Measurement of high electrical current density effects in solder joints
Microelectronics Reliability 43 (2003) 2021–2029 www.elsevier.com/locate/microrel
Measurement of high electrical current density effects in solder joints Hua Ye, Douglas C. Hopkins, Cemal Basaran
*
UB Electronic Packaging Laboratory, 102 Ketter Hall, SUNY at Buffalo, Buffalo, NY 14260, USA Received 29 January 2003; received in revised form 10 April 2003
Abstract Measuring mechanical implications of high current densities in microelectronic packaging interconnects has always been a challenging goal. Due to small interconnect size this task has typically been accomplished by measuring the change in electrical resistance of the joint. This measurement parameter is global and does not give local mechanical state information. Also, understanding strain evolution in the solder over time is an important step toward developing a damage mechanics model. The real-time, full-field, strain displacement in a eutectic Sn/Pb solder joint during electrical current stressing was measured with Moire interferometry (Post et al., High sensitivity Moire, Springer, New York, 1994) under in situ conditions. A finite element model simulation for thermal stressing was performed and compared with measured strain. The initial results show that the measured strain was largely due to thermal stressing versus the current density of 1.8 102 A/cm2 . A second Moire interferometry experiment with thermal control distinguishes deformation of solder joint due to pure current stressing above 5000 A/cm2 . Ó 2003 Elsevier Ltd. All rights reserved.
1. Introduction A perceived physical limit to increasing current density in both microelectronic and power electronic packaging is electromigration. Most research has focused on the thin, metal lines of ICs, and little on present day solder alloy interconnects [2–5] as used in higher levels of packaging. The development of a constitutive model is sought to predict the stress–strain field in a solder joint undergoing high-density electrical current flow. As a first step, the strain field in a BGA solder ball was measured using Moire interferometry. Strain evolution was derived with a natural cubic-spline interpolation method. This paper reports the preliminary findings. Previous work reported electromigration damage in solder joints undergoing current stressing of about 1.3 104 A/cm2 at ambient room temperature [2–4].
However, initial results presented here indicate that reliability may be greatly diminished at current densities much less than that due to complications from thermal stressing. The size of the tested solder bump was 3.8 mm in diameter and 1.70 mm in height. The maximum current applied to the solder bump was 10 A yielding a computed density of 175 A/cm2 . This density is much lower than the one to trigger electromigration as reported in the literature [2–4]. A finite element model (FEM) simulation of thermal stressing was performed and validated by experimental results, indicating that the strain under such current density was mostly due to thermal stressing. A second Moire interferometry experiment with good thermal control differentiates the deformation of the solder joint due to pure current stressing at a density above 5000 A/cm2 . 2. Experimental set-up
*
Corresponding author. Tel.: +1-716-645-2114x2429; fax: +1-716-645-3733. E-mail address: [email protected]ffalo.edu (C. Basaran).
Since this research solely focuses on the reliability of solder joints, two copper plates form the test vehicle
0026-2714/$ - see front matter Ó 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S0026-2714(03)00131-8
2022
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
a) Pin Test vehicle
Pin
Pin
Steel Steel
Dielectric polymer base plate b)
Fig. 2. The schematic diagram of the test vehicle fixture (a) plane view (b) the front view.
a) Cu plate b)
Cu plate
Fig. 1. The schematic diagram of the test vehicle (a) plane view and (b) side view (c) test vehicle after cross-sectioned with a diamond-wheel saw.
module instead of an FR4 carrier substrate and silicon die. The advantage of the copper plates as substrates is that the electrical connection can be made very simply and can carry a fairly large electrical current (a problem plaguing the thin-film metal lines in ICs [5]). A schematic diagram of the test vehicle is shown in Fig. 1a and b. A thin silicon dioxide layer was used as the solder mask on the copper plate. A continuous silicon dioxide layer was applied to the plate, patterned and etched. The photolithography process was done in class 1000 clean room. A hot plate was used to reflow the eutectic Pb37/ Sn63 alloy solder to form interconnect spheres. Spacers were used during reflow. Some oxidation of copper was observed and monitored. The test vehicle was first sliced through the center of the solder bump with a high precision diamond-wheel saw, further sliced to the shape that could be fit into the fixture (Fig. 1c), and finally polished with 1200 grit abrasive paper. An optical diffraction grating was replicated on the sectioned surface.
The fixture was composed of a dielectric polymer base plate with two steel plates on each end. The steel plates were affixed by screws and positioned via pins. The test vehicle was then clamped between the two steel plates and positioned by the pins. Electrical-test connections were made through the steel plates. The schematic diagram of the test vehicle fixture is shown is Fig. 2. A schematic diagram of the experimental test set-up is shown in Fig. 3. In this set-up, the applied current was computer controlled. The current was monitored through a Kelvin connection to a serial resistor. An infrared temperature sensor monitored the solder bump. All the controls, measurements and data logging were implemented in a single Labviewâ user subroutine. The monitored temperature was later used in the FEM thermal stress analysis. The module was mounted onto the Moire interferometry table and all wiring connected as shown. The optical set-up was then tuned to get a null field. Since Moire interferometry measurement relies on an initial reference field, the optical set-up was carefully protected from any disturbance during testing. The Moire interferometry technique is explained extensively by Post et al. [1]. The development of the imaging systems used for submicron displacement measurement using Moire interferometry technique
H-P6260B Power supply
Shunt resistor
H-P59501B DAC Microvolt meter
Infared thermal sensor Computer Microvolt meter Test vehicle fixture
Fig. 3. Schematic diagram of the test set-up.
