Wirebonding to multilevel metal
11/01/1997
Wirebonding to multilevel metal
Peter M. O`Neill, Dennis H. Eaton, Hewlett-Packard Co., Fort Collins, Colorado
Thomas Phua, Frankie Loh, Bernard Chin, Hewlett-Packard Co., Singapore
Changes in the materials and interfaces below bond pads have altered the requirements for wirebonding. Different pad structures were investigated by statistically designed experiments to optimize bond strength and reliability. The ball shear test continues to provide the best measure of bond integrity, including failure mode information.
Despite its age, thermosonic gold ball wirebonding to aluminum metallization is still the dominant chip-to-package connection process. Though other processes like solder bumping are becoming widely used, gold ball bonding is expected to be used well into the future, particularly for lower-pad-count chips, where it has a cost advantage over solder bumps. Thus, the IC industry needs to make gold wirebonding continue to work reliably with IC and package interconnects as they continue to develop.
Because thermosonic gold wirebonding is such an old and trusted technology, there is a tendency to take its reliable performance for granted. However, the materials and structures to which the bonds are being made have changed dramatically since thermosonic bonding was developed, and they are continuing to change in an effort to make circuits denser and faster.
To assure that the bonds are manufacturable and reliable, the effects of these changes on wirebonding must be understood and accounted for in the design of chip interconnect architectures, bonding pad structures, wirebonding setups, and packages. This paper deals only with the mechanical reliability of the bond to the IC and excludes discussion of the package bond.
Interconnect developments
Thermosonic gold wirebonding was first developed around 1970, when IC interconnect processes were much simpler. At that time, bonding pads consisted of one layer of thick (>1.0 ?m) aluminum doped with silicon on top of a thick, thermally grown field oxide. If passivation was even used, it was atmospheric pressure CVD silicon dioxide that was wet etched off the pad. The metallurgical properties and mechanics of such a structure are quite simple.
Processes currently in manufacturing and under development are far more complicated because they use more and different materials in more complex structures. Some interconnect developments that have implications for wirebonding follow.
Increasing number of metal layers. Circuit complexity and density require more conductor layers, and intermetal dielectric materials and deposition techniques are changing to accommodate this. Dice with three to five metal layers are common, with many more material interfaces where problems can occur, as well as many more possibilities for pad structures. Intermetal dielectrics are routinely planarized to reduce topographic variation, to make it easier to pattern fine-pitch metal lines and small contacts, and to stack more layers of metal.
Thinner metal layers. Finer-pitch wiring is easier to make when the metal is thinner, and the topography reduction of thinner metal makes it easier to stack multiple layers. Aluminum metal thickness affects the bond`s metallurgical, mechanical, and acoustic properties.
Cladding. Aluminum is clad on the top and bottom with layers of refractory metals and their compounds such as Ti, W, TiW, and TiN to improve electromigration and stress migration immunity, to serve as a contact barrier, and to promote adhesion. One of the consequences is that the Al no longer needs to be doped with Si. These changes also affect the bond`s metallurgical properties.
Contact plugs. Contacts are filled with metal to reduce topographic variation and contact resistance and to improve contact electromigration. The plug metal is often different from the conductor metal. W plugs with clad Al lines is the most common combination. This development affects the mechanics of the pad and allows for new pad structures.
 |
Figure 1. Alternate pad structures made possible by multilevel metal interconnect processing: a) full-metal-stack; b) waffle; c) sandwich; d) top-metal-only; and e) partial-metal-stack.
Thermosonic wirebonding
Before the effects of these interconnect developments on bonding can be evaluated, the materials science of Au to Al wirebonding must be understood. Thermosonic wirebonding uses temperature, ultrasonic agitation, and force to form Au/Al intermetallic compounds, of which there are five that range from Al rich to Au rich [1]. The Au-rich Au5Al2 phase is the most desirable because it is the strongest - stronger than either pure Al or pure Au.
In forming the intermetallic, care must be taken to avoid the formation of voids. Voids will weaken any of the compounds and are most commonly associated with the formation of Al-rich AuAl2 (known as "purple plague" because of its color and association with bond failures). Since voids are created by a high reaction rate causing local depletion of one of the reactants, the only way to assure that voiding will not happen in heat treatments after bonding is to terminate the reaction by consuming all of one reactant. Since the Au ball is much thicker than the Al-based metallization, the most stable bonds are made by forming intermetallic all the way through the Al, provided it adheres to what is underneath [2, 3].
