Bonding BGA Packages Requires Integrated Solutions

		At a Glance
	With high I/O counts, fine-pitch wirebonds, long wire lengths and laminate substrates, wirebonding in BGAs is difficult. Finding a manufacturable solution for SBGAs and PBGAs requires an integrated study of bonding wire, wire capillaries and bonding parameters.

The ball-grid array (BGA) package is expected to be the fastest-growing semiconductor package during the next five years. Estimated compound annual growth rate (CAGR) for its adoption over the 1996-2001 period is 70%.¹ Development of field support and applications has lagged, however, because of the speed at which this package has been accepted.

Because of their intrinsic design, BGAs are technically complex to bond. BGAs are designed for high I/O counts; 225 to 500 leads are common. They demand fine-pitch (<90 µm) wirebonding and require long wire lengths, straight loops and small first and second bond areas. Wire lengths of 6 mm  are common, and loop shapes must be controlled to provide clearance over ground and power busses located close to the lead tips (Fig. 1).

Fine-pitch bonding requires the use of a bottle-necked capillary, with a small tip diameter. The small capillary tip and the narrow lead width result in a small cross section for the crescent bond, making bond strength an important concern.

The BGA structure is based on laminate technology. The low glass transition temperature (T_g) of the laminate material requires bonding at lower than normal temperatures; this generally has a negative effect on wirebond strength and reliability.

Machines and materials

Experiments were conducted using two different device types. Super BGAs (SBGAs) were used, each containing an Amkor test die with 65 µm wide bond pads on a 75 µm pitch. Plastic BGA (PBGA) devices were also used, each containing a Kulicke & Soffa test die with 65 µm wide bond pads on an 80 µm pitch.

The Kulicke & Soffa Model 1488 plus ball bonder used had an adaptive ultrasonic power generator (USG) system. Its 120 kHz transducer driver adjusts ultrasonic output in real time to adapt to material variations. The adaptive USG system was chosen because it provided the necessary low-temperature bonding performance.

The wire used in this study was 25 µm (1.0 mil) AW14 designed and recommended for fine-pitch, low-medium height, long-loop applications. The wire lengths used ranged from 6.5 mm to 8 mm, and the wires were not molded.

The devices were cleaned using an argon hydrogen plasma. One advantage of the plasma cleaning method was the ability to handle parts in magazines, at eight magazines per cleaning cycle.

Description of work

The design of experiment (DOE) technique used in this bonding application optimized second bond parameters first, followed by first bond and then looping.² This sequence was used because the quality of first bond is significantly affected by the quality of second bond.

Fig. 2. The face angle of the wire capillary is optimized for second bond strength on Au-plated SBGA packages.

Optimizing second bond first assures a good subsequent tail bond. The tail bond is the weld between the short piece of wire, which subsequently will be used to form the next ball, and the substrate. A good tail bond assures that the correct volume of wire will be available to form a uniform ball. 

To optimize the second bond, pull tests were done with the hook located as close to the second bond as possible. For SBGA devices, the hook was aligned with the inner ground bar ring; for PBGA devices, the hook was placed over the ground bar. This approach applied most of the force to the second bond, providing a more sensitive measurement of the second bond strength as well as a more sensitive method of optimizing the crescent weld and tail bond areas.

After second bond optimization, first bond optimization experiments were conducted. The most important responses for first bond optimization are shear strength per unit area and ball diameter. Ball size was a limiting factor, since it must be small enough to meet 100% on-pad placement requirements. An average ball size of 57 µm was selected for the 65 µm bond pads used. The minimum acceptable shear strength was 7 mg/µm² (4.5 g/mil²). Optimization of looping completed the study.

Fig. 3. A capillary specifically designed for high frequency enhances bond strength.

In this series of experiments, the effects of plasma cleaning were of considerable interest. DOE techniques were used to optimize the cleaning process. Pull testing, with the hook located near the second bond, was the most important response used in the plasma cleaning optimization.