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
3. Experimental results The test module was first energized with 5.8 A of DC current for 20 min. The current was then raised to 9.8 A, as shown in Fig. 4. The total current stressing time was 4 h. The Moire fringes of both U and V fields were recorded in real-time. After the current was turned off, deformation fringes were recorded for another 17 h. Fig. 5 shows the measured solder bump temperature time history. The solder bump temperature increased from room temperature to 29 °C immediately due to joule heating and remained there until the current was increased to 9.8 A. The temperature abruptly increased to 34 °C and gradually rose to 36.5 °C. After the current was turned off, the temperature abruptly decreased to 29 °C and, then, progressively decreased to room temperature in an hour.
38
Temperature on Test Sample
36 34
o
Temperature ( C )
has been described in detail in Zhao et al. [6,7]. The major advantage of Moire interferometry is its high sensitivity, high resolution, and the whole field view of deformation distribution of the specimen surface. Briefly, the optical diffraction grating is replicated on the specimen surface. The specimen grating diffracts the incident two coherent laser beams with certain incident angle, and in the direction normal to the specimen surface, two strong diffracted beams are obtained. When the specimen surface deforms, the optical diffraction grating deforms with the specimen, and the two diffracted beams in the normal direction generate feature interferometry pattern that represents the in-plane displacement distribution. This scheme applies to both horizontal and vertical direction, so that deformation in the two perpendicular directions can be obtained. The feature fringe pattern generated by the vertical two beams represents the vertical deformation field, and the fringe pattern generated by the horizontal two beams represents the horizontal deformation field [7].
2023
32 30 28 26 24 22 20 0
5000
10000
15000
20000
The temperature, measured with the infrared thermal sensor (with a spotsize of 10 mm in diameter), was the average temperature of the central area on the top surface of the test module. A fine thermal couple was used to pinpoint the temperature distribution on the module during current stressing. The temperature distribution was not uniform. An appreciable heat source was found at the interface of the copper plates and clamping fixture, and needed to be included in the analysis. When the temperature atop the copper plate was 36.5 °C, the temperature near the ends of copper plates increased to 57 °C. Figs. 6 and 7 are the initial Moire fringes of the solder bump before current stressing for U and V fields, respectively. The initial fringes and non-uniformity indicates initial strains during the sample mounting process.
Current (Amps)
8
Fig. 6. Initial U field.
4 2 0 -2 0
5000
10000
15000
20000
25000
30000
Time (second) Fig. 4. Profile of the current applied.
30000
Fig. 5. Measured temperature time history.
10
6
25000
Time (Second)
Fig. 7. Initial V field.
2024
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
Fig. 8. U field (after 02 h:22 min:0 s of stressing). Fig. 12. Lines of interest and position of points A and B.
direction of lower-order fringes [1]. The direction of fringe order, Ny was determined similarly by þy direction perturbation and observing the direction for the V field movement. Once the fringe orders Nx and Ny are determined, the strains at any point are given by, Fig. 9. V field (after 02 h:22 min:15 s of stressing).
oU 1 oNx ¼ ox f ox oV 1 oNy ¼ ey ¼ oy f oy oU oV 1 oNx oNy þ ¼ cxy ¼ þ oy ox f oy ox
ex ¼
Fig. 10. U field (2 h:0 min:0 s after current turned off).
Fig. 11. V field (2 h:0 min:15 s after current turned off).
Figs. 8–11 show the Moire fringe evolution with time during and after current stressing.
4. Strain analysis To extract the strains from the Moire fringes, determination of the direction of the fringe order was needed and performed as follows. During the experiment, the test module was perturbed in the þx direction (coordinates as shown in Fig. 12) and fringes moved toward the
where Nx and Ny are fringe orders in U and V fields, respectively, and f ¼ 2fx , fx is the frequency of diffraction grating (1200 lines/mm). With this frequency, the resolution of displacement is 0.417 lm. Instead of using linear interpolation, natural cubic-spline interpolation is used to approximate the derivative of fringe order more accurately. Three lines of interest are shown in Fig. 12. The distribution of ex is extracted along horizontal Line 1 and Line 2. Similarly, the distribution of ey is extracted along Line 3. However, the shear strain cxy is only extracted for points A and B. These points are chosen because they are located near the interface of the solder and copper plate on the solder side. This process is repeated for U and V fringe patterns at specific times, thus producing the strain evolution with time. Strain fields are extracted from the Moire fringes recorded after 1 h:40 min of current stressing. The resolution of fringes in copper precludes measurement of that U field, therefore, only the ex field of the solder joint is extracted as shown in Fig. 13. The distribution of ey and cxy are plotted in Figs. 14 and 15 including both the copper plate and solder joint. The upper and lower copper plates are 0.4 mm thick. The ex is in expansion everywhere on the solder joint, while ey is mostly in contraction within the joint and in expansion within the two copper plates. The largest shear strain is observed near the center of the solder joint.
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
2025
1.6 0.0022 1.4
0.0019
1.2
0.0017
1.0
0.0014
0.8
0.0011
0.6
8.3E-4
0.4
5.5E-4
0.2
2.8E-4 0
0.0 0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Fig. 13. Normal strain ex distribution after 1 h:40 min of current stressing.
