Thermal Performance of Lead-free Packages
12/01/2003
LEAD-FREE FAILURES DIFFER SIGNIFICANTLY FROM TIN/LEAD
BY MICHAEL MEILUNAS, STEVEN O. DUNFORD AND ANTHONY PRIMAVERA
Impending legislation is forcing the pace of progress toward lead-free electronic products, and the industry is close to finalizing suitable replacement solder alloys and associated processes. Component manufacturers and assembly businesses need to gain a deeper understanding of the likely characteristics of modern package types assembled using the most popular lead-free alloys — especially thermal performance and potential failure modes.
This article compares air-to-air thermal cyclic reliability and associated failure modes of second level interconnects in lead-free 1.27-mm-pitch 256 I/O ball grid array (BGA) devices by examining three leading lead-free alloys and comparing their performance to that of ordinary tin-lead eutectic solder. The lead-free alloys selected were 96.5Sn/3.5Ag, 95.5Sn/3.8Ag/0.7Cu and 96.2Sn/2.5Ag/0.8Cu/0.5Sb.
Test Vehicle Parameters
To evaluate the alloys properly, component substrates for a generic BGA device were bumped in-house by printing flux and pre-formed 0.762-mm-diameter solder spheres. The components were designed to display the thermo-mechanical behavior of standard BGAs and were assembled to electroless nickel immersion gold and copper organic solderability preservative-coated PCBs supplied by a single manufacturer. The assemblies were divided into two test groups (Test Group 1 and 2) based upon the thermal cycle used.
 Figure 1. Failed Sn/Pb solder joint after 6,801 20-minute cycles. |
null
Thermal Testing
The assembled boards were subjected to a 20- or 30-minute 0/100°C thermal cycle until failure, which was determined using continuous event detection. Visual inspection of each failed package allowed those that failed from causes unrelated to solder performance, such as package construction failure, to be excluded from subsequent lifetime and failure analysis. The post-cycle analyses performed include calculation of characteristic lifetime (Eta), visual comparisons of cross-sectioned failed joints and failure frequency plots for each solder alloy.
Characteristic Lifetime
Eta values are summarized in Table 1, in cycles, for all packages tested, and have been divided by Test Group and PCB finish. In each case, the Sn/Pb alloy has the lowest Eta. In general, the Sn/Ag alloy has a significantly higher characteristic life than the other materials. Differences between the Sn/Ag/Cu and Sn/Ag/Cu/Sb alloys are less apparent. Overall, reliability differences between the Sn/Pb and lead-free alloys are in line with previously published results.
Failure Analysis
Nearly 85 percent of the test vehicles were subjected to post-cycle analysis. All samples were X-rayed, and many were cross-sectioned to the failure location as determined by 4-point continuity verification. Cross sections were not necessarily performed through the mid-plane of the solder joints, because many of the packages were not electrically "open," but rather demonstrated resistance increases from their initial state. Dye penetration analysis was used to compile statistical data relating to which joint locations, if any, were most susceptible to failure.
Sn/Pb Failures. Each of the Sn/Pb failures evaluated contained fractured solder joints. The fractures were predominantly attributable to fatigue through the bulk solder material located in the soldermask-defined (SMD) region of the joint, near the component body, giving rise to well-defined crack paths (Figure 1).
Lead-free Failures. Analysis of the lead-free failures revealed a variety of failure modes and many interesting and potentially important issues associated with the alloys. Most noticeable was the existence of two distinct fracture paths: well defined cracks similar to those observed in Sn/Pb failures, and fine microcracks that were difficult to view without the aid of a scanning electron microscope. The well-defined cracks were easily located by 4-point continuity measurements. However, the very fine cracks were often difficult to isolate with a 4-point meter and required careful comparisons between pre- and post-cycle resistance measurements to locate.
 Figure 2. Failed Sn/Ag/Cu solder joint after 7,598 20-minute cycles. |
null
Additionally, the event detector output corresponding to assemblies containing microcracks was often similar to output associated with noise or intermittent failure. The event detection criterion was based on IPC SM-785, with a 300-Ω resistance level and time duration of 200 ns. However, resistance changes as small as 5-25 Ω for the duration of 40 ns were measured with a more sensitive event detection system in certain lead-free samples during thermal cycling. These small resistance levels were confirmed to coincide with crack propagation. Thus, selection of lower resistance thresholds may be required for accurately testing lead-free solder.
The lead-free joint shown in Figure 2 failed after 7,598 cycles. Portions of the crack are located in the bulk solder, while other portions have propagated into the intermetallic region. The formation, location and propagation of microcracks in Sn/Ag alloys in and near the intermetallic region have been noted in other works.1,2 The effects of intermetallics on joint reliability have also been presented in earlier papers, concluding that these have the potential to strengthen the joints.3,4 Cross-sectioning the failed joints allowed extensive examination of the formation of lead-free intermetallics.
The "shattered" cracks of this experiment were often difficult to isolate with a 4-point meter, and required careful comparisons between pre- and post-cycle resistance measurements to locate. Additionally, the event detector output corresponding to assemblies containing microcracks was often similar to output associated with noise or intermittent failure. The event detection criteria was based on IPC MS-785, with a 300-Ω resistance level and a 200-ns time duration. However, resistance changes as small as 5-25 Ω for a 40-ns duration were measured with a more sensitive event detection system in certain lead-free samples during thermal cycling. These small resistance levels were confirmed to coincide with crack propagation. Thus, seleciton of low resistance thresholds may be required for testing of lead-free solder compared to eutectic Sn/Pb solder.
 Figure 3. Failed Sn/Ag solder joint on Cu pad after 10,204 20-minutecycles. |
null
Generally, the wide, single-path-style cracks were observed near the component body in the SMD-defined region of the joints. The shattered cracks, however, were just as likely to occur near the PCB pad as they were near the component pad.