Capillary considerations

For smaller bond pad widths and pitches, the role of capillary geometry is more important.³ Capillaries must be dimensioned to avoid interference between the capillary and adjacent wires and bonds, in addition to being optimized for bond quality. Solid, model-based software was used to simplify the selection of capillaries. For the SBGA devices having 57 µm ball sizes on a 75 µm pitch, the software recommended a capillary design with a 46 µm chamber diameter and a 97 µm tip diameter.

For BGA devices, the second bond is more difficult than the first bond because of the low bonding temperature, the short bond length and the contamination of the laminate surfaces. The low T_g of the laminate substrate material is the reason for the low bonding temperature, which is typically less than or equal to the T_g of the laminate. The fine pitch, along with the diameter of the capillary, lead to the short bond length. For these reasons, the face angle of the capillary had to be optimized.

Capillaries with a range of face angles were studied to determine the best choice for bonding on the BGA surface. Though increasing the face angle is normally recommended for smaller tip capillaries, the largest face angle does not always produce optimum bonds. The best procedure is to test a series of face angles to determine the optimum choice for a specific application. All the capillaries selected for the study had the same dimensions, other than face angle, so that the face angle response could be studied systematically.

Second bond optimization

A 17-trial quadratic response surface experiment using four capillary designs was carried out on both electroless and electrolytic Au-plated SBGA and PBGA devices. The same experiment was run with each capillary and both plating types. The capillaries had a 97 µm tip diameter designed for 75 µm pitch bonding. The experiments showed that the 15 face performed best on both plating types (Fig. 2).

During the course of this study, a new proprietary capillary design, significantly different from previous designs, was included. The new capillary was designed specifically for high-frequency bonding applications. Comparative tests showed that the new capillary provided higher second bond pull strength. There is less difference between the averages for the Y and X axes wires, a wider process window and a significantly better failure mode distribution. Figure 3 shows a comparison of the average pull strength, with the wires pulled at second bond.

Au plating

Fig. 4. Plasma cleaning of bonding surfaces can improve bond strength. The argon flow and arc current of the process are optimized for an H2 flow of 25 sccm.

Fig. 5. Delay between plasma clean and bonding has little effect on bond strength.

Two plating methods are commonly used for top layer metalization on BGA devices. Electroless plating is often the best method for fine-pitch applications, because it does not require a continuous electrical connection for plating to occur as is required for an electrolytic cell. An electroless-plated surface, however, is considered to be a more difficult surface to wirebond than that of electrolytic plating. Electrolytic plating is normally less expensive and has more uniform surface properties.⁴

For each of the capillaries studied, a full set of samples was replicated for both electroless and electrolytic plating. Figure 2 also illustrates these data. There were no significant differences between the bond strengths that were due to plating. It should be noted that all the parts in this study were cleaned using the optimized plasma process, and that the sample size was small.

Plasma cleaning experiments

The devices in this study were all plasma cleaned. Attempts to bond devices without an initial plasma cleaning resulted in a very large number of opens; poor tail bond welds led to a failure to form a good ball for subsequent first bonds. An experiment was performed on the plasma cleaner using cleaning time, flow rates of hydrogen and argon gases, arc current and the sweep angle of the arc filament as variables.

An initial screening experiment was performed to identify the most significant cleaning factors affecting the second bond pull strength results. No significant improvements were gained by cleaning for longer than 3 min. To improve the resolution and add confidence, the initial linear experiment was augmented into a quadratic design. Figure 4 is a contour plot of the two most significant factors in the quadratic designed experiment. The best second bond pull test results were obtained at lower argon gas flow and higher arc current. All subsequent experiments were plasma cleaned with these optimized settings.

It is common practice to strictly control the time interval between plasma cleaning and wirebonding. In some cases this limit is as low as two hours. An experiment was run to investigate this effect. Electrolytically plated PBGA devices were cleaned and bonded with a range of delay times between cleaning and bonding. Pull strength near second bond and ball shear strength were both measured, and no significant effect was observed. However, the sample size was small. Figure 5 shows the results of the time delay experiment.

