Wire bonding experts have penned several books and articles related to wire bonding. Plasma treatment is generally cited as a method to influence the bonding process or the long-term dependability of the bond.
However, plasma experts have written very few articles related to the use of plasma in microelectronics packaging, and particularly, on the applications of plasma prior to wire bonding. This is mainly because in a world that is adapted to statistical process control, perhaps plasma tackles the “unknown,” which is “out of control.” Maybe, another reason is that a plasma process is elucidated by a relatively large number of parameters and it is not clear why one combination of settings works for a specific application, whereas a second application, to all similar intents and purposes, works best with an entirely different set of parameters. To sum up, plasma has something of the aura of a “black box” or “alchemy.”
Nevertheless, there are real expectations about how plasma can contribute to the performance of a wire bonding process as well as to the long-term dependability of the packaged device. Moreover, a definite logic has to be followed when developing a plasma process, even if it is not as stringent as some may like.
The benefits of plasma can be found in two different areas — (1) the wire bonding process itself, or the so-called “the statistics”, and (2) the long-term reliability of the device, or the so-called “device reliability.” Even though both areas are interrelated to a certain extent, it is the easiest to treat them separately for the sake of clarity.
Plasma and the “Statistics” of the Wire Bonding Process
Variety of Bonding Surfaces
At present, a majority of “first” bonds are still made to an aluminum metallization on a semiconductor device, and most “second” bonds are made to gold. However, this trend is now changing quickly, with the inclusion of copper (and other) wires, the latest metallizations on the devices, and also the latest metallizations on the substrates, or leadframes, they are bonded to. In this case, a source of diversity in the plasma process is faced — a 50 nm flash gold may be treated quite differently when compared to a 2 µm thick gold. Metal leadframes can be treated very differently than BGA substrates, and a nickel/palladium metallization on an IC will be treated in a very different way than aluminum.
Although rigorous process control in the semiconductor industry has a tendency to lead to a somewhat predictable metallization, the processes employed in the plating and printed circuit board industries essentially show more variability. Therefore, it was not unusual to source BGA substrates from three different vendors, all of whom were given the same kind of specification, and it was eventually found that they result in three varied sets of wire bond statistics.
Contamination on the Surface to be Bonded
In an ideal world, bonding would be performed on clean metal surfaces, with the potential exclusion of a thin layer of oxide, on the semiconductor device and also on the leadframe or substrate that are being bonded to it.
In reality, there are a number of sources of surface contamination that have an impact on the device reliability and the wire bond statistics. These include:
- Excessive oxide on the aluminum and copper surface
- Inorganics, mostly fluorine, on the integrated circuit pad, the origins of which lie in the wafer processes that occur prior to singulation
- “Atmospheric” contamination, which is present in the air and which tends to get deposited onto the bond pads. While this is mostly organic, traces of inorganics from the air are also usually present.
- Organic contamination coming from bleeding and outgassing of the die attach adhesive. These can be found on the leadframe or substrate and also on the integrated circuit metallization.
- “Diffusion” products that result from grain boundary migration of underlying metal layers: migration of nickel through palladium and flash gold migration are the best known diffusion products.
The presence of organics and oxides mainly influences wire bond statistics, while inorganic contamination commonly influences long-term reliability. However, there is an area of overlap.
In principle, plasma is capable of supplying an impeccably clean metal surface for wire bonding, but most often, it has been shown that atmospheric organics contaminate the surface of a metal within a matter of a few hours of plasma treatment. Such an effect can be mostly seen in a cleanroom that, despite being “particle free,” can have increased concentrations of organics coming from plastics, surface finishes, and people present in the cleanroom, and these organics can be concentrated through recirculation of cleanroom air.
Effects of Surface Contamination on Wire Bond Statistics
The problem with contaminants is that they simply cover the surface that needs to be bonded, thus inhibiting an excellent bond. The presence of contaminants is also “non-systemic”; they are unevenly distributed, appear and disappear, and are incredibly difficult to locate, identify, and measure. Although a bonding process can occur for hours or days without any issues, it can abruptly go out of control without any apparent reason. The most immediate solution is to turn up ultrasonic power (change of a controlled parameter) with all the associated consequences.