A portion of a Sn/Ag solder joint on a copper OSP PCB pad is shown in Figure 3. The crack has propagated around the NSMD PCB pad through the bulk solder. No correlation between cycles to failure and the shattered crack locations (component side vs. PCB side) was observed. No correlation between crack type and cycles to failure was observed.
 |
null
Vertical crack formations traversing the height of the solder joints were observed in many lead-free samples after thermal cycling. The joints were electrically good and neither event detection nor 4-point measurements detected the cracks (Figure 4). However, visual inspection of the solder joint surfaces and X-ray analysis revealed sizable cracks that may lead to spalling in some of the lead-free samples before in-plane fracture is completed, thus introducing a new failure mode. Vertical cracks were only detected in one Sn/Pb sample after nearly 17,000 thermal cycles, which was well beyond the useful life of the Sn/Pb BGA.
Another phenomenon observed in the lead-free assemblies is "sinkholes." Sinkholes are long, thin voids that originate at the solder joint surface that appear to form during reflow and grow during thermal cycling. Sinkholes were most common in the Sn/Ag/Cu and Sn/Ag/Cu/Sb alloys. A cross section revealed a sinkhole extending approximately 0.138 mm (about 1/6th of the solder joint's diameter) into a Sn/Ag/Cu/Sb solder joint (Figure 5). The effect of the sinkhole on the thermal or mechanical behavior of the joints is uncertain. No failures were directly associated with sinkholes.
Failure Frequency Plots
The BGA samples were designed to produce a homogeneous coefficient of thermal expansion. Such a design, when assembled to a PCB, should stress the solder joints in direct proportion to their distance from the neutral point (DNP) during thermal cycling. According to existing mathematical models, the corner-most joints, with the greatest DNP, should be most susceptible to fatigue damage. This hypothesis is known to be valid for Sn/Pb assemblies, and tests performed on the Sn/Pb BGAs in this experiment confirmed what was already known. However, to evaluate the hypothesis in the lead-free context, dye penetration analysis was performed on lead-free assemblies from each group to determine which joint locations were likely to fail. Low atmospheric pressure was required for the dye to penetrate the fine cracks found in the lead-free solder joints.
 Figure 4. Vertical crack formation in a Sn/Ag/Cu/Sb solder joint after 16,700 20-minute cycles. |
null
Our failure frequency diagrams show the number of BGA joints and their locations in the solder array where approximately 25 percent or more of the joint was fatigued. If the criterion is lowered below 25 percent, a far greater number of joints must be included in the diagrams because small cracks were located throughout the solder joint arrays.
Contrary to the model, the results from 38 Sn/Ag/Cu BGAs show failure locations that are apparently randomly distributed throughout the solder array. There was no significant difference between failure location and PCB finish. Similar results were observed with the 33 samples of the Sn/Ag/Cu/Sb alloy. On the other hand, analysis of failed Sn/Ag joints showed that the corner-most solder joints fail earliest in agreement with the DNP hypothesis.
 Figure 5. Sinkhole void Sn/Ag/Cu/Sb solder joint after 12,500 30-minute cycles. |
null
Conclusion
Lead-free solder alloys can be used in place of Sn/Pb in many applications. The reliability of Sn/Ag, Sn/Ag/Cu and Sn/Ag/Cu/Sb BGAs has been shown to be greater than that of Sn/Pb BGAs in 0/100°C thermal cycle tests. Sn/Ag solder displays the highest Eta, which is double that of ordinary Sn/Pb eutectic solder.
Three fatigue mechanisms were consistently observed within the lead-free systems: typical component side fatigue, multiple crack path solder fatigue and vertical cracks. Component side fatigue was almost always located in the bulk solder at the SMD-region of the joint below the PCB/component interface. The fine, multiple-crack-path solder fatigue was observed near both the component side and PCB side of the solder joint and was often difficult to detect with 4-point resistance measurements. Vertical cracking along the surface of the solder joint can occur during thermal cycle and may lead to spalling of the solder joints. Intermetallic plate formation may improve the reliability of lead-free solder joints by acting to redirect crack propagation. Dye penetration analysis indicates the Sn/Ag/Cu and Sn/Ag/Cu/Sb failures were not directly related to the DNP, but are believed to be "randomized" by factors such as intermetallic plate formation and grain boundary orientation, which can vary significantly from joint to joint.
This study has shown that lead-free failure times, mechanisms and locations differ significantly from those of Sn/Pb and that more work is required to better understand the consequences of lead-free soldering.
References
- Igoshev, V.I. et al., "Creep Phenomena in Lead-free Solders," Journal of Electronic Materials, Vol. 29, No.2, 2000.
- Suganuma, K. et al., "Microstructure and Strength of Interface Between Sn-Ag Eutectic Solder and Cu," Japan Institute of Metals, C59, 1995.
- ALbrecht et al., "A Study of Microstructural Change of Lead-containing and Lead-free Solders," Journal of SMT, Vol. 15, Issue 2, 2002.
- Poon, N.M. et al., "Residual Shear Strength of Sn-Ag and Sn-Bi Lead-free SMT Joints After Thermal Shock," IEEE Transactions on Advanced Packaging, Vol. 2, No. 4, November 2000.
MICHAEL MEILUNAS, process research engineer, may be contacted at Universal Instruments Corp., P.O. Box 825, Binghamton, NY 13902; (607) 779-7522; e-mail: [email protected]. STEVEN O. DUNFORD, research engineer, may be contacted at Nokia Mobile Phones, (972) 374-1616; e-mail: [email protected]. ANTHONY PRIMAVERA (formerly a research engineer at Universal Instruments Corp.) may be contacted at Guidant Corp., (651) 582-7937; e-mail: [email protected].