Low Cycle Fatigue Testing of Ball Grid Array Solder Joints under Mixed-Mode Loading Conditions

Tae-Sang Park Mechatronics & Manufacturing Technology Center, Corporate Technology Operations, Samsung Electronics Co., LTD, 416, Maetan-3Dong, Yeongtong-Gu, Suwon-City, Gyeonggi-Do, 443-742, Korea e-mail: taesang.park@samsung.com Soon-Bok Lee CARE-Electronic Packaging Laboratory, Department of Mechanical Engineering, Korea Advanced Institute of Science and Technology, 373-1 Guseong-dong, Yuseong-gu, Daejeon 305-701, Korea e-mail: sblee@kaist.ac.kr Low Cycle Fatigue Testing of Ball Grid Array Solder Joints under Mixed-Mode Loading Conditions To give a proper and accurate estimation of the fatigue life of ball grid array (BGA) solder joints, a mechanical fatigue test method under mixed-mode loading is proposed. Experiments were conducted with 63Sn/ 37Pb and Sn/ 3.5Ag/ 0.75Cu solder joints in room temperature. The mechanical low cycle fatigue tests were performed under several loading angles. The loading angle is controlled by several grips which have specific surface angle to the loading direction. Constant displacement controlled tests are performed using a micro-mechanical test apparatus. It was found that the normal deformation significantly affects the fatigue life of the solder joint. Throughout the whole test conditions at room temperature, Sn/ 3.5Ag/ 0.75Cu solder alloy had longer fatigue life than 63Sn/ 37Pb alloy. Failure patterns of the fatigue tests were observed and discussed. A morrow energy model was examined and found to be a proper low cycle fatigue model for solder joints under mixed mode loading condition. DOI: 10.1115/1.1871192 Keywords: Electronic Packaging, Ball Grid Array (BGA), Solder Joints, Low Cycle Fatigue, Mixed-Mode Loading Introduction Since modern electronics technology requires complex, high density, and high-speed devices, the use of surface mount components has been increased. Due to the high degree of integration, high component density and the resulting high production cost of the surface mount technology, the modern advanced SMC assembly has imposed more stringent reliability requirements on packaging design. The ball grid array BGA package is a second level package, which is attached to a printed circuit board PCB through solder bumps to form a final second level assembly. There is an increased thermal fatigue problem with this type of attachment because the module is soldered directly to the PCB. With no lead to take up the thermal strain developed by the mismatch in the coefficient of thermal expansion CTE between the module and the PCB, the BGA packages are prone to thermal fatigue failure. This problem has initiated a considerable amount of research into the fatigue of the leadless connections. The combinations of temperature changes, either due to external environment or power switching, and materials that possess different coefficients of thermal expansion, produce substantial cyclic strains within the solder. As a result of the mismatch in the coefficient of thermal expansion CTE and thermal gradient in the device, the devices undergo not only shear deformation but also warpage and distortion during thermal excursions encountered in manufacturing process and actual operating conditions. Moreover, in some cases, such as mobile phones, the devices can be subjected to various mechanical and environmental stresses. Therefore, in most applications, solder joints are subjected to the combination of shear and tensile loading. The fatigue tests under mixed loading condition are essential to account for the solder joint reliability. Much research has been conducted on the mechanical fatigue tests of solder material such as uniaxial tension-tension 1, simple-shear 2, thermomechanical shear 3, and multiaxial tension-torsion test 4. However, the mechanical properties of the Contributed by the Electronic and Photonic Packaging Division for publication in the JOURNAL OF ELECTRONIC PACKAGING. Manuscript received October 1, 2003; revision received May 26, 2004. Review conducted by: B. Courtois. solder joints are dependent on many factors such as the size, geometry, and microstructure of the solder joint under testing and service conditions. Thus in order to get a better understanding of the solder joint behavior, it is important to study real BGA solder joints attached to printed circuit boards. Recently, micromechanical testing schemes have been developed for the structural analysis of electronic packaging