0.0030 0.0025
2.0
0.0020 0.0015 1.5
1E-3 5E-4 0
1.0
-5E-4 -1E-3 0.5
-0.0015 -0.0020
0.0 0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Fig. 14. Normal strain ey distribution after 1 h:40 min of current stressing.
to Moire results. A uniform temperature change from 22 to 37 °C is applied to the solder joint and vicinity. A higher temperature change from 22 to 57 °C is applied to both ends of the copper plate to account for local heating due to the high contact resistance between the copper plates and the fixture. The model geometry is shown in Fig. 16. Since the thickness of the sectioned solder joint is less than 2 mm compared to 20 mm of the width of the copper plates, a plane stress element is used to simulate the solder joint and plane strain element to simulate the copper plate. This simulation is more refined than a former simulation [8] where both solder joint and copper plates are simulated with plane strain elements. Both ends of the copper plates are fixed to simulate the boundary conditions of the module in the
1E-3 5E-4 2.0
0 -5E-4
1.5
-1E-3 -0.0015 -0.0020
1.0
-0.0025 -0.0030 0.5
-0.0035 -0.0040
0.0 0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Fig. 15. Shear strain cxy distribution after 1 h:40 min of current stressing.
5. FEM thermal stressing simulation To assess the contribution of thermal stresses to the strains in the solder joint, a FEM simulation is performed with ANSYS finite element code and compared
Fig. 16. Geometry and boundary conditions used in FEM simulation.
Table 1 Material parameters Solder [9]
Copper
E ¼ 62 0:06T GPa G ¼ 24:3 0:029T GPa aT ¼ 21:6 106 K1
E ¼ 117:2 GPa m ¼ 0:33 aT ¼ 16:6 106 K1
2026
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
Fig. 17. Simulated ex distribution.
Fig. 18. Simulated cxy distribution.
fixture. An elastic material model is used for solder. We do realize that this is a gross assumption. Yet, our aim here is to get a first order estimate of strains due to thermal loading. Work with more sophisticated solder model is underway. The material parameters for both solder and copper are shown in Table 1. The simulated distributions of strains are shown in Figs. 17 and 18.
6. Experimental and simulation comparison Moire fringes give the information of relative in-plane displacement. The Moire fringes in the U and V fields represent isopleths of relative horizontal and vertical displacement, respectively. Since the strain fields are the gradient fields of the displacement fields, numerical differentiation is needed to extract strain from experimental fringe measurements. A natural cubic-spline in-
terpolation method is used to improve the results of this numerical process. This numerical differentiation process generally gives good accuracy of extracted strain from fringe measurement when the data points for interpolation are sufficient. When very few data points (indicating small variation of displacements in x or y direction, but not necessary small strains) are available for interpolation, a loss of accuracy of an order of magnitude in numerical differentiation is possible [1]. When calculating the shear strain from the Moire fringes as shown in Figs. 8 and 9, e.g. the gradient of vertical displacement along the x (horizontal) direction or horizontal displacement along the y (vertical) direction, the differentiation accuracy is good due to the high data density. Figs. 19 and 20 give the distribution of shear strain cxy along Line 3 (Fig. 12) for both experimental (after 1 h:40 min) and simulated results due to thermal stressing, and are in very close agreement.
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
500
1000
1500
2000
2500
2027
0. 00010
-0.0005 0. 00005
-0.001 0. 00000
-0.0015
0
500
1000
1500
2000
2500
-0.00005
-0.002
Fig. 19. Shear strain cxy distribution along Line 3 from Moire experiment (horizontal coordinate in nm).
-0.00010
Fig. 22. Normal strain ey distribution along Line 3 from FEM simulation (horizontal coordinate in nm).
0. 0005 0. 0000 -0 .0005
0
500
1000
1500
2000
2500
-0 .0010 -0 .0015 -0 .0020 -0 .0025
Fig. 20. Shear strain cxy distribution along Line 3 from FEM simulation (horizontal coordinate in nm).
Figs. 21 and 22 give the distributions of strain ey along Line 3 (Fig. 12) for both experimental (after 1 h:40 min) and simulated results. Both the refined FEM simulation and experimental results indicate strain ey is in expansion in the regions of the copper plates and in compression in the solder region. However, the calculated strain ey from experimental fringes is over 10 times greater than predicted by simulation. Such discrepancy is attributed to the numerical differentiation from the Moire fringes. Since ey is the gradient (ey ¼ oV =oy) along y and there are few Moire fringes, poor numerical differentiation occurs. Though the calculated value of ey is not highly resolved, the tendency of Moire fringes to change direction is observable, and indicates a change of sign for the strain along the vertical line. This difference could also be due
to viscoplastic response of solder alloys and small contribution of current stressing. The results of this testing and refined FEM simulation strongly suggest that the strains observed in the experiment are largely due to thermal stressing and only minimally due to current stressing at the current density used thus far. This is because that the Moire interferometry experiment results closely resemble the finite element simulation where the only loading is temperature change. If the effect of electric current is dominant, the experiment results should deviate greatly from the simulation where only thermal stressing is considered. The agreement between the experiment and FEM simulation supports the use of Moire interferometry as a reliable approach in measuring in situ displacement of BGA solder joints under current stressing.
7. In situ evolution of strains Figs. 23–25 give the in situ evolution of strains, ex , ey , and cxy , with time during and after current stressing. Note that the current was turned off after 4 h (14,400 s). The normal strain ex increased immediately in extension after the current was applied and varied slightly during the course of stressing. When turning the current off, ex ε x at point A
0.0008
εx at point B
Strain εx
0.0006
0.001 0.0005
0.0004 0.0002 0.0000
500
1000
1500
2000
2500
-0.0005 -0.001
-0.0002 -0.0004 0
20000
40000
60000
Time (Second ) Fig. 21. Normal strain ey distribution along Line 3 from Moire experiment (horizontal coordinate in nm).
Fig. 23. Normal strain ex at point A and B.
