Abstract
Continuous power density increases and interconnect scaling in electronic packages raises risk of electromigration (EM) induced failures in high current interconnects. Concurrently, thermal cycling fatigue also places interconnects at risk of reliability failure during electronics' operating lifetime. These two differing failure mechanisms are historically treated separately, but in operation, the combination of EM effects and thermal cycling can act synchronously in accelerating failure. Presently, there is no model to predict the complexity of reliability estimation arising from these interacting failure modes but is certainly important for high current density applications. In this work, a novel testing system has been employed to facilitate the estimation of the reliability of solder interconnects under the combined influence of EM and mechanical strain. The system subjects solder interconnects to high current density, elevated ambient temperature, and a constant tensile stress while recording the change in electrical resistance and change in length of the solder over time. The solder samples were created using two copper wires connected by a eutectic Pb/Sn solder ball to imitate flip-chip or BGA packaging interconnects, allowing for controlled testing conditions in order to demonstrate the combined effects of a mechanical load and EM on the lifetime of a solder joint. A significant reduction in lifetime was observed for samples that endured the coupled accelerating factors. Comparing the experimental results of different current densities at different stress levels provided a new outlook on the nature of coupled failure acceleration in solders. This novel test methodology can inform model generation for better anticipating the failure rate of solder interconnects which naturally experience multiple stress inputs during their lifetime.
Introduction
Flip-chip bonded solder joints have contributed to the miniaturization of numerous device architectures, such as those utilized in ultra-large-scale integration, due to the high I/O density that flip-chip bonding processes can provide [1]. While reducing the size of electronic components brings many advantages including smaller and lighter-weight devices with lower power losses and parasitic effects, in several instances, the effects of electronic device miniaturization on solder interconnect reliability in next generation power electronics packages need to be understood. Recently, power electronics have seen movement toward flip-chip bonds to lower parasitic inductances, heighten integration of control circuits, and realize multidimensional thermal management techniques [2].
Due to their low melting points, common solders have low tolerance to electromigration (EM) effects among materials that are used in electronic back-end processes such as flip-chip bonding. Based on the current design recommendations for micro-electronic packaging, a single solder bump connection of a typical flip-chip (as low as 50 μm in diameter) bonded device must tolerate up to 0.4 A [3]. Under this consideration, the next generation of solder bumps may carry current densities greater than 104 A/cm2 (≈ 0.8 A per solder bump) as a result of device miniaturization [3]. Increased current densities in the context of higher operating temperatures indicate that EM effects will become a significant failure risk in solder joints as device miniaturization continues so appropriate measures must be taken to ensure reliable operation through an accurate understanding of this concerning failure mechanism.
Electromigration Failure.
One of the failure mechanisms observed at device interfaces exposed to current stresses is EM, which results in the formation of voids within connection regions (see Fig. 1). The formation of EM voids in a current carrier is a process first reported by Black, who demonstrated that EM failure is driven by high current density and elevated material temperature [4]. In Pb/Sn solders, voids are normally initiated at areas of current crowding and propagate across the solder interface [1]. In Sn-based lead-free solders with copper under bump metallization (UBM), voids appear under the stress of electrical current at the Cu6Sn5/solder interface and propagate through the Cu3Sn/Cu6Sn5 interface, which are the two intermetallic compound (IMC) layers in these types of connections [3]. For a connection with Ni UBM, voids form at current crowding areas and propagate through both the IMC and solder interfaces by forming pancake-type voids [4]. Dissolution of the UBM and growth of IMC layers is another result related to solder current stressing. Extensive growth of Cu6Sn5 and Cu3Sn IMC has been observed in connections with Cu UBM at the anode side of the connection [6–8], while for Ni UBM the growth is less extensive as a result of the slower reaction rate between Ni and Sn [5,9]. Another result of solder current stressing that has been reported is the growth of whisker hillocks on the cathode side of the connection as a result of accumulation of Sn atoms due to EM processes. As interconnect current densities continue to increase, the potential for EM failures will continue to increase.
