What You See Is What You’ve Got

Automated optical inspection (AOI) identifies faults at critical stages in the production process before they become expensive problems later. It does so without the periodic fatigue and subjectivity of a human inspector and at a much faster rate. These attributes—the capability to find a wide range of faults accurately, quickly, and repeatedly—suit AOI well to several electronics manufacturing applications.

Applying AOI to PCB photomask inspection ensures that the original photographic image of each PCB layer is correct before it is transferred to a copper-clad board for etching. After a board has been produced, the etched conductor pattern can be verified even for line widths down to 0.003². Optical and electrical test often are combined to obtain more complete fault coverage, especially for multilayer PCBs.

Flat-panel displays also benefit from AOI. In this case, errors can be due to imperfections in any of the several optical layers comprising the panel. AOI systems can assign faults to the correct layer or layers and classify defects to aid efficient correction. Flat panels are inspected by line-scan cameras that can scan at 160 MB/s or higher.

AOI systems can be calibrated to measure dimensions, and with special software, they can recognize characters. Semiconductor manufacturers need to determine that they have correctly and clearly printed descriptive information on their ICs.

For example, Figure 1 shows the display of an inspection system developed by Gorman Co. Ltd. in Taiwan. The system checks for ink splashes outside the character area, verifies that the characters are correct and not blurred, and confirms the printed pattern orientation relative to pin 1 on the package.

In another semiconductor-related application, an AOI system inspects the gold bump contacts on driver ICs for liquid-crystal displays. The Electroglas QuickSilver Inspection System incorporates special lighting and image processing developed exclusively for inspecting and measuring gold bumps. It measures the length, width, area, location, and reflectivity of every gold bump on every wafer at full production-line speed while it detects bridged bumps; missing and extra bumps; bump cracks, scratches, nodules, and pits; and extraneous material on wafers.

The high-volume production of surface-mount-technology (SMT) PCB assemblies is one of the largest markets for AOI systems. SMT manufacturing starts by screen-printing a solder-paste pattern onto the bare PCB. The alignment of the pattern to the component mounting-pad areas as well as the thickness of the paste deposit determine solder-joint integrity and component position after reflow. If the screened-pattern is incorrectly registered, during reflow the surface tension of the molten solder will pull small components out of position.

Many companies are using AOI after the screen-printing stage to inspect the deposited pattern for registration, missing areas, and excess solder. Correcting these errors before component placement will ensure a higher yield following reflow.

After the components have been placed, the PCB may be inspected again. It’s a matter of determining whether inspection before reflow is beneficial or if it is only required after soldering.

Many component placement machines have built-in vision systems that check device polarity and position as the parts are placed, making separate pre-reflow inspection less necessary. Finally, after soldering, inspection determines the quality of the solder joints, whether large flat-pack ICs have shifted position, or if small parts are tomb-stoned or billboarded, for example.

It Only Sounds Simple

The features and capabilities required by a specific machine depend on how it is used. According to Chris Kellogg Sr., industry marketing manager at Cognex, “AOI equipment first has to check for the accurate location of the correct components once they have been placed. After reflow, AOI often is used to analyze the quality of the solder joint, which can be very difficult. It may require the use of multiple cameras and lights as well as brightfield and darkfield illumination techniques. AOI systems also check for tombstoning, caused by having insufficient solder paste underneath the pads, and missing or extra components.

“When used to inspect bare PCBs for trace defects, AOI systems require high-speed defect detection and registration and the capability to classify defects to determine the proper course of corrective action,” he continued. “During PCB assembly, AOI systems need high-accuracy, robust pattern-location algorithms to reliably check for the proper amount of paste, paste registration in relation to the pads, and bridging between pads. Following component placement, speed and accuracy are critical to an AOI system’s capability to identify the correct part and determine its position in the field of view.”

So, can you use the same type of AOI machine in all these positions along the production line? Maybe, but John Arena, marketing manager of the Assembly Test Division at Teradyne, cautions against it. “Many manufacturers want to have single-system capability across multiple assembly stages. The intent is to have a Swiss army knife approach and equip the system with all combinations of lighting and imaging to support the wide range of requirements by enabling/disabling capabilities. The strategy works everywhere to a degree, but nowhere well. Compromise is implied and demonstrable in such situations,” he said.

A good reason why the one-size-fits-all strategy doesn’t work well is the need for different amounts of information at successive manufacturing stages. In general, the more variability that exists in the thing being inspected, the more difficult it is to distinguish good from bad. Determining solder-paste position is the easiest of the inspection jobs. Trying to measure solder-paste volume in addition requires thickness information and is complicated by any PCB warpage that may be present.

At the other end of the scale, using AOI after reflow typically requires the most sophisticated implementation due to the subtlety of solder-joint defects. Bob Langenberger, AOI marketing program manager at Agilent Technologies, said, “Multiple cameras and sophisticated light sources are used to increase the level of solder-defect coverage by AOI. Multicolored lighting at different angles or multiple cameras at different angles and positions allow an AOI system to capture more information so it can identify more defects. Sophisticated camera configurations and light sources also can provide additional coverage on laser marks, bent pins, tilted connectors, and marks on the sides of components.

“Smaller components and higher throughput requirements also can drive the need for multiple cameras,” he explained. “Using cameras in parallel or with different resolutions plays a part in balancing the need for throughput but with enough resolution to inspect small components. The trade-offs with sophisticated systems are greater difficulty of programming and calibration, and as you attempt to find more subtle defects, the probability of false failures also increases.”

