A number of articles and books have been written about wire bonding by wire bonding experts. Very often, plasma treatment is mentioned as a means to impact the bonding process or the long-term reliability of the bond.
Considerably fewer, however, are the articles written by plasma experts on the applications of plasma in microelectronics packaging, and specifically, on the use of plasma before wire bonding. One reason for this, in a world which is used to statistical process control, may be that plasma deals with the “unknown”, that which is “out of control”.
An additional reason may be that a plasma process is described by quite a big number of parameters, and it is not always clear why a single combination of settings works for one application while a second application, to all intents and purposes analogous, works best with a totally different set of parameters. Plasma has something of the aura of “alchemy” or a “black box”.
There are, nevertheless, true expectations about what plasma can add both to the performance of a wire bonding process and to the long-term reliability of the packaged device. There is also a definite logic to be followed in creating a plasma process, even if it is not as rigid as some may like.
The advantages of plasma are to be found in two different areas: the wire bonding process itself, or what might be called “the statistics”, and the long-term reliability of the device, or what is referred to as “device reliability”. Although the two are to a certain extent interrelated, it is easiest, for the sake of clarity, to treat them independently.
Plasma and the “Statistics” of the Wire Bonding Process
Variety of Bonding Surfaces
Today, most “first” bonds are still maed to an aluminum metallization on a semiconductor device and a majority of “second” bonds to gold. This picture, however, is changing very fast with the addition of copper (and other) wires and of new metallizations on both the devices and the leadframes, or substrates, they are bonded to. Here, a source of diversity in the plasma process is encountered: flash gold of 50 nm may be treated very differently than 2 µm of thick gold. A nickel/palladium metallization on an IC will be treated differently to aluminum. Metal leadframes can be treated in a very different way than BGA substrates.
While stringent process control in the semiconductor sector does tend to result in a slightly predictable metallization, the processes in the printed circuit board and plating industries fundamentally show more variability. It is not odd to source BGA substrates from three vendors, all who were given the same specification, and discover that they result in three different sets of wire bond statistics.
Contamination on the Surface to be Bonded
In a perfect world, bonding would be carried out on clean metal surfaces, with the possible exclusion of a thin oxide layer, on both the semiconductor device and the substrate or leadframe which are being connected to it.
In practice, there are several sources of surface contamination which impact the wire bond statistics and the device reliability.
- Inorganics, mainly fluorine, on the IC pad whose origins lie in the wafer processes which happens before singulation.
- Organic contamination arising from outgassing and bleeding of the die attach adhesive. These can be found both on the IC metallization and on the substrate or leadframe.
- “Atmospheric” contamination which exists in the air and which deposits onto the bond pads. This is mostly organic, but traces of inorganics from the atmosphere are also often present.
- Products of “diffusion” resulting from grain boundary migration of underlying metal layers: best known are nickel migration through flash gold and palladium migration.
- Excessive oxide on the surface of copper and aluminum.
Wire bond statistics tend to be influenced primarily by the presence of oxides and organics while long term reliability is more commonly influenced by inorganic contamination. There is, however, an area of overlap.
While plasma can, in principle, supply a flawlessly clean metal surface for wire bonding, it has been frequently demonstrated that a metal surface becomes contaminated with atmospheric organics in a matter of a few hours of plasma treatment. This effect is mostly noticeable in a cleanroom which, although “particle free”, can have elevated concentrations of organics arising from the surface finishes, plastics, and people present and which can be concentrated through recirculation of cleanroom air.
Effects of Surface Contamination on Wire Bond Statistics
The issue with contaminants is just that they cover the surface that has to be bonded, preventing a good bond. Furthermore, the presence of contaminants is “non-systemic”; they appear and disappear, are unevenly distributed and nearly impossible to locate, detect and quantify. A bonding process can take place for hours or days without problems and suddenly go out of control for no obvious reason. The instant solution is mostly to turn up ultrasonic power (change of a controlled parameter) with all the accompanying consequences.
