esting and inspecting electronic products during manufacturing has always required shooting at a moving target. No matter how successful your approach, you must never become complacent. Device and board architectures, manufacturing methods, and other factors can quickly render any step obsolete.
Automated optical inspection (AOI) is a case in point. Such developments as the proliferation of ball-grid arrays have reduced the effectiveness of automated inspection. Will AOI fade from the inspection landscape as fast as it emerged?
I asked Pam Lipson, CEO of Imagen and a scientific advisor to Landrex Technologies, to explore the status and likely evolution of the technique. She began by examining the past and went on to explain that software is the key to the future.
"In the beginning, AOI systems were entirely constrained by the capabilities of the hardware," commented Lipson. "Early solutions featured monochrome cameras, red LEDs, and little computing power for image analysis. Consequently, manufacturers first applied AOI to examining reflective features like solder joints—which lent themselves to that type of image capture system—and predictable bare boards—which lent themselves to that level of inspection."
As electronics manufacturers' requirements changed, systems vendors concentrated on improving the hardware, rather than looking elsewhere. Lipson contended that the systems' hardware-centric nature limited their ability to adapt to the evolution of the products being inspected. Hardware changes alone simply could not be flexible enough to allow sufficient variation in what constitutes a "good" circuit or change quickly enough to keep pace with product development.
Increased specialization of electronic products complicated the problem. "We're rapidly approaching an environment that manufactures products in lots of one," remarked Lipson.
Still, one particular hardware innovation signaled a new era. "The development of white LEDs dramatically altered the landscape," Lipson continued. "For the first time, AOI systems could 'see' more. We could strobe products under inspection with white light, and with color cameras we could detect most visible colors. Together with faster computers and more memory, AOI systems could analyze a wider range of targets and provide a larger variety of diagnostics and measurements."
Nevertheless, the key to moving forward lay elsewhere. "Traditional inspection algorithms, still-used remnants from the days when computing power was very limited, are brittle," said Lipson. "They break easily. Using image correlation, for instance, a system examines a pixel in the target image and compares it with a stored ideal, determining how closely their gray values or color values match and making a go/no-go decision. If you set the threshold of what is acceptable too high, you might accept a black component, but not an identical blue one or a black one with oxidized encaps. Set it too low and you might accept the black component, the blue component, and the oxidized encaps but not notice if the component is missing altogether. These criteria must be tweaked constantly—a poorly understood hidden cost of conducting automated optical inspection."
Technological improvements, however, have triggered a paradigm shift in the design of inspection systems. Hardware architectures can now remain relatively stable, with changes occurring primarily in system software.
"Transferring responsibility for change to the software offers fundamental advantages," Lipson explained. "Software can learn. A system can look at examples of what constitutes a good or a bad circuit and later independently draw conclusions about significant variations. Through further examples, it can quickly adapt to configurations it has never seen before. Also, as parts shrink, the signal-to-noise ratio decreases both because of the image capture system and the relative size of the part to the board features around it. Separating the target object from the noise in such images requires a good software-based analysis. And users can swap out software modules as new ones become available or as the target product mix changes much more easily than they can modify hardware."
Lipson concluded, "We find another compelling reason that advanced software will make AOI more relevant than ever before. Shrinking devices, increasing board densities, and more reliance on BGAs make board repair at the back end much more difficult. As always, finding defects earlier in the process costs much less. If we measure the process itself, we can find subtle defects as well as detect trends that may cause future defects. Manufacturers may find themselves pushing inspection further upstream to predict defects and adjust process parameters to prevent them. Measurement and analysis software in AOI systems is key to making that happen."