and have accomplished many tasks that could not have been performed by conventional experimental methods 5 7. In the present work, low cycle fatigue tests of solder joints under mixed-mode loading conditions were performed. A precise mechanical fatigue tester was constructed and several grips were designed to change the loading direction. Strain-based models, such as the Coffin-Manson model, have been generally used to characterize low cycle fatigue behaviors. In multi-axial loading states, Cortez et al. 4 suggested that the torsion and tension uniaxial fatigue lifetime data of bulk 63Sn/ 37Pb material are best correlated by the von Mises effective plastic strain. However, it is practically very difficult to obtain a single plastic strain value in a solder joint because of the complex stress state. In contrast, it is much easier to calculate energy density from the low cycle hysteresis loops for any type of solder joints under test. Therefore, in recent years, energy-based low cycle fatigue models have been increasingly used for solder alloys 2,8,9. In this study, the Morrow energy model was examined as to whether energy density can be a parameter governing the fatigue life of solder alloy under mixed-mode loading conditions. Experimental Procedures Test Apparatus. A micro-mechanical test apparatus was developed for experimental approaches on the reliability evaluation of electronic packaging. The automated testing system consisted of a testing machine part and a control and sensing part as shown in Fig. 1. The testing machine applies a load to the specimen via a servomotor attached to a ball-screw-driven rail table. The displacement on the specimen can be feedback controlled with the resolution of 50 nm by a personal computer and data acquisition system. The traveling range is up to 100 mm, which allows various gripping systems and testing styles. The capacity of the load cell is 50 kgf and its signal can be acquired with the resolution of ±2 mv in the range of 10 V. Journal of Electronic Packaging Copyright 2005 by ASME SEPTEMBER 2005, Vol. 127 / 237

Fig. 3 Reflow temperature profile Fig. 1 a Schematic diagram and b photograph of the constructed micromechanical testing system Specimen. The fatigue specimen consists of a ball grid array of reflowed solder attached to mating FR-4 substrates. Alloys 63Sn/ 37Pb and Sn/ 3.5Ag/ 0.75Cu were tested. The solder balls diameter 0.76 mm with 1.27 mm pitch were mounted to soldermask-defined FR-4 substrates. The 63Sn/ 37Pb and Sn/ 3.5Ag/ 0.75Cu solder balls are manufactured by Alpha metal and Duk San Hi-metal, respectively. The thickness of the FR-4 is 0.8 mm and the copper pads had pad diameter of 0.787 mm 31 mil, solder mask opening diameter of 0.584 mm 24 mil, and 35 m thickness of copper, plated with 5 m of nickel, and a less than 1 m gold flash. The copper lines between pads of the mating FR-4 substrate connected the nine solder joints in series to check the electrical connectivity Fig. 2 a. The specimen was formed using the two reflow processes, bumping and joining, in a N 2 environment oven. The standoff heights of reflowed 63Sn/37Pb and Sn/3.5Ag/0.75Cu were 500 and 520 m, respectively. The reflow temperature profile is shown in Fig. 3. The peak temperature was about 30 40 C above melting temperature. The fabricated specimens were subjected to thermal aging. The specimens were aged for 2 months at room temperature. A few of the Fig. 2 a Fr-4 substrate dimensions mm, b schematic diagram, and c photograph of a specimen Fig. 4 Optical image of microstructure of a 63Sn/37Pb and b Sn/3.5Ag/0.75Cu 238 / Vol. 127, SEPTEMBER 2005 Transactions of the ASME

Fig. 5 Test grip configurations. Loading angle of each specimen is a 0 deg pure tension, b 27 deg, c 45 deg, d 63 deg, e 90 deg pure shear Fig. 7 Performance curve of the testing system at displacement control: a displacement versus time and b force versus time profile specimens were aged for 72 h at 100 C. Optical images of the matrix microstructure of 63Sn/ 37Pb and Sn/ 3.5Ag/ 0.75Cu are shown in Fig. 4. After aging, the specimens were bonded to steel grips using a cyanoacrylate adhesive. To change the loading angle, several grips were prepared as shown in Fig. 5. To increase machine stiffness and reduce out-of-plane forces, no other moving stage was used. Instead of the stage, a rigid alignment grip was used to align grips. An ac-type LVDT linear variable differential transformer was used to measure the relative displacement between grips as shown in Fig. 5. The relative displacement was controlled with the resolution of 50 nm by a personal computer and data acquisition system. Various displacement control shapes were possible. In the Fig. 6 AC LVDT and grips for measuring the relative displacement between specimen grips Fig. 8 90 deg Force-displacement hysteresis under loading angle Journal of Electronic Packaging SEPTEMBER 2005, Vol. 127 / 239