Thermomechanical Failure.
In addition to EM, another reliability challenge in the field of micro-electronic packaging is thermomechanical fatigue (TMF). As multicomponent junctions, solder connections are affected by shear and peel stresses that arise from operational or environmental temperature fluctuation due to the differences in the coefficients of thermal expansion (CTEs) of the entities that the solder junctions connect. This can create grain boundary sliding and decohesion in solders leading to crack formation and propagation [10,11]. To better understand the endurance of solder under TMF conditions, accelerated life testing (ALT) methods such as thermal cycling, thermal shock, and power cycling have commonly been used in other studies [12]. Relations between cycles to failure and inelastic strain energy per volume, finite element modeling, material-percentage based, and total energy-based models have all been proposed for estimating the lifetime of solders enduring TMF [13].
Combined Failure.
Atomic movement is affected by mechanical stress, temperature, and electrical current, all of which contribute individually to EM and TM failures [14]. Although EM and TMF tests examine the lifetime effects of specific types of stresses on electronic devices, the combined effects of these mechanisms are not well understood. It is apparent that EM-based voiding and fatigue mechanisms in solder interconnects will certainly experience TMF and EM degradation simultaneously. To achieve a realistic lifetime estimation of electronic devices, the combined degradation mechanism of these damages must be examined.
While many studies have been conducted to understand the failure mechanism of solder joints under a single test condition, so far there have only been a limited number of studies that considered the coupled effects of EM and TMF in solders. By utilizing experimental results, Zuo et al. observed that EM significantly reduces TMF lifetime in SnBi solder connections when applied simultaneously, but delays crack formation under creep stresses [15]. In another work, Zuo et al. studied copper diffusion through SAC solders and the crystallographic transformation of the solder under the combined effects of thermal cycling and EM which showed that these effects facilitated void formation, grain boundary sliding and, in turn, crack formation [16]. Ma et al. used an experimental method to characterize the coupling effect of EM and TMF on the evolution of microstructures and resistance change for SnAgCu (SAC) and SnBi solder interconnects [17]. Pharr et al. suggested an analytical model to predict the complicated mechanism of coupled creep and electromigration in solders; however, this theory needed experimental data to support or modify the model [18]. Recently, Cui et al. have suggested a coupling model to describe atomic mobility based on EM-induced volumetric strain, thermally induced strain, and mechanical deformation, which aligned well with Blech experimental data but only assessed one-dimensional metal line solutions and did not account for temperature gradients [19].
The current work in this field would benefit from the introduction of a test setup that enables the study of the combined effects of temperature, electrical current, and mechanical stress on solder joints to drive the development of a lifetime model for solder interconnects in modern electrical interconnects, which is demonstrated in this work.
Design of Accelerated Test
The intent of this testing is to subject a single solder joint to electrical, mechanical, and thermal stresses simultaneously to observe:
how the combined stresses effect the overall lifetime of a solder joint
and how much of an influence on the lifetime each stress has when the three stresses are allowed to act upon the joint simultaneously.
To accomplish this, a novel testing system and method were developed to apply these conditions to four solder joints simultaneously and monitor their response to the stresses imposed upon them. A total of 72 solder joints were tested to perform this work. The failure criteria for a failed solder joint was considered to be a complete separation, or break, through the solder joint.
Sample Preparation.
Several methods have been previously suggested for fabricating test samples for EM studies [14,15]. However, the suggested methods do not enable the application of a tensile stress to the sample. In this study, a single sample consisted of two 0.01” diameter magnetic copper wires connected to each other with a 0.012” diameter eutectic Pb/Sn solder ball. The insulation layer on the magnetic copper wires served as a barrier that prevented solder connections on the lateral areas of the wires, yielding solder connections with more uniformity than samples created using uninsulated wire. The reflow profile used to create these samples consisted of preheat (25–125 °C in 1 min), preflow (125–183 °C in 2.5 min), reflow (183–210 °C in 45 s, hold at 210 °C for 45 s), and cool down (210–25 °C in 1 min) stages, totaling 6 min for a full profile run. Detailed descriptions of the samples and their fabrication can be found in Appendix A (Fig. 2) of this article.