80000
2028
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
0.0004
εy at point A εy at point B
0.0002 0.0000
Strain ε y
-0.0002 -0.0004 -0.0006 -0.0008 -0.0010 -0.0012
0
20000
40000
60000
80000
Time (Second)
Fig. 24. Normal strain ey at point A and B.
0.0000 -0.0004
decreased abruptly. At point A it dropped quickly to near its initial value before stressing. It decreased to zero at point B, where it initially was in compression. The strain ex remained nearly unchanged after cessation of current. The evolution of normal strain ey at points A and B, as recorded, shows that ey changed direction several times during current stressing. As shown in the previous section, the copper plates are in expansion while the solder joint is in contraction during stressing. The value of shear strain cxy increased gradually after the current is applied and reached a maximum in about 2 h. Immediately after current removal, the value of shear strain decreased sharply, but not to its initial value, and then remained almost unchanged. The value of shear strain is measured to be much larger (in the magnitude of 103 ) than the normal strain (in the magnitude of 104 ). In addition, the residual strains indicate permanent plastic deformation during the stressing process and are in accordance with solder being a viscoplastic material even at room temperature.
Strain γ xy
-0.0008 -0.0012
8. Experiment of lead-free solder joint with improved thermal management
-0.0016 γxy at point A
-0.0020
γxy at point B
-0.0024 -0.0028
0
20000
40000
60000
Time (Second)
Fig. 25. Shear strain cxy at point A and B.
80000
In a second Moire interferometry experiment, a wellcontrolled temperature was used during current stressing and Joule heating restricted by improving the electrical contact between the test module and fixture. This is done by applying greater clamping force so that the close electric contact between the test vehicle and fixture is
Fig. 26. U field fringe evolution (a) initial; (b) 97 h of stressing; (c) 170 h of stressing.
Fig. 27. V field fringe evolution (a) initial; (b) 97 h of stressing; (c) 170 h of stressing.
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
maintained. Lead-free solder (SnAg4Cu0.5) was used to fabricate the test module. The height of the solder joint was 1.7 mm with an approximate diameter of 1.35 mm. The joint was sectioned and polished to give an average thickness of about 0.4 mm. In the experiment, 30 A of current was applied and a current density above 5000 A/cm2 was achieved by further reducing the cross section. The width of the solder joint was reduced to 1.2 mm and the thickness is polished down to 0.5 mm while the height remains unchanged. The temperature was kept almost constant at 27 °C. The Moire fringe evolution is shown in Figs. 26 and 27 for U and V fields, respectively. Very few U and V field fringes developed after the solder joint was stressed for 170 h. This result indicates there was little shear deformation, contrary to the previous experiment. Since near constant temperature was maintained, it is appropriate to reason that current stressing caused this vertical displacement. This could explain the discrepancy between the Moire interferometry measurements and FE simulation results for ey in the previous section. Shear deformation was dominant under thermal stressing in the previous experiment (and was verified by the FE simulation). A third experiment was performed and yield results similar to the second experiment. More Moire interferometry experiments will be performed for solder joints under different current densities. Based on these experimental results, a constitutive model to predict the deformation of solder under current stressing will be proposed and calibrated for use in reliability prediction. Those results will be published in the future.
9. Conclusions In this work, strain evolution in a BGA solder joint under electrical current stressing was monitored in realtime with Moire interferometry as the first step in building a constitutive model for solder material under current stressing. This is the first study reported in the literature where strain field under current stressing is monitored by Moire interferometry. A FEM simulation of thermal stressing was performed which exhibited good agreement with Moire interferometry results. Moire interferometry is a reliable technique to measure in situ displacement of BGA solder joints under current stressing. The results of this testing and refined FEM
2029
simulation strongly suggest that the strains observed in the first experiment are largely due to thermal stressing and only minimally due to current stressing at low current density. A second experiment, where joule heating was well controlled, showed that high electric current stressing caused deformation field quite different than thermal stressing.
Acknowledgements This project is provided in part by a grant from the Sydney J. Stein Educational Foundation of the International Microelectronics and Packaging Society and by the Office of Naval Research Advanced Electrical Power Systems under the supervision of Dr. Terry Ericsen.
References [1] Post D, Han B, Ifju P. High sensitivity Moire. New York: Springer; 1994. [2] Lee TY, Tu KN, Kuo SM, Frear DR. Electromigration of eutectic SnPb solder interconnects for flip chip technology. J Appl Phys 2001;89(6):3189–94. [3] Brandenburg S, Yeh S. Electromigration studies of flip chip bump solder joints. Surface Mount International Conference Proceedings, 1998. [4] Ye H, Basaran C, Hopkins D. Experiment study on reliability of solder joints under electrical stressing––nano-indentation, atomic flux measurement. Proceedings of 2002 International Conference on Advanced Packaging and Systems, Reno, Nevada, 3-10-2002. [5] Lloyd JR. Electromigration in integrated circuit conductors. J Phys D-Appl Phys 1999;32(17):R109–18. [6] Zhao Y, Basaran C, Cartwright A, Dishong T. An experimental observation of thermomechanical behavior of BGA solder joints by Moire interferometry. J Mech Behav Mater 1999;10(3):135–46. [7] Zhao Y, Basaran C, Cartwright A, Dishongh T. Thermomechanical behavior of micron scale solder joints under dynamic loads. Mech Mater 2000;32(3):161–73. [8] Ye H, Hopkins D, Basaran C. Measurement and effects of high electrical current stress in solder joints. Proceedings of the 35th International Symposium on Microelectronics, Denver, Colorado, September 2002. p. 427–32. [9] Chandaroy, Rumpa. Damage mechanics of microelectronic packaging under combined dynamic and thermal loading (dynamic loading). Thesis (PhD), State University of New York at Buffalo, 1998. 279p.