Fig. 9 Force versus displacement of Sn/3.5Ag/0.75Cu solder joints at ±10 m displacement stroke present work, tests were performed at saw-type displacement waves with constant displacement 2 m/s. Applied displacement amplitude was changed from 4 to 25 m. The typical displacement and load profile for a controlled displacement range test of 15 m is shown in Fig. 6. The nine solder balls in a specimen are connected within a series circuit that is monitored for continuity. The resistance change resulting from the applied cyclic strain in the testing is measured during the cyclic fatigue testing. The resistance is measured using a Wheatstone resistance bridge. Using the data acquisition system, the resistance signal was monitored and recorded at the desired time interval. Experimental Results and Discussion The mechanical fatigue test under mixed loading was performed in a displacement-controlled mode in which the specimen was subjected to a fully reversed triangle mixed load at a constant displacement rate of 2 m/s Fig. 7. The applied displacement on the specimen was controlled to be constant until failure occurred. At least two specimens on each test condition were tested at various displacement amplitudes and loading angles. Figure 8 shows the first stabilized hysteresis loops with varying displacement amplitudes in loading angle 90 deg. Throughout the whole tests, every load versus displacement hysteresis was recorded with a rate of 200 scans per cycle. The material behavior under ±10 m constant displacement control is shown in Fig. 9. Force displacement response is very different between the different loading angles. In the 0 deg loading angle, i.e., pure tension compression loading case, the force displacement ratio is very stiff. In the present work, alloys 63Sn/37Pb and Sn/3.5Ag/0.75Cu were tested. Material behaviors of these two alloys are compared in Fig. 10. The slope of the force-displacement curve of Sn/3.5Ag/0.75Cu is somewhat lower than that of 63Sn/37Pb. The elastic-behavior-obtained unloading curves in Figs. 8 10 includes not only the elastic behavior of solder joints but also that of organic substrates. The elastic modulus of the organic substrate is very low comparied with that of solder material. Therefore, the elastic behavior of the specimen highly depends on the elastic modulus of the substrate and the geometric configuration of the specimen, such as the height of the solder joints. Under the shear loading, the low modulus of the organic substrate enabled the solder joint to rotate on the substrate, which made the severe dependency on the loading angle in Fig. 9. The stiffer elastic behavior of the 63Sn/37Pb solder joint in Fig. 10 is due to the smaller standoff height of the solder joint. In order to make the phenomena clear, monotonic tension tests were performed at a Fig. 10 Material behavior of 63Sn/ 37Pb and Sn/ 3.5Ag/ 0.75Cu: a loading angle 0 deg, b loading angle 45 deg, and c loading angle 90 deg constant displacement rate of 2 m/s. As shown in Fig. 11, the ultimate tangent strength of Sn/ 3.5Ag/ 0.75Cu is slightly higher than that of the 63Sn/37Pb solder joint, but the yield strength is lower. The failure patterns of the monotonic tension tests showed the typical ductile fracture as shown in Fig. 12. The fracture surface directions were about 45 and 0 deg to the loading direction under the tension and shear loading conditions, respectively, along the plane of maximum shear stress. The force amplitude decreased with the number of cycles in the course of the test. Typical hysteresis curves of the first and sub- 240 / Vol. 127, SEPTEMBER 2005 Transactions of the ASME