Test Setup.
To impose linear strain on a test solder sample, a mechanical structure was designed consisting of linear sliding bearings and a fixed pulley as shown in Fig. 3. To apply tension to the solder sample, one end of the sample was attached to a copper terminal and fixed to a wooden base plate. The opposite end of the sample was attached to a three-dimensional printed fixture which sat atop the linear bearing. A weight (ranging from 5 to 20 g, resulting in a tensile stress range from 0.96 to 3.85 MPa) was attached to the three-dimensional printed fixturing via a flexible cable and hung vertically over a pulley. This stress range was used after consulting similar mechanical stressing studies on micro-electronic interconnect materials, which utilized “peeling,” tensile, or compressive stresses ranging from 40 to 80 kgf/mm2, 0.5 to 7 MPa, and 10 to 60 MPa, respectively, [20–22]. This applied force created a change in the length of the solder sample over time which was recorded using an optical mouse sensor. The combination of the mass applied to the sample and displacement data confirmed the strength of the force being applied to the solder sample.
To apply an elevated ambient temperature to the solder sample, the testing mechanism seen in Fig. 3 was designed to allow for the addition of a ceramic heating chamber in between the linear bearing assembly and fixed end of the sample. Two 50 W heating elements (US made Bestol AC DC 24 V) and a k-type thermocouple were used to control the ambient temperature within the chamber by means of a PID control loop written to an Arduino MEGA. After calibration, this configuration had the ability to regulate solder sample temperatures of up to 120 °C, the maximum target ambient temperature considered for the study based on several device maximum operating temperatures. The two ends of the solder sample were connected to a constant current power supply (UK made aimTTi MX180TP 375 W) and the change in voltage of each solder sample was measured in situ. The data from the power supply and voltage sensors, configured for kelvin (4-wire) voltage measurement, was combined to find the change in sample electrical resistance over time.
Figure 4 depicts the final test setup after being built, capable of operating four simultaneous EM-mechanical strain experiments. Sample resistance, ambient temperature, and elongation are recorded at a user-defined sample rate. For the experiments performed in this study, a sample rate of 0.5 S/s was used.
Experimental Results
Physical Observations and Failure Analysis.
Figure 5 displays a comparison of three different types of test conditions in this experiment. A ductile deformation was observed with a considerable deformation of the material and necking for the creep test condition. Therefore, it is hypothesized that void formation in grain boundaries was expected to cause breakage due to the expectation that void formation will weaken grain boundary locations. For solder samples tested without an applied stress, it was observed that sample breakage occurred on the cathode side of the sample, as was expected (right-most image in Fig. 5). For the samples that were tested under the combined effects, the failure happened much faster than creep and EM samples by themselves. For combined stress tests, the solders typically showed less ductile deformation than samples tested solely under creep conditions. This smaller ductile deformation was attributed to the current stress causing void growth generated during the creep deformation. The raw data taken for all samples showed that adding electrical stress to a creep test reduced the time to failure by roughly four times while the time to failure for the EM sample was more than seven times higher than the creep sample and 30 times higher than the combined effect sample. For a complete set of microscopic images and their corresponding data plots, please contact the corresponding author.
