The Back-end Process: Step 2 - Die Placement
02/01/2001
BY STEVE SHERMER with KEVIN GAFFNEY
The expansive growth rate of flip-chip in packaging (FCIP) continues to task flip chip attach (FCA) equipment suppliers with new challenges. Die handling, with respect to size, bump metallurgy/size and pitch, and package handling are significant factors when determining which machine will best suit assembly needs. These needs are important to address because flip chip packaging, as one method of die placement, continues to enjoy significant growth in the IC industry. In 1999, FCIP units shipped totaled 149,000. FCIP forecasted shipments from 1999 through 2004 call for a combined average growth rate of 86.9 percent.1 This will continue to drive equipment suppliers in the flip chip attach machine market toward innovation and improvement.
Common Flip Chip Packages
Flip-chip chip scale packages (CSP) packages are generally processed in matrix strip formats, while high-performance parts are processed in carriers, or "boats." Conventional CSP strip formats contain from 8 to 100 units per strip, and CSP die sizes range from 2.5 to 11 mm square. High-performance die range in size from 11 to 26 mm square, and packages vary from 23 to 50 mm square.
Die Handling
Flip chip attach machines need to be able to handle die in various presentation formats. Waffle pack, tape feeder and wafer ring are among the most popular formats, and they each have their benefits and limitations.
Waffle pack: Waffle packs allow for populating packages with known good die (KGD). This reduces committing electrically questionable die to electrically good packages. Aspect ratio or the size of the die relative to the waffle pack cavity size should be tightly controlled to reduce die shift during handling. Ideally, the cavity size should be no more than 10 percent larger than the die size in the X and Y axes. A limiting condition of using waffle packs in high-volume assembly is the relatively few die that can be placed in either a 2 or 4-in. waffle pack. The larger the die, the fewer can be placed in the pack, which results in frequent machine reloading. Finally, using waffle packs initiates an extra process step, die sort/pick and place, upstream from the chip attach process.
 Figure 1. Sizing eject needles to die size. |
Tape feeder: Die presented to chip attach machines in tape feeders can benefit the chip attach process similarly to the waffle pack method. The use of tape feeders generally addresses KGD issues and benefits SMT machines that are configured for flip chip attach but are not capable of handling wafers. Again, tape-feeding die require the die-sort/pick-and-place process to be upstream from chip attach.
Wafer ring: Wafers presented to the machine mounted on tacky tape and a wafer ring are perhaps the most popular of die presentation formats, particularly in the conventional die attach process. This method generally is best-suited for high-volume assembly. It also requires strict attention to detail regarding die-eject optimization. Die-eject needles and die-eject cap sizes need to be selected carefully to develop a robust eject process. Other parameters, such as needle end height and eject speed, need to be characterized. If these parameters are not considered, die cracking, micro fracturing and miss-picking can occur.
Die Ejecting
To successfully eject die from wafer tape, it is critical to size the eject chuck (or cap) and space the eject needle correctly to the die size. As a general rule, the needle perimeter spacing should be no less than 80 percent of the die perimeter, and a center needle should always be present (Figure 1).
 Figure 2a. Initial eject pin position. |
Needle selection is another critical aspect of the eject process. Piercing needles can nick the die backside, which can induce cracking. Eject needles that have a radius on the tip shouldn't pierce the tape, thereby eliminating this problem. However, a dual-stage eject process is usually needed. Figure 2a illustrates the initial eject pin position. A short delay is added through the machine software to allow the tape to peel away from the die corners. When the tape in the area surrounding the pins remain in contact with the die, the needles can be raised to the programmed end-position, and the die-pick process can be completed (Figure 2b). Larger die require a longer delay for the tape to peel away from the edges.
Die Picking
Pick tools are selected according to tip material and should be sized for the die. Full-array flip chip die (where the die top surface is fully populated with bumps) require a compliant contact surface to maintain a vacuum. This is generally the case with large die (greater than 10 mm2).
 Figure 2b. Final eject pin position. |
Peripheral bumped die allow users to select hard-tipped tools that can alleviate die sticking during placement on smaller die. The material must be antistatic, so it doesn't damage the circuit.
Vision Systems
A key consideration for vision recognition is the wavelength of light used to view the fiducial. There is a vast array of materials used in IC packaging: ceramics, metals, polymers and semiconductors. Every material has unique reflection and transmission characteristics. Practically speaking, this translates to a wide range of contrast, brightness and glare considerations when trying to recognize a unique but periodic pattern on the wafer or substrate. In many cases, simply adjusting the mechanical settings on a camera (brightness, f-stop, incident light angle, aperture) are not enough to resolve the fiducial from its background. An actual change in the light wavelength, such as from white to red, may be needed to ensure accurate fiducial location. Figure 3 depicts how different LED colors with the same light settings affect die illumination and subsequent vision processing.
