is A/cm 2, where the current density is defined by the quotient of the applied current to the solder mask opening area. However, from Fig. 11 we observe that the current distribution is not uniform throughout the solder joint. In the left lower corner where the current enters from the solder joint to the copper trace, the current density j= A/cm 2 is about 1.5 times larger than the nominal value, while in the deep blue region of Fig. 11, the current density is smaller than half the nominal value. Therefore, the geometry of the solder joint is an important factor for its service life due to current crowding effect. Usually, the maximum current density at current crowding region is much higher than nominal current density, especially in drumlike solder joints. 16,5 V. TIME TO FAILURE EXPERIMENTS Test vehicles were also tested to failure to determine their time to failure. Failure was defined as 10% resistance change after taking out Joule heating and current density effects. After an experiment starts initially, resistance will increase during warm up period due to Joule heating and then stabilize and remain constant for most the experiment. This 10% change in resistance was defined after the steadystate temperature in the solder joint was reached. In solder joints resistance was also observed to be a function of the current density. Therefore the influence of current density was also removed before calculating the 10% change. Therefore 10% change was due to electromigration only. The threshold was chosen because of electrical signal integrity considerations. Time to failure TTF experiment results are listed in Table II. From these results we observe that there are three types of failure and associated evolution of the electrical resistance under different levels of current and temperature. Type 1. Under a given ambient temperature, when the current loading is small, there is no significant increase in the resistance due to electromigration. The electrical resistance remains constant after Joule heating induced part reaches a steady state. The tests were stopped after a considerably long stressing time if no failure or change in resistance was ob- FIG. 8. Color online Finite element mesh-circuit in position 1.

6 Basaran et al. J. Appl. Phys. 106, FIG. 9. Color online Temperature contour under room temperature with current loading of 10 A. served. A typical Type 1 electrical resistance evolution can be seen in Fig. 12 where joint S15P1 was stressed by a current of 10 A at room temperature. The test was terminated after 500 h with negligible irreversible resistance change. Type 2. When current density is large enough, the resistance evolution follows a bilinear curve. In this case, the influence of electromigration is noticeable, which leads to increasing resistance. For example, Fig. 13 shows the resistance evolution in solder joint S13P1, where we observe that after the warming up initial Joule heating period, the electrical resistance increases at an almost constant rate due to electromigration. We should emphasize that during these experiments temperature was continuously monitored at the solder joints. Temperature in the solder joint stays stable, see Fig. 13, after the steady state is reached. Therefore we are confident that change in resistance is not due to Joule heating but electromigration. After the 10% threshold our predefined failure limit was reached, the current loading on solder joint S13P1 was kept for another 185 h. Then, when the current was removed the solder ball resistance, while it is still hot, was measured to be m 78% increase from initial resistance. After the sample was allowed to cool down, the resistance of the test ball was measured to be m at 22.6 C, which is 16.4% irreversible change

7 Basaran et al. J. Appl. Phys. 106, FIG. 10. Color online Temperature contour of test solder ball P1 room temperature, I=10 A. compared to the initial resistance of m measured at 23.8 C. Type 3. When current is very high 12 A, heat generated by Joule heating melts the solder joint in a very short time. For example, sample S21P3 failed in 72 s, as shown in Fig. 14. Because the entire process happens in a very short period, electromigration does not play an important role in Type 3 failure. The mechanism can be explained by the de- FIG. 11. Color online Current density contour of test solder ball P1 room temperature, I=10 A.

8 Basaran et al. J. Appl. Phys. 106, TABLE II. TTF of SACN solder joints. Test I A R T=0 C m T solder C R reference m R threshold m TTF h Failure type S21P No failure 1 S18P No failure 1 S20P No failure 1 S22P S21P S22P3 a S18P S15P No failure 1 S5P No failure 1 S2P S17P S17P S16P S13P S13P S14P S12P S4P S1P S8P S6P S19P S17P S19P S11P S9P S3P S4P S4P S20P S19P S21P a We believe that this data point is anomaly. FIG. 12. Color online Sample S15P1 I=10.0 A tested under room temperature failure mode: type 1.

