BGA underfills Increasing board-level solder joint reliability BY ZANE JOHNSON AND THOMAS KOSCHMIEDER Figure 1. Dispense pat-terns for the three pack- ages used for evaluating BGA underfill materials.Click here to enlarge image A variety of industry packaging trends are affecting the board-level reliability of packages. For instance, the trend of making package body sizes smaller affects board-level reliability by decreasing the pitch between balls in the ball grid array (BGA). As the pitch decreases, the ball size is reduced, and therefore the gap (standoff height) between the printed circuit board (PCB) and package is decreased. The shorter standoff height usually reduces the board-level reliability for a package. Other industry trends include the raised expectations of customers for the overall reliability of electronic assemblies. Customers are increasingly expecting longer warranties for final products, so the manufacturers of components for these final assemblies must ensure that the assembly can meet such goals. A good example of this trend can be found in the automotive industry, where the consumer can expect long warranties for cars and a three-year warranty for parts. Finally, many large IC manufacturers no longer provide ruggedized packages because of a low volume of sales. So, the customers for these products – if they want to stay current with present microprocessor technology – must find ways of increasing the overall reliability of the assemblies. A variety of techniques are available for increasing the board-level reliability of packages. Among these are low expansion PCBs for ceramic BGA (CBGA) packages, interposers between the package and PCB, and re-packaging a device. Underfilling of BGAs to provide a compliant layer between the package and PCB is another method and the one discussed here. Package and Test PCB Characteristics A range of packages was used in these experiments because the package types available are quite varied. A 255-pin CBGA package (21 x 21 mm and 1.27 mm pitch) was one package used. Balls for this package are 90%Pb and 10%Sn and attached to the ceramic substrate using eutectic 63Sn/37Pb solder paste. Flip-chip CBGA is a common package for high-end microprocessor devices. A small tape-based plastic BGA package with 280 balls was included in the study (16 x 16 mm, 0.8 mm pitch, 9 x 9 depopulated center). A third package tested was a 196-pin mold array process (MAP) package (15 x 15 mm, 1.0 mm pitch). These small pack-ages are typically used for microprocessors for consumer electronics. Figure 2. Failure rate as a function of the number of temperature cycles and underfill material for (a) CBGA, (b) tape-based plastic BGA, and (c) MAP packages.Click here to enlarge image Unlike in CBGAs, the presence and size of the IC affects the board-level solder joint reliability of plastic BGAs. The sharp change in coefficient of thermal expansion (CTE) at the die to mold/substrate interface is the highest stress area in the package and there-fore the area most likely to fail. The CTE does not matter very much in the CBGA because of the high stiffness of the ceramic substrate. The die size for the 280-pin tape-based plastic BGA was 10.7 x 10.1 mm and 10.2 x 9.6 mm for the 196-pin MAP. The purpose of the experiments was to see if the solder joint life can be extended, so daisy chain test vehicles for the packages were used. Daisy chain patterns typically make an electrical connection between neighboring balls with a trace inside the substrate. With the PCB electrically con-necting the unconnected balls in the pack- age, a complete circuit exists in the final assembly of package to board. The electrical circuit can be monitored for changes of resistance over time to detect failures, such as electrical opens. Underfill Epoxy Characteristics A primary goal of the BGA underfull experiments was to determine which underfill properties matter for aiding solder joint reliability. The epoxy properties that might be relevant include coefficient of thermal expansion (CTE), elastic modulus, filler size, filler type, cure time, cure temperature and flow speed. The CTE and modulus are the primary factors affecting reliability, and these are the variables studied here (Table 1). Table 1. Properties of epoxies evaluated as BGA underfill materials.Click here to enlarge image All of these epoxies are “snap-cure” epoxies with cure times of less than seven minutes at temperatures of 165°C or less. Materials with short cure times were chosen to minimize the impact on cycle time in high-volume production. Board Assembly and Underfill Process After board assembly, the