Study of Electromigration of flip-chip solder joints using Kelvin probes

Transcription

1 Study of Electromigration of flip-chip solder joints using Kelvin probes Y. W. Chang and Chih Chen National Chiao Tung University, Department of Material Science & Engineering, Hsin-chu 30010, Taiwan, ROC Chih Chen, TEL: (O) ; Fax: Abstract. Kelvin structures with various layouts are employed to detect the change in bump resistance due to void formation and propagation during electromigration in flip-chip solder joints. It is found that the Kelvin structure can sense the highest voltage drop when its voltage probes are placed at the current entrance in the chip side into the bump, and thus it is the most sensitive design to monitor void formation and propagation. When the bump resistance increases 20% of its initial value, the depletion percentage of contact opening ranges from 21.0% to 65.0%, depending on the position of the probe. Keywords: Enter Keywords here. PACS: Replace this text with PACS numbers; choose from this list: INTRODUCTION To meet the relentless drive for miniaturization of portable devices, flip chip technology has been adopted for high-density packaging due to its excellent electrical characteristic and superior heat dissipation capability. As the required performance in microelectronics devices becomes higher, the operation current continues to increase, which causes the current density carried by each solder bump rises dramatically. Electromigration has become an inevitably critical reliability issue in flip-chip solder joints. Two layouts have been used to investigate the electromigration behaviour in flip-chip solder joints, including Daisy chains 3 and Kelvin structures 2,4. Daisy chains structure connects many bumps using chip and substrate metallization alternately to connect adjacent bumps. It may provide statistical results on electromigration failure, since a large number of bumps can be stressed in the same time. Yet, it is not sensitive at all to detect subtle resistance changes due to void formation or microstructure changes in solder bumps. This is because that the resistance of the chip and substrate metallization is much higher than that of the solder bumps. Kelvin structures constitute four probes, in which two of them are employed to measure the voltage drop across a solder bump. Our previous results show that the resistance change due to void formation in the initial stage is less than 1 m. 4-5 In contrast, the typical resistance for a Daisy chains structure is of the order of few Ohms and noise from the temperature

2 fluctuation may overwhelm the resistance increase attributed to the void formation and propagation. Thus, it lacks sensitivity on the void formation or subtle microstructure changes. e - (a) Daisy chain structure V I e - (b) Kelvin structure FIGURE 1. (a) Daisy chain structure (b) Kelvin structure Therefore, Kelvin structure appears to be a very promising approach to study electromigration. They have been implemented in solder joints to monitor the electromigration behaviour. 2,4,6 In addition, geometrical effect of bump resistance measured by Kelvin probes has been investigated experimentally and by simulation. 7 However, the change in bump resistance due to void propagation is not clear using different layouts of Kelvin structure. In particular, the two probes sensing the voltage drop can locate at various positions of the periphery of the bump, and thus the measured resistance changes due to void formation may vary with the design of the probes. Such an issue has not been addressed so far. EXPERIMENTAL PROCEDURE To examine how the location of Kelvin probes affects the measured resistance of solder bumps, four layouts of Kelvin structure were examined. Figure 2(a) shows one of the layouts, which matched the real solder joints with four-probe Kelvin structure. 7 The Al trace was 1.5 μm thick and 100 μm wide. Eutectic SnPb solder and under-bump-metallization (UBM) of 0.5-μm Ti/5-μm Cu/ 3-μm Ni were adopted. The two probes for applying current and the two probes for sensing the voltage drop are labeled in the figure. The voltage distribution is also shown in the figure with an applied current stress of 0.8 A. To obtain the bump resistance using Kelvin structure,

3 the voltage along part of the peripheries of the circular Al pad and the Cu pad were extracted. (a) Plan view b1 t1 b2 t2 b3 t3 b4 n1 n2 n3 n4 n5 n6 (b) Cross section view b1 t1 b2 t2 b3 t3 b4 FR5 Hot plate n1 n2 n3 n4 n5 n6 I meter Total V meter e - b3 V meter Al 1.5μm Cu 5μm Ni 3μm Ni 3 Sn 4 1μm 140 μm 85 μm 120 μm (c) e-snpb 25μm 280 μm Electroless Ni 5μm Cu 30μm FIGURE 2. (c) Cross section view of a single bump (a) Plan view (b) Cross section view (c) Cross section view of a single bump The bump resistance R bump in this letter is denoted as: R bump V top bottom,max (1) I total V V I total