Measurement of high electrical current density effects in solder joints Hua Ye, Douglas C. Hopkins, Cemal Basaran
*
UB Electronic Packaging Laboratory, 102 Ketter Hall, SUNY at Buffalo, Buffalo, NY 14260, USA Received 29 January 2003; received in revised form 10 April 2003
Abstract Measuring mechanical implications of high current densities in microelectronic packaging interconnects has always been a challenging goal. Due to small interconnect size this task has typically been accomplished by measuring the change in electrical resistance of the joint. This measurement parameter is global and does not give local mechanical state information. Also, understanding strain evolution in the solder over time is an important step toward developing a damage mechanics model. The real-time, full-field, strain displacement in a eutectic Sn/Pb solder joint during electrical current stressing was measured with Moire interferometry (Post et al., High sensitivity Moire, Springer, New York, 1994) under in situ conditions. A finite element model simulation for thermal stressing was performed and compared with measured strain. The initial results show that the measured strain was largely due to thermal stressing versus the current density of 1.8 102 A/cm2 . A second Moire interferometry experiment with thermal control distinguishes deformation of solder joint due to pure current stressing above 5000 A/cm2 . Ó 2003 Elsevier Ltd. All rights reserved.
1. Introduction A perceived physical limit to increasing current density in both microelectronic and power electronic packaging is electromigration. Most research has focused on the thin, metal lines of ICs, and little on present day solder alloy interconnects [2–5] as used in higher levels of packaging. The development of a constitutive model is sought to predict the stress–strain field in a solder joint undergoing high-density electrical current flow. As a first step, the strain field in a BGA solder ball was measured using Moire interferometry. Strain evolution was derived with a natural cubic-spline interpolation method. This paper reports the preliminary findings. Previous work reported electromigration damage in solder joints undergoing current stressing of about 1.3 104 A/cm2 at ambient room temperature [2–4].
However, initial results presented here indicate that reliability may be greatly diminished at current densities much less than that due to complications from thermal stressing. The size of the tested solder bump was 3.8 mm in diameter and 1.70 mm in height. The maximum current applied to the solder bump was 10 A yielding a computed density of 175 A/cm2 . This density is much lower than the one to trigger electromigration as reported in the literature [2–4]. A finite element model (FEM) simulation of thermal stressing was performed and validated by experimental results, indicating that the strain under such current density was mostly due to thermal stressing. A second Moire interferometry experiment with good thermal control differentiates the deformation of the solder joint due to pure current stressing at a density above 5000 A/cm2 . 2. Experimental set-up
*
Corresponding author. Tel.: +1-716-645-2114x2429; fax: +1-716-645-3733. E-mail address: [email protected]ffalo.edu (C. Basaran).
Since this research solely focuses on the reliability of solder joints, two copper plates form the test vehicle
0026-2714/$ - see front matter Ó 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S0026-2714(03)00131-8
2022
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
a) Pin Test vehicle
Pin
Pin
Steel Steel
Dielectric polymer base plate b)
Fig. 2. The schematic diagram of the test vehicle fixture (a) plane view (b) the front view.
a) Cu plate b)
Cu plate
Fig. 1. The schematic diagram of the test vehicle (a) plane view and (b) side view (c) test vehicle after cross-sectioned with a diamond-wheel saw.
module instead of an FR4 carrier substrate and silicon die. The advantage of the copper plates as substrates is that the electrical connection can be made very simply and can carry a fairly large electrical current (a problem plaguing the thin-film metal lines in ICs [5]). A schematic diagram of the test vehicle is shown in Fig. 1a and b. A thin silicon dioxide layer was used as the solder mask on the copper plate. A continuous silicon dioxide layer was applied to the plate, patterned and etched. The photolithography process was done in class 1000 clean room. A hot plate was used to reflow the eutectic Pb37/ Sn63 alloy solder to form interconnect spheres. Spacers were used during reflow. Some oxidation of copper was observed and monitored. The test vehicle was first sliced through the center of the solder bump with a high precision diamond-wheel saw, further sliced to the shape that could be fit into the fixture (Fig. 1c), and finally polished with 1200 grit abrasive paper. An optical diffraction grating was replicated on the sectioned surface.
The fixture was composed of a dielectric polymer base plate with two steel plates on each end. The steel plates were affixed by screws and positioned via pins. The test vehicle was then clamped between the two steel plates and positioned by the pins. Electrical-test connections were made through the steel plates. The schematic diagram of the test vehicle fixture is shown is Fig. 2. A schematic diagram of the experimental test set-up is shown in Fig. 3. In this set-up, the applied current was computer controlled. The current was monitored through a Kelvin connection to a serial resistor. An infrared temperature sensor monitored the solder bump. All the controls, measurements and data logging were implemented in a single Labviewâ user subroutine. The monitored temperature was later used in the FEM thermal stress analysis. The module was mounted onto the Moire interferometry table and all wiring connected as shown. The optical set-up was then tuned to get a null field. Since Moire interferometry measurement relies on an initial reference field, the optical set-up was carefully protected from any disturbance during testing. The Moire interferometry technique is explained extensively by Post et al. [1]. The development of the imaging systems used for submicron displacement measurement using Moire interferometry technique
H-P6260B Power supply
Shunt resistor
H-P59501B DAC Microvolt meter
Infared thermal sensor Computer Microvolt meter Test vehicle fixture
Fig. 3. Schematic diagram of the test set-up.