The intermetallic should be uniform in thickness and cover as much of the area of contact between the ball and pad as possible. Intermetallic formation can be impeded by surface contamination, but this can be overcome by high temperature, force, time, and higher ultrasonic frequency.
Contrary to popular belief, the purpose of the ultrasonic energy is not to scrub the Au ball over the Al pad surface, but to soften the Al and Au by moving vacancies and interstitials (promoting lower force/temperature reaction). Actual movement of the ball over the pad is not desirable. Intermetallic does not only grow during bonding; in plastic packages, it grows during the post-mold cure. Soldering packages onto PC boards grows still more intermetallic.
Bonding applies considerable mechanical (static/normal capillary force and dynamic/parallel ultrasonic vibration) and thermal energy to the bond pad, so there is risk of cracking or delaminating the metal and dielectric layers beneath the pad. Bonding can even create a crater in the silicon substrate.
The risk of damage depends on the pad structure and material and the bonding parameters. It is commonly believed that damage can be reduced by thickening the pad Al to cushion the forces applied by the ball and capillary. If damage does occur during bonding, it might not be immediately detectable by the bond-strength tests that are commonly done for process control prior to encapsulation.
Reliability requirements and testing
There are really two issues with mechanical bond reliability: the strength achieved by the bond, and the strength required of the bond. The latter depends largely on the package. In a cavity package, the forces on the bond are primarily due to mechanical shock and thermal expansion of the wire relative to the package and chip. In a molded plastic package, forces are caused by stress in the molding operation and molding compound, and thermal expansion between the molding compound and the chip. Because the corners of the chip are farther from the neutral point than the sides, bond failures due to thermal stress will be more common at the corner pads. As a result of these mechanisms, cavity packages do not require bonds as strong as those in molded packages.
Bond strength can be seen as having two components, which are not necessarily dependent on each other:
Break strength - the ability to resist a one-time static force. This is measured directly with the common ball shear and wire pull tests [4, 5] and, indirectly, with liquid thermal shock (LTS).
Fatigue strength - the ability to withstand repeated applications of force, even with reversing direction, below the initial break strength. This can be affected by the propagation of cracks and other latent defects. It is also stressed by LTS.
Break strength should be measured and optimized independently of the package force, so as to give the bonds the best chance of holding up to the package force. If optimized bonds fail in encapsulated packages, then the package must be blamed for applying more force to the bonds than they can be expected to endure.
Pad structures
Multilevel interconnect process flows make new pad structures possible and potentially alter the bond`s metallurgical properties. The following structures (Fig. 1) were investigated:
Full-metal-stack. Older processes use pads consisting of all-metal layers with pad-sized contacts between them. As much aluminum as possible is provided to react with the Au ball, to cushion the mechanical forces of bonding, and to remove the fragile deposited dielectrics between the metal layers. This structure permits low resistance contact to the pad on any side on any layer. However, it can only be built in processes that permit large contacts. Many newer processes, particularly those with contact plugs, only allow minimum-size contacts.
 |
Figure 2. Statistically designed experiments are essential to optimize wirebonding. For the partial-metal-stack, this yielded a) a Pareto effects graph of average shear, and b) a response contour over the two strongest effects marked with maximum and range of experiment.
Waffle. The simplest perturbation of the full-metal-stack, when only minimum contacts are allowed, replaces the large contacts with offset arrays of minimum contacts in a waffle arrangement. A process with plugged contacts may only allow minimum contacts. However, there are concerns about the effects on the bond of the topographic variation created by these arrays of contacts when they aren`t filled with metal plugs.
Sandwich. The topographic variation of the waffle pad under the bond in an unplugged process can be eliminated by only placing contacts around the perimeter of the pad, provided they are numerous enough to handle the required current. Even with this sandwich design, the mechanical strength of alternating metal and dielectric layers under the bond is still a concern.
Top-metal-only. Using only the top layer of metal under the bond, connecting it to the lower layers with rings of contacts at its perimeter eliminates the concerns about contacts and alternating material layers under the bond. However, this structure may not have adequate Al thickness for good bond metallurgical properties and for mechanical cushioning. Though it has the lowest capacitance of any of the structures considered, it also has the least electromigration robustness.
