By Steve Shermer and Kevin Gaffney, Amkor Technology Advanced Package Development Center, Chandler, Arizona
As flip-chip packaging achieves growing acceptance throughout the industry, flip-chip producers are faced with solving manufacturing challenges, typically related to handling, for optimum yield and throughput. Flip-chip CSPs are generally processed in matrix strip formats containing anywhere from 16 to 100 die per strip at a pitch between 8 and 11 mm. High-performance die are typically processed in carriers holding die ranging from 11 to 26 mm square. Non-CSP package sizes may vary from 23 to 50 mm on a side. Presentation Formats Die are presented to flip-chip attach machines in waffle packs, tape-feeders and wafer ring formats. Wafers mounted on tacky tape and wafer rings are perhaps the most popular of die presentation formats, particularly in the conventional die attach process. Waffle packs and tape allow for populating packages with known-good die (KGD). Issues in waffle-pack handling include controlling the aspect ratio and size of the die relative to the waffle pack's cavity size. This requirement is needed to reduce die shift during handling, to shorten search time for die fiducials and to allow the optimum pick tool to be used. Ideally, the cavity size should be no more than 10 percent larger than the die size in each axis. Throughput, when using waffle packs, is an issue due to the relatively few die that can be placed in either a 50 mm or 100 mm waffle pack. Both waffle-pack and tape-feeding require die sort/pick-and-place processes upstream from chip attach, but tape-feed is better suited to higher volumes. For larger production volumes, wafer rings are generally preferred. This presentation method, however, is fussy about die-eject needle and cap sizes, needle end-height and eject speed. Without attention to these considerations, die cracking, micro fracturing, and mis-picking may occur. With wafer tape, it is important to carefully match the eject chuck (or cap) and eject needle spacing to the die size. There should always be a center needle, and the needles around the perimeter should generally be spaced on no less than 80 percent of the die perimeter, as in Figure 1.
Avoiding Cracking The critical issue for the needle-eject parameter is to avoid nicking the back side of the die, which can induce cracking. Eject needles with a radiused tip will not pierce through the tape, which helps avoid nicking; however, a two-stage eject process is also necessary (Figure 2). A short delay after the first ejection stage allows the tape to peel away from the die corners, while the tape around the pins remains in contact with the die. Then the needles can be raised to the second position. Larger dies require a longer delay for the tape to peel away from the edges. In addition to being sized for the die, pick tools for full-array flip-chip die (where the die-top surface is fully populated with bumps) and large die (greater than 10 mm2) should possess a compliant contact surface to maintain vacuum without damaging surface features. On the other hand, for peripheral bumped die, tools tipped with a harder material, such as Vespel or Delrin, are appropriate. The harder tip alleviates die-sticking during placement of smaller die. In either case, the material must be anti-static to avoid ESD damage to the circuit. Die Indexing Regardless of the handling method, one of the most common challenges in flip-chip-in-package assembly is adjusting the placement machine's vision system so that it can reliably index the die. Interestingly, the wavelength of light used to view the fiducial may need to be matched to the IC packaging material. Ceramics, metals, polymers and semiconductors all exhibit unique reflection and transmission characteristics. These characteristics translate to a range of contrast, brightness and glare considerations. Therefore, in addition to adjusting the brightness, f-stop, incident-light angle, aperture, etc., it may be necessary to change the illumination from white to red, for example, to ensure that the vision system can reliably locate the fiducial. Vision Recognition In addition to optimizing lighting, it is highly desirable to have dedicated fiducials, or vision recognition points, on the individual die and substrates. Although any unique pattern can be used as a fiducial, the probability of a locating a truly unique pattern can be difficult. For instance, bond pads are an obvious choice for vision recognition; however, if the die has been previously probed, the residual probe marks will confound vision recognition efforts. A specific example for development activities is the use of die circuitry as a fiducial. In this case, mask revisions or passivation changes will confound the vision recognition. Solder Fluxing Methods for applying flux to flip-chip bumps and pads include drum or stamp fluxing, print fluxing and flood or dispense fluxing. The method selected depends somewhat on the material properties of the fluxing agent, as well as process time and the equipment cost. In addition, the volume of flux per bump and the total surface area over which the flux acts can affect both the downstream process and product reliability. The drum (stamp) method employs a small tray inside the flip-chip attach machine. Flux in the tray is leveled to the desired height with a doctor blade and each die is picked from the feed source, moved to the flux tray, dipped or "stamped" into the tray and placed on the substrate. This is a simple and efficient method that localizes the flux to the chip bumps. Process control with this method can be a challenge, however, due to difficulties in measuring the thickness of the flux in the tray. The print method is a standard screen-print process that applies flux to the substrate rather than the die itself. This method can apply flux to a number of die sites quickly, but it requires upstream equipment and processing. As with the stamp method, it is difficult to accurately measure flux volumes with the print method.
The dispense (flood) method offers the virtue of simplicity. Liquid flux from a pneumatic syringe is dispensed at the center of each flip-chip site on the substrate and flows out across the substrate, fluxing each pad. Process time is short, but the volume of flux deposited can be large, and interactions between the flux and substrate may be complex. Flip-Chip Self-Alignment Flip-chip die tend to "self-align" during bump reflow. The surface tension of the liquid solder wetting the bond pads tends to pull the chip into perfect alignment with the bond pads. Due to this factor, flip-chip bumps can be as much as 25 percent out of perfect alignment from the exact center of the bond pad. The larger the bumps, the higher the placement tolerance. This feature also requires that substrate bond-pad locations and tolerances be tightly controlled. For example, even though a 25 percent misalignment can be tolerated, the offset must be in the same direction for every bond pad. If the bond-pad registration drifts in the +X direction on one side of the bond site but the -X direction on the opposite side, the liquid solder will still wet to the pads. Now, however, instead of having the desired self-aligning feature, the solder balls will actually stretch and severely deform as they pull in opposite directions. This caveat is only applicable to bond-pad definition processes that act independently at each bond pad, such as in laser ablation techniques. The speed with which die can be placed depends not just on processing, but on machine accuracy and architecture, as well. For instance, machines with 10-micron placement accuracy run more complex software than less accurate machines. More complex algorithms increase computation time and result in slower axis motion. This is of particular concern when the placement machine is PWB equipment that has been reconfigured for package assembly. Such reconfigured machines typically possess work envelopes large enough to accommodate circuit boards. This, in turn, translates to much larger X and Y excursions than are necessary for flip-chip package substrates. A modified PWB placement machine is also likely to have a larger than necessary footprint, which is undesirable in a Class 10,000 package assembly cleanroom with its expensive floor space. Regardless of whether the place machine was originally designed for PWBs or was custom designed for flip-chip-in-package, if it incorporates integral fluxing capability one to two seconds will be added to processing time per die placement. This integral fluxing capability requires a trade-off against an up-stream fluxing system with its associated costs. The Future Looking ahead, as bump technology advances reduce bump size and pitch, machine accuracy and die recognition capability will become a larger factor in successfully placing flip chips at ever-increasing speeds. Even in the current cyclical market downturn, flip-chip-in-package continues to enjoy significant growth.
Estimates call for as much as an 87 percent growth rate over the long term. This will continue to drive innovation and improvement on the part of assemblers and equipment suppliers. During this phase of high growth it will often be necessary to use one piece of equipment for multiple products and to perform process optimization on the production line. Since flexibility and high-volume manufacturing features are often in competition on a given machine, managers in charge of capital equipment choices are in an increasingly challenging position. i
|
Copyright © 2001