- This article is filed under:
- » View All
Machine Vision-Based AOI in Electronic Assembly. . . .by Nello Zuech, President, Vision Systems International, Consultancy
by Nello Zuech, Contributing Editor - AIA Posted 03/04/2004
As far back as the early 80’s machine vision companies developed systems targeted at assembled board applications in the electronic industry. At the time the prevalent component attachment technique was ‘‘lead-thru-hole.’‘ The requirement was to verify that a lead had come through the appropriate hole on the board and that it was clinched in the right direction. The second requirement was to inspect the board following wave solder to make sure that the solder joint was a good one and that the solder did not cause any shorts or other conditions that might lead to board failure.
Just as the machine vision industry was responding to these applications with some degree of robustness, the approach to manufacturing the boards also changed. The electronic industry adopted surface mount packaged devices. The good news was that there were additional inspection requirements – inspect the solder paste before component placement and inspect after reflow solder. The bad news was that the old approaches to lead-thru-hole inspection were not longer valid. Again, the machine vision industry responded and systems emerged with reasonable robustness to address the applications of solder paste, pre-reflow/assembly inspection, post reflow and post wave.
Over the last couple of years we have seen the electronic industry embrace another new packaging approach – leadless chip carriers or bumped packages. These designs lend themselves to board design with increased functionality per square inch of real estate. At the same time even passive components have become much smaller. The net result is that tolerances for placement accuracy have become more demanding and the inspection requirements have become even more demanding. Simple part presence/absence logic is not longer sufficient. Both relative and absolute part positioning data are important as the density of pads increases. In the case of solder paste inspection, the increased interconnect density now requires 3D analysis of solder volume, not just presence and relative pad coverage. Being able to inspect interconnects that are underneath the body of the package as one has with leadless chip carrier such as BGA packages, is all but impossible with conventional optical approaches. The machine vision industry has responded with product offerings based on x-ray imaging.
Alongside of these requirements one also finds that both board warp and component body warp can be factors in achieving high first pass yields after soldering. The co-planarity of the interconnecting bumps or whatever is also critical. Again, the machine vision industry has responded with products that perform 3D-based analysis.
As consumers demand ever more functionality out of what had traditionally been single function devices, size becomes an issue forcing the semiconductor industry to increase density, which in turn results in increased interconnect density. Significantly, many of these devices (cell phones with picture taking and display, cell phones with PDA functionality, DVDs, etc.) addressing consumers result in high-volume, low-mix production lines. These are the very lines that can easily justify machine vision-based AOI systems and the very lines where these systems work the best and most cost-effectively.
We’ve come a long way -
Machine vision-based AOI systems for assembled board applications have evolved over the years thanks largely to the advances in the technologies that serve as the infrastructure for such systems. As personal computers have taken advantage of the increased power and speeds of the microprocessor, so, too, have machine vision systems. There is no question that performance robustness correlates to number of compute-intensive image enhancement and image processing algorithms one can perform within the time required. For the most part, these algorithms were around even back in the early days of machine vision. The challenge was that with the compute power of the time they would take forever to do and consequently could not be considered for any applications in manufacturing, which generally require decisions be made in seconds or less and not hours.
At the same time strides have been made in solid-state imagers. Today one finds high-resolution (greater than 1000 x 1000) imagers and color-based imagers embodied in cameras. Significantly, while some of this imager technology was even around 10 years ago, today cameras embodying these high-performance imagers cost under $10K and even under $5K as opposed to $40K - $50K ten years ago. Another issue ten years ago was the compute power to handle the additional data (more pixels and color) just was not available. The migration of the machine vision industry from analog cameras to digital cameras has also yielded performance gains significantly reducing digitizing noise and pixel sampling repeatability.
Along the way the cost of the optics compatible with the higher resolution and color cameras has also come down. Another major development that has lead to increased system robustness has been the development of the LED. Today their cost makes it possible to develop designs that result in optimized application-specific lighting arrangements. In the case of the assembled board applications this might include different directions of light as well as different colors. The lower cost of the ‘‘staging’‘ (lighting, optics, camera design) makes it possible to offer systems with multiple camera arrangements, often required to be able to consistently image around a component.
At the same time the user interfaces have become far friendlier, as the machine vision industry embraced Microsoft Windows. These friendlier man-machine interfaces not only make it easier for an operator to take ownership of the technology but also to ‘‘train’‘ systems on new board designs, interface to rework stations, etc.