Partial-metal-stack. Large contacts are more feasible to make in the upper layers than in the lower layers, so, if the top metal layer alone is not thick enough, it might be possible to use the top two metal layers with a pad-sized contact between them.
Experiments
A large number of experiments were conducted to determine which pad structures produced the strongest, most reliable bonds. A statistically designed experimental (SDE) [6] matrix was used, which varied the bonding parameters and sometimes the materials. The output of the SDE matrix was a response surface (usually shear force) on which the bond could be optimized.
The usual bonding parameters were force, ultrasonic power, temperature, and time, but the make of the bonding machine, its ultrasonic frequency, and the type of capillary were also significant. Al thickness and the fabrication site were the major material variables. Most of the work used 0.5-?m, three-layer metal CMOS chips assembled in plastic quad flat packs, but some other wafer processes and packages were used for comparison. The applied thermal stress was 235?C preconditioning (to emulate solder reflow) followed by LTS cycling 200 times between -55 and +125?C.
Bond quality was evaluated primarily by shear-strength testing, but also by pull-strength testing, and etching off the pad metal to inspect visually for damage to the lower dielectric and metal layers. Testing occurred both before encapsulation and stress and after stress and decapsulation. Precision cross-sectioning of the bonds gave tremendous insight into bond metallurgical properties.
 |
Figure 3. Optical cross section of top-metal-only pads showing a) a good bond, and b) a bad bond.
 |
Figure 4. SEM cross section of the center of a bond to a partial-metal-stack pad.
 |
Figure 5. SEM cross section of the center of a bond to a top-metal-only pad.
The optimization of partial-metal-stack bonding illustrates the usefulness of the SDE approach. This is a D-optimal design in four variables, where the average shear strength is the response. ECHIP [7] SDE software conveniently ranked the effects (Fig. 2a) and optimized the response (Fig. 2b). Because the strongest effect is an interaction (power ? time), this optimization
is difficult to do without the methodical approach of SDE.
 |
Figure 6. Shear force of a top-metal-only pad: after vs. before the thermal stress of molding.
Metallurgical insights
These experiments provided a wealth of information on the metallurgical properties of bonding. The most consistently strong bonds had uniform, thick Au/Al intermetallic (Fig. 3a), while weak bonds had spotty intermetallic (Fig. 3b). In the strongest bonds, all of the Al was consumed by the grown intermetallic alloy.
Intermetallic formation stopped at the TiW cladding between Al layers in both full- and partial-metal-stack structures. Though the cladding layer was distorted by the bonding operation, it never ruptured. Also, there was no deformation or cracking of the dielectric layers below the metal. Thus, even in stacked-metal pads (Fig. 4), only the Al in the top layer of metal is available for intermetallic formation.
Even when there is only one layer of metal in the pad, no damage to the underlying stack of deposited dielectrics is found where TEOS-precursor oxide is the interlevel dielectric (Fig. 5). Al thickness had an insignificant effect on bond strength, with a slight decrease in bond strength as Al thickness increased. The high shear forces measured on this type of pad, where all the Al was consumed by intermetallic, prove that intermetallic sticks to the TiW underlayer and does not cause any damage to it or the dielectric layers.
 |
Figure 7. Delamination of D2 from M2 in a sandwich pad. (Photo courtesy of Jeff Schoper, Hewlett-Packard Co.)
Additional heat treatment after bonding made the bonds stronger. Specifically, the post mold cure and thermal preconditioning stress increased the ball shear strengths in molded packages (Fig. 6), most likely due to additional intermetallic growth.
Pad structure evaluations
Full-metal-stack. Experiments proved that this is the trusted structure it was thought to be. The shear test force was high and tightly distributed around 70 gm, while the failure mode was overwhelmingly ball shear. The intermetallic stops at the first barrier layer, and there is virtually no change in strength after stress. No damage to the dielectric layers was visible after etching back.
Waffle. The waffle structure only makes sense when the contacts are plugged. Experimental bonding to a pad where the plug operation had been skipped was a disaster, due to the small amount of metal in contact with the Au ball on top of the ribs between the contacts, and cracking of the ribs themselves.