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
3. Experimental results The test module was first energized with 5.8 A of DC current for 20 min. The current was then raised to 9.8 A, as shown in Fig. 4. The total current stressing time was 4 h. The Moire fringes of both U and V fields were recorded in real-time. After the current was turned off, deformation fringes were recorded for another 17 h. Fig. 5 shows the measured solder bump temperature time history. The solder bump temperature increased from room temperature to 29 °C immediately due to joule heating and remained there until the current was increased to 9.8 A. The temperature abruptly increased to 34 °C and gradually rose to 36.5 °C. After the current was turned off, the temperature abruptly decreased to 29 °C and, then, progressively decreased to room temperature in an hour.
38
Temperature on Test Sample
36 34
o
Temperature ( C )
has been described in detail in Zhao et al. [6,7]. The major advantage of Moire interferometry is its high sensitivity, high resolution, and the whole field view of deformation distribution of the specimen surface. Briefly, the optical diffraction grating is replicated on the specimen surface. The specimen grating diffracts the incident two coherent laser beams with certain incident angle, and in the direction normal to the specimen surface, two strong diffracted beams are obtained. When the specimen surface deforms, the optical diffraction grating deforms with the specimen, and the two diffracted beams in the normal direction generate feature interferometry pattern that represents the in-plane displacement distribution. This scheme applies to both horizontal and vertical direction, so that deformation in the two perpendicular directions can be obtained. The feature fringe pattern generated by the vertical two beams represents the vertical deformation field, and the fringe pattern generated by the horizontal two beams represents the horizontal deformation field [7].
2023
32 30 28 26 24 22 20 0
5000
10000
15000
20000
The temperature, measured with the infrared thermal sensor (with a spotsize of 10 mm in diameter), was the average temperature of the central area on the top surface of the test module. A fine thermal couple was used to pinpoint the temperature distribution on the module during current stressing. The temperature distribution was not uniform. An appreciable heat source was found at the interface of the copper plates and clamping fixture, and needed to be included in the analysis. When the temperature atop the copper plate was 36.5 °C, the temperature near the ends of copper plates increased to 57 °C. Figs. 6 and 7 are the initial Moire fringes of the solder bump before current stressing for U and V fields, respectively. The initial fringes and non-uniformity indicates initial strains during the sample mounting process.
Current (Amps)
8
Fig. 6. Initial U field.
4 2 0 -2 0
5000
10000
15000
20000
25000
30000
Time (second) Fig. 4. Profile of the current applied.
30000
Fig. 5. Measured temperature time history.
10
6
25000
Time (Second)
Fig. 7. Initial V field.
2024
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
Fig. 8. U field (after 02 h:22 min:0 s of stressing). Fig. 12. Lines of interest and position of points A and B.
direction of lower-order fringes [1]. The direction of fringe order, Ny was determined similarly by þy direction perturbation and observing the direction for the V field movement. Once the fringe orders Nx and Ny are determined, the strains at any point are given by, Fig. 9. V field (after 02 h:22 min:15 s of stressing).
oU 1 oNx ¼ ox f ox oV 1 oNy ¼ ey ¼ oy f oy oU oV 1 oNx oNy þ ¼ cxy ¼ þ oy ox f oy ox
ex ¼
Fig. 10. U field (2 h:0 min:0 s after current turned off).
Fig. 11. V field (2 h:0 min:15 s after current turned off).
Figs. 8–11 show the Moire fringe evolution with time during and after current stressing.
4. Strain analysis To extract the strains from the Moire fringes, determination of the direction of the fringe order was needed and performed as follows. During the experiment, the test module was perturbed in the þx direction (coordinates as shown in Fig. 12) and fringes moved toward the
where Nx and Ny are fringe orders in U and V fields, respectively, and f ¼ 2fx , fx is the frequency of diffraction grating (1200 lines/mm). With this frequency, the resolution of displacement is 0.417 lm. Instead of using linear interpolation, natural cubic-spline interpolation is used to approximate the derivative of fringe order more accurately. Three lines of interest are shown in Fig. 12. The distribution of ex is extracted along horizontal Line 1 and Line 2. Similarly, the distribution of ey is extracted along Line 3. However, the shear strain cxy is only extracted for points A and B. These points are chosen because they are located near the interface of the solder and copper plate on the solder side. This process is repeated for U and V fringe patterns at specific times, thus producing the strain evolution with time. Strain fields are extracted from the Moire fringes recorded after 1 h:40 min of current stressing. The resolution of fringes in copper precludes measurement of that U field, therefore, only the ex field of the solder joint is extracted as shown in Fig. 13. The distribution of ey and cxy are plotted in Figs. 14 and 15 including both the copper plate and solder joint. The upper and lower copper plates are 0.4 mm thick. The ex is in expansion everywhere on the solder joint, while ey is mostly in contraction within the joint and in expansion within the two copper plates. The largest shear strain is observed near the center of the solder joint.
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
2025
1.6 0.0022 1.4
0.0019
1.2
0.0017
1.0
0.0014
0.8
0.0011
0.6
8.3E-4
0.4
5.5E-4
0.2
2.8E-4 0
0.0 0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Fig. 13. Normal strain ex distribution after 1 h:40 min of current stressing.