The main issues are:
- Non-Stick on Pad (NSOP)
- Decrease in average shear strength
- Decrease in average wire-pull strength
- Decrease in bonded area
- Cratering and other damage arising from a too aggressive bond process
Lifts are viewed differently by different operators. For certain operators, a lift which is within specification is considered as acceptable. For most applications requiring some degree of reliability, a lift is never an acceptable failure mode. The question is whether one will capture a few erratically distributed “lifts” which have not yet revealed themselves as such. Decreases in bonded area, pull strength and average shear strength are all indicators that when the bond is made, the area of contact between the bond wire and the pad will be sub-optimal.
This is because of the surface contaminant obstructing the process of “welding”. Furthermore, it is known that, at the moment the bond is achieved, the contact area is never 100%, but it increases with time to deliver a sturdier bond through the process of grain boundary migration. Surface contamination will interfere with this process, resulting in a bond which does not reach its highest potential strength. This is one area where sub-optimal wire bond statistics will impact device reliability. If the bond area is not exploited the bond will fail prior to reliability testing.
Influence of Plasma Treatment on Wire Bond Statistics
Turning the above around, it is easy to infer that the advantages of plasma include the following:
- Reduction or elimination of lifts
- Reduction or elimination of NSOPs
- Increase in bonded area
- Increase in average shear strength
- Increase in average wire-pull strength
- Reduction or elimination of cratering and other damage arising from a bond process that is too aggressive
- Increase in process window for the bonding process
As mentioned before, contaminants cover the bonding surface spoiling a good bond; it is quite possible that plasma will act to remove short-term, non-systemic excursions from a process that is otherwise functioning very stable with a high process capability. However, even an extremely capable process which is functioning below its vital potential will deliver sub-optimal device reliability. An improved wire bond process which maximizes bonded area and reduces pad damage will always deliver improved reliability than a sub-optimized process.
Plasma and Device Reliability
Failure Modes During Reliability Testing
Despite the substantial variety currently found in combinations of bond wires and metallizations, the failure modes which are observed during reliability testing are inclined to be common to all metallurgical systems. Monometallic systems form a “class within a class” in this case rather than an exception.
The difference from one metallurgical system to another is not so much the failure mode as the susceptibility to this failure mode: the time to failure.
With some simplification, it can be said that the predominant causes of failure involve the following:
- Weakly formed bonds fail a lot faster than well-formed bonds
- Well-formed bonds which are free of contaminants will eventually fail by Kirkendall voiding (polymetallic systems), but usually well beyond the required lifetime of the package
- Failures of “well-made bonds” which sacrifice the reliability of the package are typically the consequence of contaminant accelerated voiding (Horsting Effect) and/or corrosion because of contaminants
Framed in this way, the connection with bond pad contamination is readily apparent. Assuming that Nordson MARCH’s bonding process is well enhanced and that the company’s metallizations are “in order” in terms of thickness, surface topography, adhesion, density and so on, the cause of a weakly formed bond, i.e. a bond with a low bond area containing largely of non-coalesced microbonds, is nearly always organic contaminants on the surface of the bond pad. The reasons why such bonds fail a lot faster than one might expect are rather complex and go beyond the scope of this overview.
However, the solution to the issue - “clean your bond pad” - is pleasantly simple.
Most typical polymetallic systems form intermetallics. The most extensively studied is unquestionably the gold-aluminum combination. The formation of an intermetallic is the vital first step of forming the bond and intermetallics will be created during the production process, burn-in and use of the components. Given that intermetallic formation is fundamentally the diffusion of one metal into the other (with the development of chemical compounds between the two metals), and that ultimately one of them will be exhausted by this process, the end result is always failure.
When the bond is free of contaminants, however, the “time to failure” is typically much longer than the design lifetime of the part and so intermetallic development and the resultant Kirkendall voiding is not an issue in practice. The intermetallics are robust and electrically conductive.