Sandwich. The sandwich pad was a fragile structure that was prone to delamination in shear testing and thermal cycling (Fig. 7). This fragility is easily explained by the pad being composed of alternating layers of soft, malleable metal and hard, brittle dielectric. Since stress degraded the shear strength by a third to a half (with corner pads being weaker than side pads), and since no cracks were found by etching back after bonding, while cracks were abundant after stress, the likely failure mode was ball-bond torque fatigue due to differences in thermal coefficients of expansion between the chip and the encapsulant.
Top-metal-only. The top-metal-only structure proved to be the strongest and most versatile of all the designs tested. The shear averages around 90 gm, is tightly distributed, and varies little from its global optimum, even when the Al thickness is constrained to 6000 ? or to a practical bonding temperature of 235?C. The delamination shear mode, which we do not believe to be a problem at high shear force [8, 9], is observed with this structure with thin Al, but it is eliminated at 9000 ?. The shear strength after molding and stress is highly correlated to, but greater than, the shear strength as bonded (Fig. 6), so this structure does not fatigue. It does not vary significantly from side to corner. The strongest bonds have uniform intermetallic that consumes all the Al.
Partial-metal-stack. The partial-metal-stack also works very well. Bonds to this pad shear at averages between 60-80 gm and do not fatigue. When the ball does not shear, it lifts, removing some of the pad metal in keeping with the shear pin effect. This is the major difference between these last two pad structures: the secondary shear mode for the partial-metal-stack is pad lifting (only metal lifting), while for the top-metal-only pad, it is pad delaminating (dielectric and metal lifting). The partial-metal-stack also has no dielectric damage and no pattern of strength with pad position.
Strength testing and failure mode insights
This work found that the best measure of bond strength is the ball shear test, which gives both force and mode information. The wire pull test is nearly useless except for qualitative mode information on very weak bonds.
Traditionally, a good bond has been defined as one where the shear test chisel only shears through the Au ball and does so with high force/unit weld area, but this view is based on experience with older processes with simple interconnect and pad structures. Research since 1990 [1, 2, 3, 8, 9] shows that some other shear modes that were thought to always indicate latent damage can be caused by the shear test alone. Thus, it is necessary to etch back in conjunction with ball shear to assess initial bond strength and latent defects.
In more advanced processes with many layers of metal and intermetal dielectric, other shear modes are possible if the stress at failure is high enough. The shear test is destructive, so the weakest material or interface will break. With so many materials and interfaces in an advanced interconnect structure, it is logical to require that all of them exceed a minimum shear strength. However, it is not necessary to specify that all shear failures occur at only one interface.
Based on current knowledge of interconnect materials` properties, several acceptable shear modes are plausible. Though Au-rich intermetallic is brittle, it has a higher shear strength than either Au or Al. Thus, if the shear chisel is set above the intermetallic layer, it may shear through the Au. If the intermetallic has not consumed all the Al, it is possible to shear through the Al when the chisel is set against the intermetallic (or even if it is set against the Au when the intermetallic formation is spotty, such that the Au can withstand more force than the bonded Al).
Pads composed of multiple metal layers always have lower Al layers separated by refractory barrier metals that do not react with Au. A lower, unbonded Al layer can serve as a "shear pin" to protect the other materials in the interconnect stack. If the intermetallic grows all the way through the Al, it can bond to what lies beneath it and carry greater shear stress to the lower layers. In shear testing this configuration, it is then possible to fracture an intermetal dielectric layer or to separate metal and dielectric layers, both of which would be classified as delamination.
If the metal and dielectric layers hold together well enough to transmit the shear force all the way down to the silicon substrate where its strength is exceeded, a crater will result, even if there was no prior damage from the bonding operation. Thus, delamination and cratering do not always imply latent damage, as has often been assumed, and can be acceptable shear modes.
Conclusion
Au to Al thermosonic wirebonding is being continually challenged by developments in IC interconnect. These developments must be understood and accounted for in the bonding pad structures and bonding processes. A bonding process is at least a four-variable problem that must be optimized for each new combination of pad structure, interconnect process, and bonding machine type. The most systematic way to perform this optimization is with a response surface statistically designed experiment.