0.0030 0.0025
2.0
0.0020 0.0015 1.5
1E-3 5E-4 0
1.0
-5E-4 -1E-3 0.5
-0.0015 -0.0020
0.0 0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Fig. 14. Normal strain ey distribution after 1 h:40 min of current stressing.
to Moire results. A uniform temperature change from 22 to 37 °C is applied to the solder joint and vicinity. A higher temperature change from 22 to 57 °C is applied to both ends of the copper plate to account for local heating due to the high contact resistance between the copper plates and the fixture. The model geometry is shown in Fig. 16. Since the thickness of the sectioned solder joint is less than 2 mm compared to 20 mm of the width of the copper plates, a plane stress element is used to simulate the solder joint and plane strain element to simulate the copper plate. This simulation is more refined than a former simulation [8] where both solder joint and copper plates are simulated with plane strain elements. Both ends of the copper plates are fixed to simulate the boundary conditions of the module in the
1E-3 5E-4 2.0
0 -5E-4
1.5
-1E-3 -0.0015 -0.0020
1.0
-0.0025 -0.0030 0.5
-0.0035 -0.0040
0.0 0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Fig. 15. Shear strain cxy distribution after 1 h:40 min of current stressing.
5. FEM thermal stressing simulation To assess the contribution of thermal stresses to the strains in the solder joint, a FEM simulation is performed with ANSYS finite element code and compared
Fig. 16. Geometry and boundary conditions used in FEM simulation.
Table 1 Material parameters Solder [9]
Copper
E ¼ 62 0:06T GPa G ¼ 24:3 0:029T GPa aT ¼ 21:6 106 K1
E ¼ 117:2 GPa m ¼ 0:33 aT ¼ 16:6 106 K1
2026
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
Fig. 17. Simulated ex distribution.
Fig. 18. Simulated cxy distribution.
fixture. An elastic material model is used for solder. We do realize that this is a gross assumption. Yet, our aim here is to get a first order estimate of strains due to thermal loading. Work with more sophisticated solder model is underway. The material parameters for both solder and copper are shown in Table 1. The simulated distributions of strains are shown in Figs. 17 and 18.
6. Experimental and simulation comparison Moire fringes give the information of relative in-plane displacement. The Moire fringes in the U and V fields represent isopleths of relative horizontal and vertical displacement, respectively. Since the strain fields are the gradient fields of the displacement fields, numerical differentiation is needed to extract strain from experimental fringe measurements. A natural cubic-spline in-
terpolation method is used to improve the results of this numerical process. This numerical differentiation process generally gives good accuracy of extracted strain from fringe measurement when the data points for interpolation are sufficient. When very few data points (indicating small variation of displacements in x or y direction, but not necessary small strains) are available for interpolation, a loss of accuracy of an order of magnitude in numerical differentiation is possible [1]. When calculating the shear strain from the Moire fringes as shown in Figs. 8 and 9, e.g. the gradient of vertical displacement along the x (horizontal) direction or horizontal displacement along the y (vertical) direction, the differentiation accuracy is good due to the high data density. Figs. 19 and 20 give the distribution of shear strain cxy along Line 3 (Fig. 12) for both experimental (after 1 h:40 min) and simulated results due to thermal stressing, and are in very close agreement.
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
500
1000
1500
2000
2500
2027
0. 00010
-0.0005 0. 00005
-0.001 0. 00000
-0.0015
0
500
1000
1500
2000
2500
-0.00005
-0.002
Fig. 19. Shear strain cxy distribution along Line 3 from Moire experiment (horizontal coordinate in nm).
-0.00010
Fig. 22. Normal strain ey distribution along Line 3 from FEM simulation (horizontal coordinate in nm).
0. 0005 0. 0000 -0 .0005
0
500
1000
1500
2000
2500
-0 .0010 -0 .0015 -0 .0020 -0 .0025
Fig. 20. Shear strain cxy distribution along Line 3 from FEM simulation (horizontal coordinate in nm).
Figs. 21 and 22 give the distributions of strain ey along Line 3 (Fig. 12) for both experimental (after 1 h:40 min) and simulated results. Both the refined FEM simulation and experimental results indicate strain ey is in expansion in the regions of the copper plates and in compression in the solder region. However, the calculated strain ey from experimental fringes is over 10 times greater than predicted by simulation. Such discrepancy is attributed to the numerical differentiation from the Moire fringes. Since ey is the gradient (ey ¼ oV =oy) along y and there are few Moire fringes, poor numerical differentiation occurs. Though the calculated value of ey is not highly resolved, the tendency of Moire fringes to change direction is observable, and indicates a change of sign for the strain along the vertical line. This difference could also be due
to viscoplastic response of solder alloys and small contribution of current stressing. The results of this testing and refined FEM simulation strongly suggest that the strains observed in the experiment are largely due to thermal stressing and only minimally due to current stressing at the current density used thus far. This is because that the Moire interferometry experiment results closely resemble the finite element simulation where the only loading is temperature change. If the effect of electric current is dominant, the experiment results should deviate greatly from the simulation where only thermal stressing is considered. The agreement between the experiment and FEM simulation supports the use of Moire interferometry as a reliable approach in measuring in situ displacement of BGA solder joints under current stressing.
7. In situ evolution of strains Figs. 23–25 give the in situ evolution of strains, ex , ey , and cxy , with time during and after current stressing. Note that the current was turned off after 4 h (14,400 s). The normal strain ex increased immediately in extension after the current was applied and varied slightly during the course of stressing. When turning the current off, ex ε x at point A
0.0008
εx at point B
Strain εx
0.0006
0.001 0.0005
0.0004 0.0002 0.0000
500
1000
1500
2000
2500
-0.0005 -0.001
-0.0002 -0.0004 0
20000
40000
60000
Time (Second ) Fig. 21. Normal strain ey distribution along Line 3 from Moire experiment (horizontal coordinate in nm).
Fig. 23. Normal strain ex at point A and B.
80000
2028
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
0.0004
εy at point A εy at point B
0.0002 0.0000
Strain ε y
-0.0002 -0.0004 -0.0006 -0.0008 -0.0010 -0.0012
0
20000
40000
60000
80000
Time (Second)
Fig. 24. Normal strain ey at point A and B.
