Detailed vision—Not all black or white
The images analyzed in inspection systems are moving away from simple high-contrast silhouettes and into complex gray-scale images.
Irene Leszkowicz, Matrox Imaging -- Test & Measurement World, 12/1/2004
Advanced automated optical inspection (AOI) and automated x-ray inspection (AXI) systems used as part of a production line are extremely adept at functional pass/fail, or flag, testing of dense complex assemblies of surface-mount components. If you need more than a pass/fail result, though, perhaps to quantify factors that may impact reliability, you may have to inspect components offline in a QA lab or at a development bench. Must you replicate the factory-floor vision system costing hundreds of thousands of dollars, or can you use something much more modest? Will you need the same image-analysis algorithms, or will a different set do?
| Accuracy considerations 10x rule |
| Read all of the Dec. 2004/Jan. 2005 features:
The Best in Test DSM fault models Ethernet in the First Mile could simplify networks and reduce costs. And most of the test infrastructure is already in place. |
We have all witnessed the physical changes in electronic assemblies over the last 10 years. Increased densities came with surface-mount technology as through-hole components were replaced by quad flat packs (QFPs) bordered with very fine-pitch J leads. In the quest for more interconnects in a given footprint, technologies have evolved with connections on the underside of the package, hidden from the optical camera eye. These include ball-grid arrays (BGA, µBGA, PBGA) as well as flip chips and other "bumped" chip-scale packages (CSPs).
That same 10-year period also represents the typical amount of time it takes for a new software technology to progress from research labs and technical journals through high-end customer products and finally to more general commercial availability. Image-analysis software technology is nearing the end of such a 10-year stretch and is undergoing a final developmental stage: optimization, which involves converting critical portions into specialized processor instruction sets (MMX, SSE, or co-processor code) or into hardware-accelerated implementations.
Image-analysis software represents a key component of today's mature, sophisticated AOI systems. To inspect today's components and connections, advanced AOI systems bring to bear sophisticated motion control that positions programmable arrays of hundreds of vari-directional LEDs and up to five cameras, all with micron precision.
The cameras can provide a top view and oblique angle views of all four sides of any component. Ranging techniques measure any warpage in a board, for example, ensuring that optimal focus distances can be maintained at any location during a scan over the board. Accurate positioning of LEDs eliminates unwanted shadows and reflections.
Because AOI systems tightly control the illumination direction and camera angle, most assembly and solder defects can be detected through simple image-analysis operations performed in small "inspection windows" set up for each pin, gap, or joint. Table 1 lists some common defects and typical AOI approaches for detecting them.
These simple analysis techniques remain popular because they have been used for a long time and are trusted. Simple analyses can also be performed quickly, a trait that is essential for inspections of printed circuit boards.
Consider that a reliable inspection of very small components (0402 and 0201) and very fine-pitch QFPs requires an AOI system to capture at least 10 pixels across each feature. For large board sizes, therefore, a lot of pixels need to be captured and processed. Some manufacturers of AOI systems claim to collect and process over 100 Mbytes of image data per board. It becomes evident that speed of analysis is important. Applying hardware solutions to the problem—wider buses, higher bandwidths and multiple processors—can more readily speed up simple techniques than complex ones.
Catching solder-paste defectsBy placing an AOI system after the soldering stage, a manufacturer can identify defects that would render a board nonfunctional. Some sources, in fact, suggest that 80% of defects are solder related (Ref. 1).
Other companies are looking to improve yield (percentage of boards that pass post-solder AOI and in-circuit testing [ICT] without rework) by trying to identify potential problems before the solder stage—that is, just after solder-paste deposition. They point to the 10x rule, which encourages defect detection as early as possible. Consequently, they use solder-paste inspection systems for early defect detection. (See "10x rule ," below.)
Defects have plenty of opportunities to creep into the highly mechanical process of forcing a sticky substance through the thousands of tiny holes in a silkscreen. Failure analysis has indicated that the critical parameters in deposited solder paste are its height and volume. An overhead camera can determine the area of the paste spot, and you can derive a height and volume measurement from the area based on assumptions about the shape of the spot, such as that it is a perfect hemisphere.
Area measurementsTypically, area calculations employ a standard technique, classifying pixels as either "background" or "object" based on their brightness (thresholding) and then counting the connected "object" pixels in a blob. The discretization errors of thresholding can be significant, especially in the presence of nonuniform lighting and at lower magnifications (see "Accuracy considerations ," below). Such errors in estimating volume can be overcome with more sophisticated edge-detection algorithms that have subpixel accuracy. Even with controlled lighting, sophisticated edge detection may allow a lower magnification to be used, which decreases the number of images required to cover the entire board and increases throughput.
An alternate inspection approach is to move beyond the 2-D image. Volume determination and other inspections (like determining the shapes of fillets on solder joints, the bumps on CSPs, and the balls on BGAs) have given rise to a number of "profilometry" systems that measure the height as well as the shape of objects protruding from a flat substrate.
 |
| Figure 1. Structured light in the form of a fan of laser light reflects from a flat surface as a straight line. However, viewed from a camera at an oblique angle, any raised surfaces cause portions of the line to appear to be deflected by an amount proportional to their height. |
Typically, the light scans over the entire board to build up a height map line by line. Frame grabbers with some onboard processing can do these calculations on the incoming data and can generate the height map in real-time. Profilometry approaches are sometimes classified as "21/2-D," because they only look at the upper surfaces of the components, rather than forming a complete 3-D representation of the board and components. Variations on this technique use multiple cameras to overcome problems of concavity and of tall components casting shadows. Interference patterns between two structured white light sources are sometimes used instead of laser illumination.
Of course, many PCBs today include solder connections invisible to any optical technique. Part 2 of this article (scheduled for April 2005) will examine the automated x-ray techniques that you can apply to inspect such connections.
| Author Information |
| Irene Leszkowicz is the Group Leader for Interactive Imaging Software at Matrox Imaging, and has over 20 years of experience with the company. She holds a BEng in Electrical Engineering from McGill University and an MASc from the University of Toronto, and she is also an Adjunct Professor in Electrical and Computer Engineering at McGill University in Montreal. ileszkow@matrox.com. |
| REFERENCE |
|
|