Recent Improvements

Conflicting requirements for an engineering solution are nothing new. However, many of the problems that confronted AOI developers and users in the past have been solved. Today’s systems use better algorithms, have faster image subsystems, and display more pixels. Teradyne’s Mr. Arena said, “Given the pace of change in the computer industry, the implication is that AOI buyers should evaluate the relative age of a system’s architecture. If it’s more than a couple of years old, it’s probably built on constraints that have been vastly improved or eliminated.”

Image Acquisition Time

Newer cameras with larger pixel arrays can capture more information about a PCB in each frame. This means that the board doesn’t have to be moved so many times to get complete coverage. Fewer frames have to be transferred for processing although each one is larger. Also, downloading camera data in parallel will increase throughput compared to serial data transfer.

Image Processing Time

Distributed processing places computing power where it’s needed, avoiding the bottleneck of the single CPU in a typical PC. As an example of what is available, the Advanced Pixel Processor from Coreco Imaging operates on pixel data at rates up to 200 MB/s.

Yvon Bouchard, applications manager at the company, said, “At those speeds, the PCI bus of the computer becomes a problem. As a result, designs that incorporate on-board frame buffer memory and dual-ported architectures will become the norm. As acquisition speeds increase to the 800-MB/s range, embedded vision processing will be mandatory.”

The Advanced Pixel Processor includes frame buffers and a reprogram-mable 800,000-gate processing core. In operation, the core logic is programmed to execute the required image-processing algorithms such as fast Fourier transforms (FFTs), run-length encoding, and a real-time Bayer filter function. Of course, faster PCs also help when extreme speeds aren’t required.

The organization of the image-processing algorithms affects execution time as well. According to Mr. Langenberger from Agilent Technologies, “Algorithms implemented as large statistical database models tend to be affected more by PC speed than algorithms that are more focused on feature extraction. However, increased PC speed also has enabled alternative processor-intensive algorithms that previously were impractical.”

Board Movement

AOI machines are being built with linear motors, drives, and bearings. When used with antivibration software, settling time can be improved. Although these changes may not significantly improve throughput, they will increase accuracy.

Open Systems

Moving from proprietary to commercial, off-the-shelf PC-based systems is a major cause of recent price reductions.

Easy Programming

If you plan to manufacture batches of dissimilar PCBs, then quick programming is important to you. Many machines are self-learning in two ways: they use part information from the computer-aided design (CAD) pick-and-place file and learn by automatically stepping through a known-good board.

Summary

As impressive as the capabilities of modern AOI systems are, they can’t inspect what they can’t see. CR Technology has combined X-ray and optical inspection in the XRV Combo system. According to Richard Amtower, the company’s president, “Our business mission is to add value for our customers through yield improvement. We do this by detecting defects with speed and accuracy and quick and effective reporting of defect information for corrective action.

“Because of the widespread adoption of ball-grid arrays (BGAs), chip-scale packages (CSPs), and flip chip technology, SMT manufacturers have become the major customers for high-resolution industrial X-ray systems,” he continued. “They are driven by the need to verify the integrity of solder joints that cannot be viewed visually.”

The XRV Combo completely inspects PCBs, not only checking visual features such as component type and polarity, but also hidden features such as BGA solder connections that only X-ray can see.

Just as the XRV Combo system has only recently become available, today there may be viable AOI solutions to your electronic manufacturing inspection problems that were impractical only a couple of years ago. Look for systems that provide more information through the use of color and higher-resolution digital cameras. Also consider how that information is processed, whether a fast PC is adequate or additional power is required in the form of a frame grabber with built-in digital signal processing (DSP) capabilities, for example.

According to Mr. Amtower, “AOI now is accepted as an essential component of test and inspection strategies for PCB manufacturers. The explosion of competitive systems in the last three years reflects a major change in the status of inspection systems driven by customer requirements and equipment capabilities.”

Acknowledgements

The following companies provided information for this article:

How They Do What They Do

Correlation is accomplished by shifting an image in both the X and Y axes with respect to a reference image while determining the alignment giving the best correspondence between the two sets of pixels. Algorithms based on correlation provide an accurate means of matching patterns, but results degrade with variation in contrast, rotation, and scale of images.

The full name of the algorithm, Normalized Grayscale Correlation, confirms the fact that it works best for live and reference images having similar contrast ranges. Figure 2 shows a correlation example and the bright response peak that occurs when the two images are most closely aligned to each other.

An alternative type of algorithm finds edges and shapes. This type of algorithm can be limited to counting edges along a user-specified line profile. For example, if a connector were inspected that was supposed to have four wires attached, counting edges along a straight line that intersected all four wires should yield eight edges. Any other number is a result of a faulty part or such poor contrast or resolution that the algorithm can’t work properly.¹

An example of a pattern-matching algorithm based on the geometric properties of an image is SMART Search™, developed by Imaging Technology, now part of Coreco Imaging. How accurately can a feature be located? According to the company, “Two techniques are involved. First, a geometric search tool locates a pattern…. Once the pattern is located to within a pixel, then a subpixel calculation is done using neural-net based techniques…. [SMART] can…provide better than 1/40 of a pixel accuracy.”²

References

1. Hanks, J., “Using Edge Detection in Machine Vision Gauging Applications,” National Instruments, Application Note 125, August 1998.
2. “SMART Search,” Coreco Imaging, 2000, www.imaging.com.

Return to EE Home Page

Published by EE-Evaluation Engineering
All contents © 2001 Nelson Publishing Inc.
No reprint, distribution, or reuse in any medium is permitted
without the express written consent of the publisher.

May 2001