The issues arise when the surfaces which are bonded are not free of contamination, but filled with inorganic materials, especially halogens. The halogens may be there as a result of environmental contamination, wafer processes, or may be present in the molding compound used in the device package. By a fairly complicated mechanism known as the Horsting Effect and, like Kirkendall voiding, a consequence of the diffusion of one metal into the other, the halogens become concentrated into zones at the metal interface where they significantly quicken the process of void formation, but also signify a site at which corrosion will occur.
Halogens which are not combined in the bond interface, but which touch the metal surface, mainly at the perimeter of micro-welds, are able to form electrochemical cells which cause rapid corrosion. A weakly formed bond, with a large number of micro-welds and low bond area will have a greater perimeter length than a well-formed bond and so be more vulnerable to this corrosion mechanism.
Of the three failure mechanisms stated above, this latter is the most predominant since the majority of integrated circuits today make use of gold wire bonding onto an aluminum metallization. Despite the intricacy of the failure mechanism, the solution is once again pleasantly simple: “clean your bond pad”.
Figure 1. Bench-top plasma cleaning system for surface activation and adhesion improvement (AP600).
Developing a Plasma Process
It has been concluded that it is easy to believe that by guaranteeing a clean surface to which one can wire bond plasma it will always result in enhanced “wire bond statistics” (even if it is only by removing non-systemic excursions) and improved device reliability. In the “ideal world”, one would use a plasma which sputters all the inorganic and organic contaminants from Nordson MARCH’s bond pads and delivers the flawlessly clean bond pad resulting in maximized “wire bond statistics” and maximized “device reliability”. This, however, is where the complexity begins.
This solution, which would be a high power argon direct plasma at relatively low pressure, is occasionally used. It can be used on some metal leadframes with power devices. The problems that restrict its applicability are the effects of overheating and sputtering. Metal leadframes are not usually vulnerable to overheating, and the re-deposition of sputtered material is only a problem when it is significant and results in an alteration in surface resistivity or device performance, i.e. leakage currents or changes in device features. Sputtering away organic contamination has the largest effect on wire bond statistics, however much slower than removing it in a “chemical plasma” such as oxygen, but oxygen plasma is almost ineffective against inorganic contaminants. Oxygen is also not indicated in most instances where oxidizable metals such as palladium or copper are involved.
Here is the first “polarity” that has to be considered when designing a plasma process: sputtering plasma or chemical plasma? In a majority of cases, Nordson MARCH chooses a process which integrates both effects, i.e. mixtures of argon and oxygen. For oxidizable surfaces, this is most usually a mixture of hydrogen and argon. This selection leads directly to the subsequent polarity. Chemical plasmas function best at higher pressures (250 – 2000 mT) while sputtering plasmas require low pressures (150 – 250 mT) to exploit the mean free path of the energetic ions that accomplish the sputtering.
In fact, since organic contamination is the most common problem, they normally start with a higher pressure oxygen plasma and tend to make it more forceful (by adding argon, lowering pressure and increasing plasma power) in case they notice significant quantities of inorganic contamination or if there is a sign that the cycle time can be reduced, and thus increase production throughput, by using a more forceful plasma process.
Turning up the power excessively, particularly with organic substrates, can cause overheating and can result in extreme sputtering, for example of flash gold. So based on the parts, one should want to wire bond, leadframe, or organic substrate, thick or thin gold, robust or sensitive components, very fine pitch or large pitch (at very fine pitch changes in surface resistivity become more significant) the “levers” of the plasma process are manipulated.
The objective is to describe a process window where the cleaning effect can be maximized, but the potential “downside” of an over-aggressive plasma can be avoided. Given the number of variables involved, both in the nature of the parts and the plasma parameters that can be diverse, a Design of Experiment (DoE) is often a useful approach. One has to keep in mind that DoEs, like SPC, are programmed to work in a world of controlled and predictable “cause and effect,” which contamination normally is not.