Fig. 11 Monotonic tension test of 63Sn/ 37Pb and Sn/ 3.5Ag/ 0.75Cu under loading angle 0 deg and loading angle 90 deg sequent cycle are shown in Fig. 13. The area inside the hysteresis refers to the plastic work dissipated by the solder material. This work affects either microstructural changes or crack propagation 7. Throughout all the loading angles, the first cycle hysteresis is symmetric in compression and tension loading areas. However, in later subsequent cycles, asymmetric hysteresis is shown except for the pure shear test. This asymmetry regarding tension and compression is due to the contact of two corresponding fracture surfaces in the cracked area. Both surfaces are pressed against each other in compressive loading. Excluding this distortion in compression loading, the hysteresis is almost symmetric in compression and tension loading. The load drop in the hysteresis loop is a measure of fatigue cracking and can be used as a measure of fatigue failure 10,11. This load drop is characterized by the load drop parameter = P M P N / P M, where P M is the maximum load range and P N is the load range for the nth cycle. In small displacement range tests of Sn/3.5Ag/0.75Cu, the load range increases during a few cycles and decreases as the number of cycles increases as shown in Fig. 14. However, with 63Sn/ 37Pb alloy, the hysteresis load decreased from the first cycle. Figure 15 shows the typical values of versus cycle number with Sn/3.5Ag/0.75Cu alloy under the loading angle of 90 deg. These load drop curves are used to determine the fatigue life as a function of applied displacement range. In Fig. 16, the fatigue lives for 20, 50, and 70 percent drops in the hysteresis load are displayed. In order to determine the lifetime of the specimens, the change in resistance was recorded at a rate of 10 scans per second. A Fig. 13 Applied force decreases as cycle increases: a loading angle 0 deg, b loading angle 45 deg, c loading angle 90 deg for Sn/3.5Ag/0.75Cu at ±10 m displacement stroke Fig. 12 SEM photographs of monotonic tension test of 63Sn/37Pb solder joint under a loading angle 0 deg and b loading angle 90 deg and Sn/3.5Ag/0.75Cu under c loading angle 0 deg and d loading angle 90 deg typical resistance versus cycles curve is compared with load drop in Fig. 17. As the cycles increased, the magnitude of the resistance became larger. The increase in the resistance is caused by fatigue damage accumulation in the solder bumps. As shown in Fig. 17, there are abrupt increases in the resistance at 50% load drop. However, this abrupt increase in the resistance occurred arbitrarily from 50% to 80% load drop depending on the loading angle. This phenomenon was not consistent throughout the whole experiment and had large deviation in fatigue life compared to the load drop method because the resistance of the specimen was mainly affected by the most damaged solder joint. On the other hand, the Journal of Electronic Packaging SEPTEMBER 2005, Vol. 127 / 241

Fig. 14 Cyclic behavior of Sn/3.5Ag/0.75Cu at ±10 m displacement stroke 90 deg loading angle load drop was affected by the averaged damage of the nine solder joints in the specimen. Therefore, in this study, the load drop parameter was employed as failure criteria. The fracture patterns of the fatigue test were shown in Fig. 18. The fatigue crack propagated in solder material along the solder mask irrespective of the loading angles. The correlation of life cycles versus displacement stroke is shown in Figs. 19 and 20. The fatigue life curves of solder bumps with various displacement ranges become straight lines within the same loading angle. Between the loading angles, however, there are apparent discrepancies. All over the whole test conditions, the Sn/ 3.5Ag/ 0.75Cu solder has longer a fatigue life. In this study, it is assumed that the plastic energy density dissipated in a solder joint can be closely related to the fatigue life of a solder joint. As shown in Fig. 18, the plastic strain energy was almost dissipated in the stress-concentrated area of the solder joint due to the geometric shape of the specimen. As a fatigue model, the Morrow energy mode was examined. The model predicts fatigue life N f in terms of the plastic strain energy density W, as shown below: N m f W = C, 1 where m is the fatigue exponent, and C is the material ductility coefficient. The plastic strain energy density was determined from the area within the maximum force applied hysteresis loop. Figure 21 shows the fatigue life of 63Sn/37Pb and Sn/3.5Ag/0.75Cu alloy as a function of strain energy density. In Fig. 21 hollow symbols indicate the data of 63Sn/ 37Pb, while the solid symbols indicate Sn/3.5Ag/0.75Cu alloy. The results in Fig. 21 were obtained at a fixed test temperature 25 C but at different loading Fig. 16 Fatigue life versus displacement range for Sn/ 3.5Ag/ 0.75Cu, for different load drop definitions of failure a loading angle 0 