The major problems are:
- Non-stick on pad (NSOP)
- Reduction in the bonded area
- Reduction in average wire-pull strength
- Reduction in average shear strength
- Cratering and other similar damage coming from an extremely aggressive bond process
Different operators view lifts differently. For some operators, a lift that is within specification is believed to be satisfactory. However, in a majority of applications where some degree of reliability is required, a lift is definitely not an acceptable failure mode. The question is whether one would be able to capture some unevenly distributed “lifts” that are yet to reveal themselves as such. Reductions in average shear strength, pull strength, and bonded area are all indicators that when the bond is established, the contact area between the pad and bond wire will be only sub-optimal. This is due to the surface contaminant inhibiting the “welding” process.
In addition, it is known that the moment the bond is realized, the area of contact is never 100%; however, it increases in due time to provide a stronger bond via the process of grain boundary migration. This process will be obstructed by surface contamination, leading to a bond that fails to achieve its maximum potential strength. This is one area of concern where sub-optimal wire bond statistics will affect the reliability of a device. Failure to exploit the bond area will cause the bond to fail before reliability testing.
Influence of Plasma Treatment on Wire Bond Statistics
Turning the above around, it is easy to infer that the benefits of plasma include the following:
- Increase in the average wire-pull strength
- Increase in the average shear strength
- Increase in the bonded area
- Increase in process window for the bonding process
- Elimination or reduction of NSOPs
- Elimination or reduction of lifts
- Elimination or reduction of cratering and other damage coming from an extremely aggressive bond process
As described before, bonding surface is covered by contaminants, which tend to spoil a good bond; plasma can possibly act to eliminate non-systemic, short-term excursions from a process that is otherwise working quite stable with a high process capability. Conversely, even a highly capable process that is operating below its key potential will still provide sub-optimal device reliability. Therefore, an enhanced wire bond process that improves the bonded area and also reduces pad damage will invariably provide better reliability when compared to a sub-optimized process.
Plasma and Device Reliability
Failure Modes During Reliability Testing
At present, a considerable variety can be found in combinations of metallizations and bond wires, but despite this fact, the failure modes, which are noticed at the time of reliability testing, are bound to be common to all metallurgical systems. Here, monometallic systems create a “class within a class” in this case instead of an exception.
The variation from one metallurgical system to another is not so much the failure mode as the vulnerability to this failure mode — the time to failure. With certain simplification, it can be assumed that the following factors contribute to the major causes of failure:
- Compared to well-formed bonds, weakly formed bonds fail relatively faster
- Failures of “well-made bonds” that compromise the package reliability are usually the result of contaminant accelerated voiding (called Horsting Effect) and/or corrosion due to contaminants
- Well-formed and contaminant-free bonds will ultimately fail by Kirkendall voiding (polymetallic systems), but often much beyond the needed lifetime of the package
Framed in this fashion, the connection with the contamination of bond pad becomes instantly obvious. Considering that Nordson MARCH’s bonding process is suitably improved and that the company’s new metallizations are “in order” with respect to density, thickness, adhesion, surface topography, to name a few, the cause for a weakly created bond — that is, a bond with a low bond area consisting mostly of non-coalesced microbonds — is almost invariably organic contaminants on the bond pad surface. The reasons why these bonds fail relatively faster than one might anticipate are quite complicated and go much beyond the scope of this article. Conversely, “clean your bond pad” provides the solution to this problem — a solution that is pleasantly simple.
Intermetallics are formed by a majority of standard polymetallic systems. Undoubtedly, the combination of gold and aluminum is the most widely studied intermetallic. The development of an intermetallic is the first important step toward creating the bond, and intermetallics are formed at the time of the production procedure, burn-in, and utilization of the components. During the formation of intermetallics, one metal will diffuse into the other (with the formation of chemical compounds between the two metals), and eventually, one of the metals will be sapped by this process and, therefore, failure will always be the end result.
However, when the bond does not have contaminants, the “time to failure” will be usually much longer when compared to the design lifetime of the part and, therefore, the development of intermetallics and the ensuring Kirkendall voiding is not a problem in reality. The intermetallics are both electrically conductive and robust.