Pad structures must be evaluated for both initial bond quality and bond reliability. The best measure of initial quality is the ball shear test, which gives both force and mode information. A good bond is one that shears with a consistently high force, regardless of its mode. Different pad structures will have different distributions of shear modes, depending on the relative strengths of the various material layers and their interfaces, and the location of any damage caused by the bonding operation. Bonding damage to the layers lying under the pad is best assessed by inspection of the pad after etching off the bond and pad metal.
The metallurgical properties of a bonding process and pad structure combination should also be evaluated by optical and SEM observation of bond cross sections, which can reveal hidden bonding damage. Au/Al intermetallic formation will stop on the refractory metals that commonly clad Al. Though the intermetallic will not penetrate the cladding, it will adhere strongly to it. Strong, stable bonds can be made by consuming all the Al with intermetallic; some researchers contend that this process makes the most stable bonds. Thinner Al, at least as thin as 6000 ?, makes for stronger, more reliable bonds when the metal is supported by high-quality dielectric layers such as TEOS-precursor oxides.
The mechanical reliability of a bond with good initial quality is determined by its strength relative to the force that will be applied to it in use, which is primarily due to thermal expansion differences between the chip and its package, and its ability to withstand the fatigue caused by repeated thermal cycling. LTS is the most severe form of thermal cycling. As with initial bonding, the ball shear test is the best way to evaluate the degradation of bond strength after stress. n
References
1. T. Ramsey, C. Alfaro, H. Dowell, "Metallurgy`s Part in Gold Ball Bonding: Information You Can Use to Assess Gold-to-Aluminum Bond Quality and Reliability," Semiconductor International, April 1991, pp. 98-102.
2. H. Ueno, "Influence of Al Film Thickness on Bondability of Au Wire to Al Pad," Materials Transactions, JIM, Vol. 33, No. 11, 1992, pp. 1046-1050.
3. H. Ueno, "Reliable Au Wire Bonding to Al/Ti/Al Pad," Japanese Journal of Applied Physics, Vol. 32, Part 1, No. 5A, May 1993, pp. 2157-2161.
4. "Test Methods for Destructive Shear Testing of Ball Bonds," ASTM F 1269-89, 1995.
5. "Standard Test Methods for Measuring Pull Strength of Microelectronic Wire Bonds," ASTM F 459-84, 1995.
6. Y-S. Chen, H. Fatemi, "Au Wire Bonding Evaluation by Fractional Factorial Designed Experiment," Proc. 1986 Internatl. Symposium on Microelectronics, pp. 76-82.
7. ECHIP: Software for Statistically Designed Experiments, ECHIP Corp., Yorklyn Rd., Hockessin DE 19808.
8. R.A. Clark, V. Lukatela, "Inadequacy of Current MIL-STD Wire Bond Certification Procedures Applied to Au-Al Ball Bonds," The International Journal of Microcircuits and Electronic Packaging, Vol. 15, No. 2, Second Quarter 1992, pp. 87-96.
9. G.V. Clatterbaugh, H.K. Charles, Jr., "The Effect of High-temperature Intermetallic Growth on Ball Shear-induced Cratering," Transactions on Components, Hybrids, and Manufacturing Technology, Vol. 13., No. 1, March 1990, pp. 167-175.
Peter M. O`Neill received his BS and MS degrees in electrical engineering from Purdue University in 1977 and 1978, respectively. He is currently a reliability engineer with Hewlett-Packard`s Integrated Circuit Business Division. 3404 E. Harmony Road, MS 64, Fort Collins, CO 80525; ph 970/229-4562, fax 970/229-3450, e-mail [email protected].
Dennis H. Eaton received his BS degree in physics from Harvey Mudd College, and his MS and PhD degrees in physics from the University of Illinois. He is reliability manager of the Integrated Circuit Business Division of Hewlett-Packard.
Thomas Phua received his BS degree in mechanical engineering from the University of Texas in 1989. He joined Hewlett-Packard in 1990, and is currently engineering section manager in the Integrated Circuit Business Division of Hewlett-Packard Singapore.
Frankie Loh received his BS degree in mechanical engineering from Texas A&M University in 1993. At Hewlett-Packard Singapore, his responsibilities include process enhancement and product development in the Integrated Circuit Business Division.
Bernard Chin received his BS degree in electrical engineering from the State University of New York at Buffalo. He is presently a failure analysis
eliability engineer at Hewlett-Packard Singapore.