deg, b loading angle 45 deg, and c loading angle 90 deg Fig. 15 Load drop parameter versus cycles of Sn/ 3.5Ag/ 0.75Cu alloy under the loading angle of 90 deg angles and different total strain level. The Morrow model predicts a linear relationship between log N f and log W irrespective of loading angles. The material constant m of Sn/ 3.5Ag/ 0.75Cu and 63Sn/ 37Pb was 0.76 and 0.87, respectively. The constant m of the 63Sn/37Pb solder alloy obtained by Shi 9 and Solomon 2 was 0.7 and 0.57, respectively. This difference is due to the specimen shape, loading condition, and microstructure of solder. Summary and Conclusions To apply various micromechanical loads, a precise testing apparatus was developed. Isothermal low cycle fatigue testing was 242 / Vol. 127, SEPTEMBER 2005 Transactions of the ASME

Fig. 19 Fatigue life versus displacement range for Sn/3.5Ag/0.75Cu by 50% load drop Fig. 17 A typical resistance versus cycles curve and corresponding load drop curve as cycle increases loading angle 45 deg performed with the machine that provides cyclic mixed-mode strain in the solder balls with varying loading angles. The loading angles are controlled by the angle of loading directions. The low cycle fatigue test of solder joints at room temperature under mixed-mode loading cases was performed. The failure of the solder bump was detected with the load drop method. Force displacement responses are very different for different loading angles. Normal deformation affects significantly the fatigue life of the solder joints. Throughout all the test conditions at room temperature, Sn/ 3.5Ag/ 0.75Cu solder alloy had a longer fatigue life than 63Sn/ 37Pb alloy. Fig. 20 Fatigue life versus displacement range for 63Sn/ 37Pb by 50% load drop Fig. 21 Fatigue life versus strain energy density As a fatigue model, the Morrow energy model was examined. The model predicts a linear relationship between log N f and log W irrespective of loading angle. Fig. 18 Photographs of a cross-sectional view of the failure path and substrate pads after b 70% load drop under loading angle 0 deg, c 50% load drop and d 90% load drop of Sn/ 3.5Ag/0.75Cu solder joint under loading angle 90 deg Acknowledgments This work has been supported by KOSEF The Korea Science and Engineering Foundation through CEPM Center for Electronic Packaging Materials at KAIST. Journal of Electronic Packaging SEPTEMBER 2005, Vol. 127 / 243

References 1 Shi, X. Q., Pang, H. L. J., Zhou, W., and Wang, Z. P., 2001, Low Cycle Fatigue Analysis of Temperature and Frequency Effects in Eutectic Solder Alloy, Int. J. Fatigue, 22, pp. 217 228. 2 Solomon, H. D., and Tolksdorf, E. D., 1995, Energy Approach to the Fatigue of 60/40 Solder: Part I-Influence of Temperature and Cycle Frequency, ASME J. Electron. Packag., 118, pp. 130 135. 3 Palmer, M. A., Redmond, P. E., and Messler Jr., R. W., 2000, Thermomechanical Fatigue Testing and Analysis of Solder Alloys, ASME J. Electron. Packag., 122, pp. 48 54. 4 Cortez, R., Cutiongco, E. C., Fine, M. E., and Jeannotte, D. A., 1992, Correlation of Uniaxial Tensio-Tension, Torsion, and Multiaxial Tension-Torsion Fatigue Failure in a 63Sn-37Pb Solder Alloy, Proc 42nd Electronic Components and Technology Conf., pp. 354 359. 5 Lee, S. B., Park, T. S., and Ham, S. J., 2000, Experimental Techniques for Fatigue Reliability of BGA Solder Bumps in Electronic Packaging, JSME Int. J., Ser. B, 43 4, pp. 400 407. 6 Darveaux, R., and Banerji, K., 1992, Constitutive Relations for Tin Based Solder Joints, IEEE Trans. Compon., Hybrids, Manuf. Technol., 15 6, pp. 1013 1024. 7 Wiese, S., Jakschik, S., Feustel, F., and Meusel, E., 2001, Fracture Behaviour of Flip Chip Solder Joint, Proc. 51st Electronic Components and Technology Conf., pp. 2199 1306. 8 Pang, J. H. L., Tan, K. H., Shi, X., and Wang, Z. P., 2001, Thermal Cycling Aging Effects on Microstructural and Mechanical Properties of a Single PBGA Solder Joint Specimen, IEEE Trans. Compon. Packag. Technolo., 24 1, pp. 10 15. 9 Shi, X. Q., Pang, H. L. J., Zhou, W., and Wang, Z. P., 1999, A Modified Energy-based Low Cycle Fatigue Model for Eutectic Solder Alloy, Scr. Mater., 41 3, pp. 289 296. 10 Guo, Z., and Conrad, H., 1993, Fatigue Crack Growth Rate in 63Sn37Pb Solder Joints, ASME J. Electron. Packag., 115, pp. 159 164. 11 Solomon, H. D., 1991, Low Cycle Fatigue of Sn96 Solder With Reference to Eutectic Solder and a High Pb Solder, ASME J. Electron. Packag., 113, pp. 102 108. 244 / Vol. 127, SEPTEMBER 2005 Transactions of the ASME