9 Basaran et al. J. Appl. Phys. 106, FIG. 13. Color online Sample S13P1 I=10.5 A tested under room temperature failure mode: type 2. generative amplification effect of Joule heating to electrical resistance. Increasing electrical resistance produces more Joule heat under the constant current loading. Once the rate of heat generation exceeds the heat removal capacity of the structure, temperature rises and conversely leads to the increasing electrical resistance, which again raises the temperature until the solder joint melts. Type 3 failure observed in solder joint SP22P3 is an aberration probably due to manufacturing defect, such as a void which is common in BGA solder joints. Using standard least squares method, an empirical time to failure formula is developed from the data presented in Table II. For consistency, Type 3 failure data points are excluded. In addition data points S4P1, S17P2, and S13P3 were also excluded because they were way outside the norm. The following electromigration time to failure equation is 1 TTF = A ev/kt 1 j e R T=0 C Moreover, in numeric terms the equation can be given by ev ln TTF = kt ln j ln R T=0 C. 6 7 FIG. 14. Color online Sample S21P3 I=12 A tested under room temperature failure mode: type 3.

10 Basaran et al. J. Appl. Phys. 106, TABLE III. Activation energy for SnAgCu solder alloy. Q kcal/mol The first three terms of Eq. 6 are similar to that of Black s TTF equation, where A is a constant, which depends on the definition of failure threshold. For 10% net electrical resistance change after temperature reaches steady state, A =e T is the solder joint temperature in kelvins and j is nominal current density in A/cm 2. The last term in Eq. 6 accounts for the manufacturing defect in the specimens, where R T=0 C has been defined in Sec. IV. Manufacturing defects lead to the scattered data we measure for the initial resistance. The coefficient of determination of Eq. 7 is R 2 =0.81. The activation energy over the range of C is measured to be ev, which is very close to the values reported in the literature for SAC solder alloy, Table III 0.72, , , 14,17 and 0.76 ev Ref. 15 for SAC solder alloy. The current density exponent is found to be ,20 It was also observed that TTF of the SACN solder joint is sensitive to its initial manufacturing defects, which leads to different initial resistance values. 19,20 VI. CONCLUSIONS Reference 0.72 Reference Reference Reference Reference 15 In this study electromigration failure of SnAgCuNi BGA solder joints was studied experimentally. It is observed that resistance of solder joints is not just a function of the temperature but also a function of current density. An equation is proposed that correlates current density and temperature to resistance of a solder joint. It is also shown that Black s TTF equation cannot be used for solder joints. An electromigration time to failure equation for SACN BGA solder joints is proposed. Activation energy for SACN solder alloy is found to be very close to published values of activation energy of SAC solder alloy. In TTF equation current density exponent for SACN solder joints is 8.6. Initial manufacturing defects cause serious differences in initial resistance of solder joints. ACKNOWLEDGMENTS This research project has been partly sponsored by the STMicroelectronics packaging team ATM which belongs to PTM/CPA groups in Grenoble, France. The project has also partly been sponsored by US Navy ONR under the direction of program Director Terry Ericsen. 1 C. Basaran, M. Lin, and H. Ye, Int. J. Solids Struct. 40, I. A. Blech, J. Appl. Phys. 47, M. Abdulhamid, C. Basaran, and Y. S. Lai, Thermomigration vs. electromigration in lead-free solder alloys, IEEE Trans. Adv. Packag. in press. 4 A. F. Bastawros and K. S. Kim, J. Electron. Packag. 120, C. Basaran, S. Li, and M. Abdulhamid, J. Appl. Phys. 103, J. R. Black, Sixth Annual Reliability Physics Symposium, 1967 unpublished, pp M. Lin, A damage mechanics framework for electromigration failure, Ph.D. thesis, University at Buffalo, State University of New York, H. Ye, C. Basaran, and D. Hopkins, Appl. Phys. Lett. 82, H. Ye, C. Basaran, and D. Hopkins, Int. J. Solids Struct. 40, H. Ye, C. Basaran, and D. Hopkins, Microelectron. Reliab. 43, M. Abdulhamid and C. Basaran, J. Electron. Packag. 131, S. Li, M. Abdulhamid, C. Basaran, and Y. S. Lai, Damage mechanics of low temperature electromigration and thermomigration, IEEE Trans. Adv. Packag. 32, C. Basaran and M. Lin, Mech. Mater. 40, S. Gee, L. Nguyen, J. Huang, and K.-N. Tu, International Wafer-Level Packaging Conference, San Jose, CA, USA, 2005 unpublished. 15 L. Xu, J. Pang, F. Ren, and K. Tu, J. Electron. Mater. 35, S. Li and C. Basaran, Mech. Mater. 41, S. L. Allen, M. R. Notis, R. R. Chromik, and R. P. Vinci, J. Mater. Res. 19, N. Dariavach, P. Callahan, J. Liang, and R. Fournelle, J. Electron. Mater. 35, H. Tang and C. Basaran, Int. J. Damage Mech. 10, H. Ye, C. Basaran, and D. C. Hopkins, Int. J. Damage Mech. 15,