The problems will emerge when the bonded surfaces have contaminants and are filled with inorganic materials, particularly halogens. The presence of halogens can be attributed to wafer processes, environmental contamination, or they might be present in the molding compound utilized in the device package. Through a relatively complex mechanism called the Horsting Effect and, similar to Kirkendall voiding, a result of one metal diffusing into the other, the halogens are concentrated into zones at the metal interface where they considerably speed up the process of void formation, and yet further indicate a place at which corrosion will take place.
Halogens that are not integrated into the bond interface, but which make contact with the surface of the metal, mostly at the perimeter of micro-welds, have the potential to form electrochemical cells, which lead to quick corrosion. Compared to a well-formed bond, a weakly formed bond, with a low bond area and a large number of micro-welds, will possess a greater perimeter length and hence will be more susceptible to this mechanism of corrosion.
Among the three aforementioned failure mechanisms, the latter is the most predominant one because most of the integrated circuits available today utilize gold wire bonding onto an aluminum metallization. In spite of the complexity of the failure mechanism, the solution is once again agreeably simple — that is “clean your bond pad.”
Figure 1. Bench-top plasma cleaning system for surface activation and adhesion improvement (AP600).
Developing a Plasma Process
In conclusion, it is simple to consider that ensuring a clean surface to which plasma can be wire bonded will always lead to optimized device reliability and improved “wire bond statistics” (even if it is simply by eliminating non-systemic excursions). In the “ideal world,” a plasma would be preferably used which sputters all the organic and inorganic contaminants from Nordson MARCH’s bond pads and provides the impeccably clean bond pad leading to maximized “device reliability” and maximized “wire bond statistics.” However, this is where the real problem starts.
This solution — which would be a high-power argon direct plasma at comparatively low pressure — is sometimes employed. Moreover, it can be utilized on certain metal leadframes with power devices. The issues limiting its applicability are the effects of sputtering and overheating. Generally, metal leadframes are not sensitive to overheating, and the re-deposition of sputtered material poses an issue only when it is substantial and causes changes to device performance or surface resistivity — that is, leakage currents or modifications in device features.
When organic contamination is sputtered away, there is a maximum possible impact on wire bond statistics, but this impact is relatively lower than eliminating it using a “chemical plasma” like oxygen; yet, oxygen plasma is virtually ineffective against inorganic contaminants. Moreover, in a majority of the cases where oxidizable metals like copper or palladium are involved, oxygen is not indicated.
Either chemical plasma or sputtering plasma may be the first “polarity” to be taken into consideration when designing a plasma process. Nordson MARCH selects a process that combines both effects, that is, mixtures of oxygen and argon. With regards to oxidizable surfaces, this is generally a mixture of argon and hydrogen. This choice directly leads to the next polarity. While chemical plasmas work best at higher pressures (250–2000 mT), sputtering plasmas need low pressures (150–250 mT) to leverage the mean free path of the energetic ions that achieve the sputtering.
As a matter of fact, organic contamination is the most standard problem, and therefore, they usually begin with a higher pressure oxygen plasma and are inclined to make it more forceful (by increasing plasma power, lowering pressure, and adding argon) in case they happen to observe considerable amounts of inorganic contamination or if there is a sign for reducing the cycle time, and thus boost production throughput by applying a more powerful plasma process.
If the power is turned up excessively, especially with organic substrates, overheating can occur, and this would lead to extreme sputtering of flash gold, for instance. Therefore, based on the parts, one must prefer wire bond, organic substrate, leadframe, thin or thick gold, sensitive or robust components, large pitch or very fine pitch (variations in surface resistivity become more substantial at very fine pitch) so that the “levers” of the plasma process are exploited.
The aim is to elucidate a process window where the effect of cleaning can be increased as much as possible; however, the possible “downside” of an over-aggressive plasma can be prevented. Considering the number of variables involved — the nature of the parts as well as the plasma parameters that can be varied — a Design of Experiment (DoE) is generally regarded as a useful method. It must be remembered that DoEs, similar to SPC, are programmed to operate in a realm of predictable and controlled “cause and effect,” which is generally not the case with contamination.
Going into Production
From Process Development to Production
According to Nordson MARCH, contamination is as, “almost impossible to locate, identify, and quantify,” but in a majority of the cases, it does obey its own “statistics,” moving within particular ranges. The result of setting up a plasma process on a restricted sample of parts is that the “sample average” will not correspond with the “process average.” Therefore, when migrating from process development to production, the plasma process should be re-positioned to handle the “worst case” that can possibly be encountered in production.