Going into Production
From Process Development to Production
While Nordson MARCH has described contamination as “almost impossible to locate, identify and quantify” it does tend to obey its own “statistics” in most cases, moving within specific ranges. The consequence of setting up a plasma process on a limited sample of parts is understandably that the “sample average” will not equate to the “process average”. Consequently, when moving from process development to production it is usually necessary to re-position the plasma process to handle the “worst case” that might be found in production.
Furthermore, when starting a plasma process one will, for the first time, be dealing with a normalized, clean surface, a surface which fundamentally is known and “always the same”. This virtually always requires a re-centering of the wire bond process.
It is always a nervous moment when talking about introducing a plasma process. There's always someone who wants to measure the surface before and after plasma to be sure that it has done what it should. It can be done. In setting up the process, many customers will try to understand why their wire bonding process is “out of control”; what's on the surface? X-Ray photoelectron spectroscopy (XPS) alone or in combination with Time-of-Flight Single Ion Mass Spectrometry (TOF-SIMS) can provide insights into what is on the surface.
This information can also be practical in deciding which plasma process to use. Neither method, however, is practicable for production QC and both are very costly to operate. Furthermore, today’s contaminant may not be tomorrow’s contaminant so any QC routine set up to monitor contamination may end up measuring the wrong thing or miss something which has appeared today, but was not there yesterday. The approach which is followed in setting up a plasma process nonetheless provides a level of security which in most cases is adequate.
The plasma process is fixed to clean the bond pads in the “worst case”. Without asking precisely what the contaminants are today, the plasma process is fixed to deliver a normalized, clean surface. This surface will deliver the wire bonding statistics that were determined during process development. Eventually, all operators end up using wire bonding statistics (which they were following anyway) as the indicator that the plasma process is performing what it should do. Plasma systems are engineered to run a very repeatable process, with accurate control of RF power, gas flow, process pressure and process time.
The experience in practice, with several machines running plasma before wire bonding, is that if the input to the process remains unaffected (nature of semiconductor device and leadframe/substrate) and the wire bonding process is stable, then the wire bonding statistics are very stable.
Compatibility and Concerns
Questions frequently arise about the effects of the plasma on the semiconductor devices; are there issues with ESD, charging, device parameter changes, etc? Some of these, such as sputtering and overheating with redeposition, are the possible consequences of selecting inappropriate process parameter settings and have been talked about. As a reference point, it may be noted that almost every microprocessor and every memory device passes through a direct plasma before wire bonding (or in wafer level package processing) without sacrificing function or reliability.
Used properly, plasma is safe and effective. There are, however, classes of semiconductor devices which are sensitive to direct plasmas. Devices with image sensors, open junctions, EEPROMs and certain types of power devices cannot be uncovered to direct plasma without causing performance alterations or, in certain cases, disastrous damage. In some cases, a plasma which fundamentally eliminates the RF field, the existence of charged and energetic particles, and the light which is inherent in the plasma process, can provide a solution. Nordson MARCH’s “ion-free plasma” is one such system.
It will always be the responsibility of a (potential) plasma user to confirm that there is no compatibility problem with the devices, but the plasma equipment manufacturer can assist in making this assessment.
Figure 2. High Throughput Plasma Treatment System (FlexTRAK).
The introduction of an appropriate plasma process prior to wire bonding will always deliver a cleaner surface to bond to. Potential advantages are better wire bond statistics, enhanced device reliability and the elimination of excursions due to non-systemic influences (random pollution of the surfaces to be bonded by unrestrained factors).
Since 2006, John Maguire has been the Business Manager in Europe for Nordson March. He has a degree in Chemistry from the University of Bath (UK) and a doctorate in Polymer Chemistry from the University of the South Bank (London, UK). With that Mr. Maguire combines a solid technical background with three decades of experience in the European PCB, microelectronics, electronics and semiconductor industries. Working along with specialized distributors in each region and segment, Mr. Maguire’s first strategic objective is to guarantee that plasma processing emerges from its “Black Box” and takes its place as a well-established and well understood solution to many challenges in the electronics sector 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.