4 Where V top is the voltage obtained from the top Al pad at a specific angle, V bottom,max is maximum voltage value obtained from the bottom Cu pad, and I total is the total current applied in the simulation model. Since the bump resistance was calculated from the voltage drop divided by the applied current, the measured bump resistance is proportional to the measured voltage drop. 180 A B O A 90 A O B (a) B - (b) FIGURE 3. (a) Voltage distribution in the solder joint with a Kelvin structure. The possible positions for the voltage probe on the chip side are also shown in the figure, including 0, 45, 90, and 180. (b) Cross section view of all different stages Nevertheless, the trend reverses in terms of the percentage in voltage increase. Figure 5(b) shows the percentage increase in the measured voltage. The percentage in bump resistance change, R bump (%), can be expressed as R R bump bump, initial R (%) 100 bump (2) R bump, initial

5 RESULT AND DISCUSSION Figures 5(a) and 5(b) show the voltage distributions along the A-A -A arc in the Al pad and along the B-B -B arc in the Cu pad, respectively. The voltage ranges from -8.0 mv to -0.6 mv along the A-A -A arc of the Al pad; while it varies from mv to mv along the B-B -B arc of the Cu pad. The voltage distributions on the left-hand side of both curves show abrupt changes, which is attributed to the fact that current crowding effect occurs seriously on the left-hand side of the joints. 8 (a)stage 0:R = R (b)stage 1:R = R (c)stage 2:R = R (d)stage 3:R = R (e)stage 4:R = R (f)stage 5:R = R FIGURE (mv) (a) ~ (f) The voltage distribution of six different stages Since the voltage distributions are not uniform on both Al and Cu pads, the voltage drop detected by Kelvin probes may vary with difference probe layouts, causing the measured bump resistance to differ with different probe layouts. Furthermore, the voltage variation on the Al pad may be over 3 mv at different positions; whereas it is only less than 0.3 mv on the Cu pad, which means that the position of the voltage probe on the Al pad plays a crucial role on the measured bump resistance. Therefore,

6 only the position of the voltage probe on the Al pad was changed in this study. The position of the voltage probe on the Cu pad was fixed, as shown in Figure 3(a). (a) A A A B B B (b) FIGURE 5. (a) The voltage profile along the A-A -A arc in the chip side. (b) The voltage profile along the B-B -B arc in the substrate side. To investigate how the void propagation influences the bump resistance by simulation, a series of models with various sizes of voids were constructed. Figures 3(b) illustrate the cross-sectional view of the solder joints for five of the constructed 29 models. Pancake-type voids were assumed to propagate along the interface between Ni 3 Sn 4 intermetallic compound (IMC) and eutectic SnPb solder. The depletion percentage of the UBM opening is 0%, 6.1%, 20.0%, 50.0%, and 75.0%, respectively, for the five models. As voids form and propagate, more current was forced to drift longer to the right side along the Al pad and the UBM layer, causing the voltage to redistribute. It is found that different layouts of Kelvin probes detect different resistance increases even for the same void size. Figure 6(a) and 6(b) depicts the measured voltage drop as a function of depletion percentage of the UBM opening when the voltage probe was placed at 0, 45, 90, and 180 positions in Figure 2(a). The mean voltage at each position was obtained by averaging its neighboring values within 10 m. The results indicate that the Kelvin structure detects a higher voltage increase when the voltage probe was placed at 0, which was the entrance point of electron flow. For example, when the void depleted 50% of the UBM opening, the voltage increase detected by the four layouts is 0.97 mv, 0.83mV, 0.60 mv, and 0.38 mv, respectively. It is concluded that the probe would sense higher voltage changes upon placing the voltage probe closer to the current crowding region for the same size of void. This is because higher current passage will render higher voltage drop for a given resistance increase. Since there is a current feeding line at position 0, a voltage probe cannot be fabricated there. It is suggested to add the voltage probe on the current feeding line, slightly away from the bump. Although some of the Al-line resistance would be included for the measurement by this layout, it would be the most sensitive method to detect the resistance change due to void formation.