In addition, whenever a plasma process is started, one would be handling a clean, normalized surface for the first time; this surface is essentially known and is “always the same.” This almost always requires the wire bond process to be re-centered.
Introducing a plasma process is invariably an anxious moment. There will always be someone who prefers to determine the surface both before and after the plasma treatment to make sure that it has done what it was meant to do. While setting up the process, several customers will attempt to figure out why their specific wire bonding process is “out of control” so, what is on the surface? Here, a better understanding of surface contamination can be obtained through X-ray photoelectron spectroscopy (XPS) alone, or in tandem with time-of-flight single ion mass spectrometry (TOF-SIMS).
Such a piece of information can prove useful in deciding the type of plasma process to use. However, both techniques are not feasible for production and QC, and both are also extremely costly to operate. In addition, today’s contaminant may not be tomorrow’s contaminant, and therefore, any QC routine that is set up to track contamination may end up determining the incorrect thing or overlook something which has emerged today but was not there yesterday. Nevertheless, the method followed in setting up a plasma process offers a level of security that is sufficient in a majority of the cases.
To clean the bond pads in the “worst case,” the plasma process is simply fixed. However, without actually asking about the existing contaminants, the plasma process is fixed to provide a clean and normalized surface. Such a surface will provide the wire bonding statistics that were established at the time of the process development. Ultimately, all operators will end up utilizing wire bonding statistics (which they were following nonetheless) as the indicator that the plasma process is executing what it is supposed to do. Plasma systems are designed to run an extremely repeatable process, with precise control of gas flow, RF power, process time, and process pressure.
The experience in practice — with a number of machines running plasma prior to wire bonding — is that if the wire bonding process is stable and the input meant for the process continues to be unaffected (nature of substrate/leadframe and semiconductor device), then the wire bonding statistics can be said to be extremely stable.
Compatibility and Concerns
The questions that often arise regarding the impacts of the plasma on the semiconductor devices are whether there are problems with charging, ESD, device parameter changes, and so on. Some of these, like overheating and sputtering with re-deposition, are the potential outcomes of choosing unsuitable process parameter settings and have been discussed in detail.
As a reference point, it can be observed that nearly all microprocessors and all memory devices are subjected to a direct plasma prior to wire bonding (or alternatively in wafer level package processing) without compromising reliability or function.
Plasma is both safe and effective if it is utilized properly. However, there are groups of semiconductor devices that are susceptible to direct plasmas. Devices with EEPROMs, open junctions, image sensors, and specific kinds of power devices cannot be exposed to direct plasma without promoting changes in performance or, in some cases, catastrophic damage. In some instances, a plasma that essentially removes the existence of energetic and charged particles, RF field, and the light which is integral in the plasma process can offer a solution. One such system is Nordson MARCH’s “ion-free plasma.”
It is the (potential) plasma user who is responsible for confirming that there is no compatibility issue with respect to the devices; however, the manufacturer of plasma equipment can help in making this evaluation.
Figure 2. High-Throughput Plasma Treatment System (FlexTRAK).
Introducing a suitable plasma process before wire bonding will invariably provide a cleaner surface to bond to. Improved device reliability, enhanced wire bond statistics, and the removal of excursions owing to non-systemic effects (arbitrary pollution of the surfaces to be bonded by uncontrolled factors) are the potential benefits.
John Maguire has been Nordson March’s Business Manager in Europe since 2006. He has a doctorate in Polymer Chemistry from the University of the South Bank (London, United Kingdom) and a degree in Chemistry from the University of Bath (United Kingdom). With that, Mr Maguire integrates a strong technical background with 30 years of experience in the European semiconductor, electronics, microelectronics, and PCB sectors.
Mr Maguire works together with specialized distributors in every region and segment, and his first strategic aim is to make sure that plasma processing comes out of its “Black Box” and takes its designated place as a well-interpreted and well-established solution to a host of challenges faced in the electronics industries and other market segments.
This information has been sourced, reviewed and adapted from materials provided by Nordson MARCH.
For more information on this source, please visit Nordson MARCH.