0.0000 -0.0004
decreased abruptly. At point A it dropped quickly to near its initial value before stressing. It decreased to zero at point B, where it initially was in compression. The strain ex remained nearly unchanged after cessation of current. The evolution of normal strain ey at points A and B, as recorded, shows that ey changed direction several times during current stressing. As shown in the previous section, the copper plates are in expansion while the solder joint is in contraction during stressing. The value of shear strain cxy increased gradually after the current is applied and reached a maximum in about 2 h. Immediately after current removal, the value of shear strain decreased sharply, but not to its initial value, and then remained almost unchanged. The value of shear strain is measured to be much larger (in the magnitude of 103 ) than the normal strain (in the magnitude of 104 ). In addition, the residual strains indicate permanent plastic deformation during the stressing process and are in accordance with solder being a viscoplastic material even at room temperature.
Strain γ xy
-0.0008 -0.0012
8. Experiment of lead-free solder joint with improved thermal management
-0.0016 γxy at point A
-0.0020
γxy at point B
-0.0024 -0.0028
0
20000
40000
60000
Time (Second)
Fig. 25. Shear strain cxy at point A and B.
80000
In a second Moire interferometry experiment, a wellcontrolled temperature was used during current stressing and Joule heating restricted by improving the electrical contact between the test module and fixture. This is done by applying greater clamping force so that the close electric contact between the test vehicle and fixture is
Fig. 26. U field fringe evolution (a) initial; (b) 97 h of stressing; (c) 170 h of stressing.
Fig. 27. V field fringe evolution (a) initial; (b) 97 h of stressing; (c) 170 h of stressing.
H. Ye et al. / Microelectronics Reliability 43 (2003) 2021–2029
maintained. Lead-free solder (SnAg4Cu0.5) was used to fabricate the test module. The height of the solder joint was 1.7 mm with an approximate diameter of 1.35 mm. The joint was sectioned and polished to give an average thickness of about 0.4 mm. In the experiment, 30 A of current was applied and a current density above 5000 A/cm2 was achieved by further reducing the cross section. The width of the solder joint was reduced to 1.2 mm and the thickness is polished down to 0.5 mm while the height remains unchanged. The temperature was kept almost constant at 27 °C. The Moire fringe evolution is shown in Figs. 26 and 27 for U and V fields, respectively. Very few U and V field fringes developed after the solder joint was stressed for 170 h. This result indicates there was little shear deformation, contrary to the previous experiment. Since near constant temperature was maintained, it is appropriate to reason that current stressing caused this vertical displacement. This could explain the discrepancy between the Moire interferometry measurements and FE simulation results for ey in the previous section. Shear deformation was dominant under thermal stressing in the previous experiment (and was verified by the FE simulation). A third experiment was performed and yield results similar to the second experiment. More Moire interferometry experiments will be performed for solder joints under different current densities. Based on these experimental results, a constitutive model to predict the deformation of solder under current stressing will be proposed and calibrated for use in reliability prediction. Those results will be published in the future.
9. Conclusions In this work, strain evolution in a BGA solder joint under electrical current stressing was monitored in realtime with Moire interferometry as the first step in building a constitutive model for solder material under current stressing. This is the first study reported in the literature where strain field under current stressing is monitored by Moire interferometry. A FEM simulation of thermal stressing was performed which exhibited good agreement with Moire interferometry results. Moire interferometry is a reliable technique to measure in situ displacement of BGA solder joints under current stressing. The results of this testing and refined FEM
2029
simulation strongly suggest that the strains observed in the first experiment are largely due to thermal stressing and only minimally due to current stressing at low current density. A second experiment, where joule heating was well controlled, showed that high electric current stressing caused deformation field quite different than thermal stressing.
Acknowledgements This project is provided in part by a grant from the Sydney J. Stein Educational Foundation of the International Microelectronics and Packaging Society and by the Office of Naval Research Advanced Electrical Power Systems under the supervision of Dr. Terry Ericsen.
References [1] Post D, Han B, Ifju P. High sensitivity Moire. New York: Springer; 1994. [2] Lee TY, Tu KN, Kuo SM, Frear DR. Electromigration of eutectic SnPb solder interconnects for flip chip technology. J Appl Phys 2001;89(6):3189–94. [3] Brandenburg S, Yeh S. Electromigration studies of flip chip bump solder joints. Surface Mount International Conference Proceedings, 1998. [4] Ye H, Basaran C, Hopkins D. Experiment study on reliability of solder joints under electrical stressing––nano-indentation, atomic flux measurement. Proceedings of 2002 International Conference on Advanced Packaging and Systems, Reno, Nevada, 3-10-2002. [5] Lloyd JR. Electromigration in integrated circuit conductors. J Phys D-Appl Phys 1999;32(17):R109–18. [6] Zhao Y, Basaran C, Cartwright A, Dishong T. An experimental observation of thermomechanical behavior of BGA solder joints by Moire interferometry. J Mech Behav Mater 1999;10(3):135–46. [7] Zhao Y, Basaran C, Cartwright A, Dishongh T. Thermomechanical behavior of micron scale solder joints under dynamic loads. Mech Mater 2000;32(3):161–73. [8] Ye H, Hopkins D, Basaran C. Measurement and effects of high electrical current stress in solder joints. Proceedings of the 35th International Symposium on Microelectronics, Denver, Colorado, September 2002. p. 427–32. [9] Chandaroy, Rumpa. Damage mechanics of microelectronic packaging under combined dynamic and thermal loading (dynamic loading). Thesis (PhD), State University of New York at Buffalo, 1998. 279p.