packages were ready for epoxy underfill dispensing using any of the many available dispensers. Important features in dispensing equipment for this application are a method of heating the board/packages to between 80 and 100°C, and an auger screw type of dispense pump. Heating the PCB is useful because most underfill epoxies have a low viscosity (fast flow time) somewhere in the 80 to 100°C range. An auger screw dispense pump (sometimes known as a positive displacement pump) is important because of excellent dispensing repeatability. Figure 3. A resistor processed with underfill in a (a) “glob-top” and (b) “side” configuration.Click here to enlarge image Choosing a dispense pattern for putting the epoxy next to the pack-age can be complex. Possibilities include a single line dispense along one side of the package, multiple dis-penses along one side of a package, and dispenses along multiple sides of the packages. In this experiment, dis-pense patterns with some version of an “L” along two sides and a final seal pass around all sides were used to create a relatively uniform fillet (Figure 1). The same dispense pat-terns were used for each epoxy. It was straightforward to watch the dis-pensed epoxy flow through the BGA and out from under the package, to verify the suitability of the dispense pattern. It was also important to watch for signs of voids, because trapped air can be a reliability concern. The epoxy curing process consisted of placing the PCB in an oven at 165°C. After seven minutes, the PCB was removed and placed on a metal table to cool down to room temperature. To ensure a proper cure profile for actual manufacturing, it is appropriate to create a profile board with a thermocouple buried in the epoxy between the package and PCB. Accelerated Temperature Cycle Testing Once the PCBs were underfilled, a method of stressing them was needed so that differences in board-level reliability life could be determined. Air-to-air temperature cycling (AATC) was chosen. Other methods, such as air-to-air temperature shock or liquid-to-liquid thermal shock, are viable alternatives. AATC means that the air temperature inside a chamber is changed. Typical temperature ranges are 0 to 100°C and -40 to 125°C. The 0 to 100°C condition was used for the CBGA exper-iment and has a 30-minute cycle with five-minute dwells at the temperature extremes, and 10-minute ramps in-between. The two plastic packages used the -40 to 125°C condition with 15 minute dwells and ramps. Historically, CBGA packages have been tested at 0 to 100°C, which Motorola adopted for its own uses. The automotive industry uses -40 to 125°C, which Motorola has adopted for most plastic substrate packages. Several techniques exist for monitoring the resistance through the daisy chain nets as a function of total temper-ature cycles experienced. The simplest is to probe the resist-ance nets by hand at regular intervals (hours, days, elapsed cycles, etc.). This method provides inspection data and has been commonly used for board-level reliability testing. A more complex method using in situ monitoring was used here. This method monitors the resistance through the daisy chain nets on a continuing basis. When the resistance goes through a particular threshold (such as that defined in IPC-SM-785), the monitoring equipment logs the package as failed and records the elapsed cycles. The data acquired with in situ moni-toring is continuous (i.e., all cycles to failure exactly known). Temperature Cycling Results Figure 2a shows the final results for the temperature-cycled parts with various underfills and the control (no underfill). Comparing epoxy A with the control shows that a high CTE epoxy can detrimentally impact the board-level reliability, because the control lots fare better. The two epoxies with CTEs in the 40 ppm/K range (epoxies B and C) show almost the same reliability. Both exceed the board-level life found for the control units. Epoxy D shows a significant lifetime improvement when compared to the control. A goal of 3,000 cycles of 0 to 100°C AATC for pole-mounted telecommunications applications is easily met with epoxy D. The 280-pin tape-based plastic BGA results (Figure 2b) were somewhat similar to those for the CBGA. The lowest CTE mate-rial (epoxy D) performed significantly better than the control. One result that was different from the CBGA experiment was that the two 40 ppm/K materials (epoxies B and C) performed differently. The higher modulus material (epoxy C) failed earli-er than epoxy B, and it is interesting that epoxy C actually had lower reliability than the high CTE material (epoxy A). For the -40 to 