Seventy-two total samples were tested for this work, but several samples were tested under the same conditions to reduce time to failure (TTF) standard deviation. The 14 experimentally tested conditions are summarized in Table 3 (see Appendix B) and plotted in CDFs in Figs. 6 and 7. The interaction between creep deformation and current varied based on the input conditions. For example, for two of the conditions at 5893 A/cm2 (with applied stresses of 0.96 and 1.93 MPa, respectively), added tensile stress increased deformation. The opposite observation was made for two conditions at 6875 A/cm2 (with applied stresses of 2.89 and 3.85 MPa, respectively).
k (eV/K) . | j (A/cm2) . | A . | n . | Q (eV) . |
---|---|---|---|---|
8.6137 × 10–5 | 5893 | 6.85 × 10–70 | 2.33 | 6.15 |
6875 | 7.75 × 10–65 | 2.52 | 5.8 |
k (eV/K) . | j (A/cm2) . | A . | n . | Q (eV) . |
---|---|---|---|---|
8.6137 × 10–5 | 5893 | 6.85 × 10–70 | 2.33 | 6.15 |
6875 | 7.75 × 10–65 | 2.52 | 5.8 |
Using the estimated parameters from Table 1 as a control group, comparisons between results for EM alone and EM combined with mechanical stressing were done in Fig. 8 to demonstrate the effects that applied stress had on lifetime in this study. All cases for combined stressing demonstrated a reduced lifetime compared to estimates from Black's equation for their respective current density and temperature conditions.
In Fig. 8, the slope of the blue line is one, meaning that the parameters in Table 1 match the observed data very well for EM conditions without applied tensile stress. For all conditions tested with applied tensile stress, the slopes are much less than one, confirming that Black's equation for EM does not predict the TTF appropriately when EM conditions are combined with externally applied mechanical stresses. The predicted TTFs based on Black's equation for samples tested under tensile stress overestimate the time to failure by 83–94%. This observed reduction in lifetime emphasizes the need for current and future efforts to be made to understand the complicated relationship between electromigration and other interacting mechanical and thermal failure mechanisms, as the potential for these interactions will increase as device miniaturization continues.
Statistical Analysis.
Simple analysis of variance (ANOVA) and a general linear regression analysis was used to compare the 72 data points collected in this study in addition to the lifetime plots in Figs. 6 and 7. This simple linear comparison revealed that the combination of EM and mechanical stress conditions reduced the expected lifetime of a eutectic Pb/Sn solder joint when compared to electromigration conditions alone. The linear regression equations obtained in Fig. 9 demonstrate that for each current density and temperature combination, the TTF is reduced by the introduction of a tensile load. This is expected for the first two conditions with no applied current density, as these were essentially crept tests. The remaining four cases, on the other hand, indicate that the interaction between EM and tensile stress failure modes do have a negative impact on the lifetime of the solder joint (Fig. 9).
A covariance analysis was also conducted to demonstrate the effects of subjecting the samples to multiple simultaneous stresses. Table 2 shows key results of the covariance analysis, which was carried out to show the effective reduction in TTF (in minutes) as a result of the combined conditions imposed on a solder joint. In Table 2, the negative values for each of the combined stress conditions indicate that the combination of these stresses reduced the expected lifetime when compared to the other combined scenarios. Therefore, the combination of current density and tensile load had the largest impact on solder joint lifetime, followed by EM stresses alone (the combination of current density and temperature). The case for the combination of temperature and tensile stress had the smallest impact on lifetime.
Discussion of Results
The results of this work demonstrate that the combined stress environment experimental setup used in this study is a viable option for creating a variety of conditions to observe possible failure mechanisms associated with eutectic Pb/Sn solder joints. From the data collected in this study, the interaction between current density, temperature, and mechanical stress conditions on a eutectic Pb/Sn solder joint was shown to be complicated in nature but their combined presence did demonstrate a reduction in lifetime compared to EM conditions by themselves. Furthermore, it is clear from the results that this testing method is also capable of producing data that confirms that the combination of these various stresses will change the predicted lifetime of a sample when compared to a single enacting failure mode. From these results, it is clear that other forms of mechanical stress must also be combined with EM conditions in order to more accurately estimate solder interconnect lifetimes in high power density configurations.