First bond optimization

Optimizing the ball bond is largely a matter of working with design and process models to determine which targets to set. Once the targets are determined, response surface modeling of the process is used to map the process and predict the parameters that will meet the targeted responses with optimum strength bonds.

After capillary dimensions are specified, it is necessary to run an experiment to model the process. For fine-pitch bonding it is necessary to simultaneously optimize several responses. The shear strength must be optimized at a specific ball diameter. In addition, ball height and shape are also important quality criteria. ECHIP software was used to perform this optimization.

Figure 6 illustrates the simultaneous optimization of both the bond diameter and the ball shear strength per unit area to get the best strength at the design specified ball diameter. The most significant process control variables are ball size ratio (BSR, the ratio of the expected free air ball diameter to the input wire diameter) and ultrasonic power. BSR affects the diameter of the free air ball, the ball that is formed before bonding. Ultrasonic power affects the bonding deformation. In this case, there was a very large process window for producing bonds of acceptable strength. Producing a maximum bond strength at the targeted bond diameter requires a well-characterized process and the correct specification of capillary feature dimensions.

Bond looping

The BGA device has presented a whole new set of looping issues. The device requires long, low, straight loops. There are normally two rings within the second bond periphery to distribute power and ground, and the wires must have a significant separation from them to prevent shorting.

The looping problems were solved using new software algorithms that controlled the wire shape by controlling the capillary motion. Figure 7 shows the shape of the wire. The wire has a steep angle going into the second bond, providing good standoff distance between the wire and the distribution rings.

A wire like this, with a long flat portion parallel to the die surface that descends sharply to the second bond, has been shown to have improved thermal cycling reliability. This occurs because as it cycles, the wire is able to flex at the outer bend, not in the heat affected zone (HAZ) above the ball.⁵

Figure 8 is a close-up of wires bonded to the pads, with the distribution ring shown. The additional bend in the wire near second bond is required to provide the necessary clearance. This shape is especially important for SBGAs, because the die surface can be lower than the height of the leads (Fig. 1).

Conclusion

Fig. 7. The required loop shape is flat over the die, with a sharp bend down to the second bond.

Fig. 8. An extra bend is required near the second bond to provide clearance over the distribution rings.

Experiments have verified that high-yield, fine-pitch BGA processes are feasible with modern, high-speed automatic wirebonding equipment. High-volume, robust processes have been established for devices with pitch as low as 70 µm. It was demonstrated that optimized capillary designs improve the bond quality and provide stronger bonds, even when the bonds are small in size. New capillary designs are being introduced that improve the robustness and increase the process capability of high-frequency bonding by providing a larger process window.

Developing a successful technology of this nature requires considerable understanding of the bonding process, the materials and the equipment. Collaborative efforts between equipment and materials suppliers and the end-use customer or assembly facility greatly enhance the product/process development cycle, increasing the speed of development and the quality of the end result.

Cleaning BGA devices prior to bonding is another requirement for high-yield BGA manufacturing. The experiments demonstrated that once cleaned optimally, the devices were capable of a robust process.

This article was adapted from a paper presented at IMAPS '97 in Philadelphia, Pa., Oct. 12-16, 1997.

References

1. Worldwide Package Demand by Package Type, VLSI Research Inc., 1996.

2. M. Sheaffer, L. Levine, "How to Optimize and Control the Wire Bonding Process: Parts I & II," Solid State Technology, Nov. 1990, pp. 119-123.

3. I. Hadar, "Tradeoffs in Fine Pitch Wire Bonding Technology," 11th European Microelectronics Conference, May 1997, pp. 211-218.

4. H. Charles, et al., "Wire Bonding on Various Multichip Module Substrates and Metallurgies," 1997 Electronic Components and Technology Conference, pp. 670-675.

5. P. Hoffman, et al., "Development of a High Performance TQFP Package," 1994 Electronic Components and Technology Conference, pp. 57-62.