What does one have to consider when investigating machine vision-based AOI?
The first thing that must be done is to determine the leading causes of failure within your own facility. Note I said ‘‘causes of failure’‘ and not failures by type alone. Can the causes be traced back to component co-planarity issues or board warpage issues? Can the causes be traced back to screen-printing issues – bridges, excess, insufficient, registration, uneven, etc.? Can the failures be traced back to board assembly component positioning issues – component correctness, orientation, placement registration on pad, etc.? Can the causes be traced back to post reflow solder – missing solder, too much solder, shorts, cracks, voids, flux issues, etc.? Can the causes be traced back to post wave solder – solder fillet geometry, shorts, voids, flux issues, etc.?
One issue to consider is that some reject conditions detected after screen paste and after component placements sometimes correct themselves during reflow solder. These are particularly thorny issues as generally the degree of failure condition determines whether or not reflow will have a chance at correcting. In the end it is really better to have detected all these conditions before there is even any probability the condition will ultimately lead to board failure.
Having documented your operation’s specific failure modes one can then made a decision as to where to deploy a machine vision system along the board assembly line.
What are other issues?
Some other issues largely relate to the specific assembly operation. Is it a high-volume, low-mix operation? As suggested these are the ones that most lend themselves to using machine vision-based AOI systems as ‘‘training’‘ these systems on a specific board design generally takes time. In recent years machine vision companies have developed ‘‘training’‘ approaches that have greatly reduced the amount of time required to train on a given board – from days to hours. Nevertheless, training still takes a long time. Even after training there usually is a requirement for ‘‘tweaking’‘ to optimize the performance of the system – reduce false rejects. This tweaking is generally based on the properties of the specific components that are used in a given board design, the actual board design itself and markings on the board, affects of system lighting on the components and boards, etc. In any event, tweaking requires running a reasonable number of boards. Hence, if one operates in a low-volume, high-mix scenario, while machine vision-based AOI systems are probably even more important, one must recognize the challenges.
As one can appreciate, system downtime while using the system to train it to inspect another board could be another concern. Some companies have responded to this concern by offering companion offline training systems, from which programs can ultimately be downloaded to the floor machine.
Another issue is that given all the variables associated with assembled board inspection (component appearances themselves, component markings, board appearances, board markings, etc.), these systems still experience false rejects. Consequently, systems that come integrated to rework stations are an advantage. The location of reject conditions can be directly downloaded to the rework station greatly improving the productivity of the rework operator as he is directed to specific sites of concerns and can, therefore, quickly make a judgment as to whether the condition is a reject or not and conduct the corresponding rework if it is.
Input for article
We canvassed input for this article from well over 20 companies understood to be selling machine vision-based AOI systems into the North American electronic industry. Regrettably not all responded. We have prepared the following three tables based on those who did respond. Table 1 reviews the specific applications that each company claims they address with one or more of their products. Table 2 looks at the specific features associated with their systems. The third table reviews the solder joint properties that can be detected. In most cases the caveat is that one has to provide samples reflecting your facilities typical results. Then testing will determine which of the specific solder joint concerns can reliably be detected.
Now more than ever it is realistic to consider machine vision-based AOI systems to inspect assembled printed circuit boards. Their performance is now far more rigorous than it has ever been. The result in improved yield and productivity will more than pay for the machines in short order. In buying a system it is important to understand your specific needs, which are a function of your own process experiences. It is also important to understand the features that you require in a system as not all systems come with the same features or capabilities. It is also important to understand that different applications along a board assembly line have different requirements and, therefore, more than one system may be required to gain the optimal performance for the specific application requirements.
The following tables are meant to provide an insight into what is available and what the respective companies claim they can do. The tables do not include all the companies offering these type products but only those that volunteered to provide input for this article. As noted far more companies were asked to contribute but did not have the courtesy to respond. Clearly those that did respond feel strongly about their capabilities and I leave it up to you to conclude why the other companies did not respond. My sense is that those that did respond are probably those that have the most competitive products today.
There are currently no comments for this article.
Leave a Comment:
All fields are required, but only your name and comment will be visible (email addresses are kept confidential). Comments are moderated and will not appear immediately. Please no link dropping, no keywords or domains as names; do not spam, and please do not advertise.