In real-world applications, a single solder joint will not solely be subjected to a linear mechanical stress. In a thermal-electric environment, shear, torsion, compression, and tensile stresses are all potentially acting on various solder joints as a result of estimates of stresses. The methods used in this study solely sought to demonstrate that mechanical stress will indeed affect the lifetime of a solder joint that is already subjected to accelerated failure conditions and show the importance of characterizing the lifetime effects that arise from package-induced mechanical forces on electronic assemblies.
Conclusion
In this study, a novel testing system was used to document the effects of combined electrical, thermal, and mechanical stresses on eutectic Pb/Sn solder joints in an accelerated testing environment to better understand their lifetime. It was shown that:
- (i)
The introduction of creep deformation onto a solder joint already experiencing EM conditions accelerated the time to failure, but the overall rate of damage accumulated to the joints was dependent on the three stress factors of:
Current density
Temperature
Tensile Stress
While creep deformation was shown to accelerate EM effects in the samples tested, the available accelerated reliability testing methods used in this study did not necessarily capture the decrease in lifetime of the solder during significantly heightened current stresses. From this study, it can be concluded that:
(ii) The complicated nature of tensile stress combined with EM effects must be adequately understood when assessing reliability of systems that naturally experience combined stress environments
Moreover, these findings illustrate the need and significance of reliability models that can accommodate the coordinated damage induced by interacting stresses.
Acknowledgment
The authors would like to thank the University of Arkansas Mechanical Engineering Department and Engineering Research Center for providing access to all of the facilities and resources that made this study possible. The authors would also like to thank Collin Ruby for his work done to collect solder failure images.
Funding Data
National Science Foundation (NSF) (Grant No. 2014-00555-04; Funder ID: 10.13039/100000001).
Nomenclature
- ANOVA =
analysis of variance
- CDF =
cumulative distribution function
- CTE =
coefficient of thermal expansion
- EM =
electromigration
- IMC =
intermetallic compound
- I/O =
input/output
- TTF =
time to failure
- PID =
proportional/integral/derivative
- SAC =
SnAgCu
- SiC =
silicon carbide
- TMF =
thermomechanical fatigue
- UBM =
under bump metallization
Appendix A
Sample Fabrication.
Samples were typically created in batches of 12 and were created using the following steps:
Cut 24 2 in. long lengths of 0.01 in. diameter magnetic copper from a spool of wire.
Remove the insulation layer along 1 in. of one end of the lateral surfaces of the wire on the side of the wire which is not to be bonded to solder using 300 grit SiC sandpaper.
Polish the base surface of the lengths of wire which were selected as the side to be bonded to as follows:
(i) Begin polishing with 5000 grit SiC sandpaper using water as lubricant.
(ii) Polish the same ends using a 30 μm alumina polishing pad with water as lubricant for deburring after the first polishing step.
(iii) Polish the same ends using a 0.3 μm alumina polishing pad with water as lubricant.
(iv) Clean the ends of the wires using acetone.
Fix 12 of the polished copper wires to a ceramic plate using Kapton tape.
Apply solder flux to the bonding end of each of the 12 polished wires on the ceramic plate.
Using a microscope and tweezers, place an individual 0.012 in. Pb/Sn solder ball on the end of each wire which was coated with the solder flux.
Coat the polished ends of the remaining 12 wires with solder flux and place the coated ends on the ceramic plate touching the already fixed solder ball and wire assembly. Place Kapton tape across all samples after placement to maintain contact between the wires and solder balls.
Place the batch of samples (still fixed to the ceramic sheet) into a Sikama Falcon reflow oven and run the appropriate temperature profile for the solder (see Sample Preparation section for reflow information).
Remove the ceramic sheet after the samples have cooled and removed the samples from the sheet.
Clean the samples in acetone to remove any remaining solder flux and adhesive from the Kapton tape and place the finished samples in a refrigerator until testing begins. See Fig. 2 for a microscope image of a single solder joint fabricated using these steps.