125°C AATC, a good rule of thumb is 2,000 cycles without failure to make the package available for automotive applica-tions. So, with either epoxy B or D, the tape-based plastic BGA package can be considered for these applications. Figure 4. Weibull plot of underfilled resistors.Click here to enlarge image For the MAP BGA, only two epoxies B and D were evaluated, and another variable introduced was the PCB pad size (because plastic BGA package solder joint reliability is known to depend upon pad size). Figure 2c shows that the smaller PCB pad size results in a longer lifetime than the larger PCB pad when there is no underfill. For epoxy B, the PCB pad size did not affect the reliability of the underfilled MAP. With underfill, the stresses are spread over the entire surface area covered by the underfill epoxy. This relief to the solder joint areas makes the pad size less important. Again using the automotive application rule of thumb, (2,000 cycles of -40 to 125°C), epoxies B and D would both make the reliability of the 196 MAP BGA acceptable. PCB Component Clearance Consideration PCBs for daisy chain packages for this kind of experiment are typically designed to have separation between the packages, making access for failure analysis easier. When comparing this type of PCB to a true final assembled PCB, it is apparent that packages to be underfilled on a PCB will preferably have the PCB designed to allow for the needle to run next to the package and for a fillet region around the package. A space of at least 3 mm all around the target component is a useful spacing guide. For many existing PCBs, such spacing will be difficult to maintain, often because of passive components that fill available spaces. To understand the impact on reliability of underfilled passive components, resistors were assembled to PCBs, underfilled and then temperature cycled. Two underfill patterns were tried: “glob-top,” which was an attempt to bury the resistor in epoxy, and “side,” which had the epoxy run into the resistor as if the fillet ran out (Figure 3). Results show that underfilling the resistors by either method adds life to the solder joints (Figure 4). This is a positive result for the many existing PCBs that might benefit from underfilling the BGA and yet have passives very close to the package of interest. Underfill Finite Element Simulation To help identify the underfill properties that would maximize the solder joint life for the 280-pin plastic BGA package, a finite element (FE) model of the finished assembly was built. Similar models have been used extensively in the industry to understand the importance of design and material property parameters. In the previous section, underfill D was identified as the best choice of the four snap-cure materials. Using the model with the properties from Table 1, it is possible to conduct a parametric study of underfill properties – varying the modulus and CTE – to identify the modulus and CTE that would maximize the solder-joint life of the package. A full factorial numerical study was per-formed, using elastic modulus values of 2, 5 and 10 GPa, and CTE values of 5, 15, 25, 35, 55 and 75 ppm/K. The optimum underfill material (having a characteris-tic life in excess of 10,000 cycles) would have a CTE near 15 ppm/K, but the modulus could range between 2 and 10 GPa without adversely affecting fatigue life. This 15 ppm/K value is close to that of typical mold compounds and PCB laminates, and appears to minimize the thermo-mechanical loads applied to the solder joints. While a CTE of 15 ppm/K would be difficult to obtain with current epoxy resins and filler materials, it does provide clear direction for underfill development. Underfill D is in a high-gradient zone, and any reduction in CTE would bring a significant improvement in fatigue life. Conclusions Underfilling BGA packages with commercially available “snap-cure” epoxies can lead to significant board-level reliability improvement. It has been found that the CTE of the underfill epoxy should be low, approaching the PCB's CTE. Additionally, a low modulus for the underfill epoxy aids reliability. Such improvements open both ceramic and plastic BGA packages to application environments that were previously closed. Further development work, both experimental and numerical, will enable the adoption of CBGA and PBGA packages in many new areas. AP Zane Johnson, senior staff engineer, and Thomas Koschmieder, senior staff scientist, can be contacted at Motorola SPS, 3501 Ed Bluestein Boulevard, Austin, TX 78721; 512-933-5813; Fax: 512-933-5444; E-mail: email@example.com and firstname.lastname@example.org.