Fluxing Systems
The method by which the flip chip bumps and pads are fluxed can also be varied. Typical methods are drum or stamp fluxing, print fluxing, and flood or dispense fluxing. Again, each method has its pros and cons. Not only must one consider the material properties of the desired fluxing agent, but also the equipment investment and process time associated with each type of process. Additionally, the volume of flux used per bump and the total surface area over which the flux acts can have a significant effect on downstream process and ultimately product reliability. Even if a flux is touted as a "no-clean" flux, a poorly designed fluxing processes could nullify the "no-clean" nature of a flux.
Drum or stamp method: In this method, a small tray is located inside the FCA machine. Flux is put into the tray and a doctor blade is used to level the flux to the desired height. As each die is picked from the feed source, it moves to the flux tray where it is lowered into the flux tray, or "stamped," and then placed on the substrate. This method has the advantage of localizing the flux to the chip bumps, using simple equipment and being integrated into the FCA process. The primary disadvantage to this method is the accuracy of the flux height, because there are few simple and reliable integrated methods for measuring the thickness of the flux in the tray.
 Figure 3a. (left) Red LED illumination. Figure 3b. White LED illumination. (right) |
Print method: The print method of fluxing is a standard screen printing process. A stencil is brought to within a few mils of the substrate and a squeegee drives a large volume of flux across the stencil. Flux is thereby deposited on the substrate where holes exist in the stencil. This method can quickly flux a large number of die sites, but requires upstream equipment and processing. As with the stamp method, accurate measurements of flux volumes are difficult.
Dispense or flood method: Dispensing is perhaps the least complicated method to distribute flux, but it also may have the greatest negative effect on reliability. In this method, a liquid flux is dispensed at the center of each flip chip site. The flux then flows out across the substrate, fluxing each pad. The equipment for this method is a simple, pneumatic syringe, which can be incorporated directly into the FCA equipment. Process time can be minimal depending on how the sequence of placement steps is programmed and the parallel capabilities of the equipment. One major drawback to this method is that the volume of flux deposited greatly exceeds the theoretical minimum volume required to cover the bond pads. Additionally, the flux can interact with the flip chip system in unwanted ways. For example, the solder mask can absorb the flux that is volatilized in later steps and redeposit it on the die surface. Excess flux can crystallize during the reflow, causing surface contamination.
Die Placement Accuracy
A significant feature of flip chip attach is the ability of the flip chip device to "self-align" during bump reflow. As the solder bumps reach liquidus, the force generated by the liquid solder wetting the bond pads is enough to pull the device into perfect alignment with the bond pads. Because of this, the initial placement of the flip chip device has a slightly larger tolerance than one might initially expect. As a percentage of pad size, flip chip bumps can be as much as 25-percent misaligned from the exact center of the bond pad. The absolute value of this misalignment depends on pad and bump diameters, because a large bump has a larger placement tolerance. Most of today's FCA systems are capable of placement repeatability of ± 10 µm or better.
Productivity
Die placement rates are generally a product of machine accuracy and architecture, as well as processing steps. A high-accuracy machine (below 10 microns) relies on motion control profiles through machine software for a more precise and predictable placement point. These additional algorithms increase axis motion times and this is a concern depending on the active work area or envelope of the machine.
Many surface-mount machines have been reconfigured for flip chip attach capability. Typically, SMT machines have the capability to produce printed circuit boards (PCBs) that are large relative to microelectronic packaging. Large work envelopes equate to time consuming X-Y motions, which impact productivity. The PCB processing capability will also impact the machine footprint. Class 10,000 clean room assembly floor space is considerably more costly per square foot than SMT assembly floor space. Finally, a machine with integral fluxing capability will generally add 1 to 2 seconds per die placement. This additional processing time must be considered and gauged against an up-stream fluxing system and the associated costs.
Summary
As flip chip packaging continues to grow in popularity in the semiconductor industry, attach machines must address needs of silicon size, type and package construction. Such factors, along with machine footprint, productivity, process capability and types of die presentation format, are among the catalysts for selecting equipment. Looking ahead, machine accuracy and die recognition capabilities will be a factor to consider, as well, as bump technology advancements reduce bump size and pitch.
Reference
- Sandra Winkler, Advanced IC Packaging Markets and Trends, 4th ed., pp. 5-31, Electronic Trends Publications 2000.
STEVE SHERMER is director of operations, and KEVIN GAFFNEY is senior process engineer, flip chip development, at Amkor Technology. For more information, contact Steve Shermer, Amkor Technology, 1900 South Price Road, Chandler, AZ 85248; 480-821-2408 x5443; Fax: 480-821-5222; E-mail: [email protected].