7 (a) (b) FIGURE 6. (a) The difference of delta voltage (b) Variation of voltage Where R bump,initial is the bump resistance before current stressing. As the voltage probe moved far away from the current crowding region, the percentage increase in bump resistance due to the same void size is higher. This is attributed to the fact that the initial bump resistance is much lower as the probe moves away from the current crowding region. The bump resistances are 9.2, 3.5, 0.5, and 0.4 m for the probes at 0, 45, 90, and 180, respectively. As voids form, although the probe at position 180 would sense smaller voltage drop than that at 0, its percentage increase in bump

8 resistance would become higher than that of 0 because it is normalized to its initial resistance. It is noteworthy that the percentage change in bump resistance rises abruptly at later stages of electromigration using the probes far away from the current entrance point. For example, with the use of the probe at 180, the percentage of bump resistance increases rapidly when the depletion area is larger than half of UBM opening. It can be as large as 900% when the voids deplete about 89% of the UBM opening; whereas only a 100% increase is detected by the probe at 45. This is because the effective conduction area is pushed forward to the right-hand side in Fig. 2, which is located at 180 position in Fig. 1(a). More current was forced to drift longer in the Al pad and in the UBM as the voids grow bigger. Thus, the probe at 180 would sense a dramatic increase in voltage after voids depletes about 60 % of the UBM opening, as shown in Fig. 3(a). Therefore, compared with the probes near 0 position, the probe at 180 position would detect a much higher percentage increase in bump resistance for the same size of voids. The above results may provide a new criterion for electromigration failure in the flip-chip solder joints. In Al and Cu interconnects for microelectronic devices, failure criterion for electromigration is often defined when their resistances reach 1.2 times (20% increase) of their initial values. Figure 6 shows the magnified figure for the initial stage of electromigration up to 25% resistance increase in Figure 6. When the bump resistance increases 20% of its initial value, the depletion percentage of UBM opening is only 21.0%, 22.0%, 39.0%, and 65.0% with the Kelvin probe measured at 180, 90, 45, and 0, respectively. Yet, in most of the recent studies for solder joint electromigration, failure was taken when open-circuit occurred. Moreover, it is reported that degradation of Al trace often takes place at later stages of electromigration of solder joints. 9 Damage may also occur in other components rather than the solder bump itself. Therefore, this new failure criterion using Kelvin structure may provide a better way to measure mean-time-to-failure, and thus may facilitate the measurement of activation energy for electromigration failure in solder bumps. In summary, effect of void formation and propagation on bump resistance due to electromigration in flip-chip solder joints has been investigated using Kelvin structures with various layouts. The Kelvin structure can sense the highest voltage drop when its voltage probes are positioned at the current entrance into the bump in the chip side, whereas it only detects about a quarter of the voltage drop with the probes far away from the current entrance for the same size of voids. Therefore, the Kelvin structures with the probes near the current entrance have higher sensitivity on void formation. When the bump resistance increases 20% of its initial value, the depletion percentage of UBM opening is only 21.0%, 22.0%, 39.0%, and 65.0% with the Kelvin probe measured at 180, 90, 45, and 0, respectively. The results may provide a new failure criterion for electromigration test for flip-chip solder joints.

9 ACKNOWLEDGMENTS The authors would like to thank the National Science Council of the Republic of China for financial support through Grant No. NSC E In addition, assistance from the simulation facility of the National Center for High-performance Computing (NCHC) in Taiwan is highly appreciated. REFERENCES 1. K. N. Tu, J. Appl. Phys., 94, 5451 (2003). 2. S.L. Wright, R. Polastre, H. Gan, L.P. Buchwalter, R. Horton, P.S. Andry, E. Sprogis**, C. Patel, C. Tsang, J. Knickerbocker, J.R. Lloyd, A. Sharma, M.S. Sri-Jayantha, Proceeding of Electronic Components and Technology Conference, IEEE Components, Packaging, and Manufacturing Technology Society, San Diego, CA, 633 (2006). 3. Annie T. Huang, A. M. Gusak, K. N. Tu, and Y-S Lai, Appl. Phys. Lett., 88, (2006). 4. Y. W. Chang, S.W. Liang, and Chih Chen, Appl. Phys. Lett., 89, (2006). 5. S. W. Liang, Y. W. Chang, T. L. Shao, Chih Chen, and K. N. Tu, Appl. Phys. Lett. 89, (2006). 6. S. N. Gee, N. Kelkar, J. Huang, and K. N. Tu, Proceedings of IPACK 2005 ASME InterPACK, San Francisco, California, USA. 7. S. W. Liang, Y.W. Chang, and Chih Chen, J. Electron. Mater. 35(8), 1647 (2006). 8. E. C. C. Yeh, W.J. Choi, and K.N. Tu, Appl. Phys. Lett., 80(4), 580, (2002) 9. S. W. Liang, S. H. Chiu, and Chih Chen, Appl. Phys. Lett., 90, (2007) I