Evaluating machine vision for SMT

Machine vision plays a critical role in systems that perform surface-mount-device (SMD) mounting during PCB manufacture, as it locates PCB fiducial marks and verifies device position. In a typical SMD placement guidance application (Figure 1), a vision system uses an upward-viewing camera to view the

underside of devices and a downward-viewing camera to capture images of fiducials and placement sites on each PCB for board alignment.

The SMD mounter picks the device from a waffle pack, tape, or other source and rapidly passes over the device inspection camera. The vision system locates and examines each lead to ensure that devices with slightly deformed leads are placed correctly and those with missing, damaged, or out-of-tolerance leads are rejected.

Vision systems have improved greatly in the past few years, and many now offer a range of features to help SMD manufacturers. New capabilities include:

• increased alignment yields, which reduce operator intervention and increase mean time between alignments;
• high accuracy for handling ever-shrinking device geometries and device variability across manufacturers;
• high throughput to meet demanding placement speeds,
• increased productivity to reduce time to market, and
• an open software architecture.

Based on these capabilities, I've compiled a list of criteria you can use when evaluating vision systems for an SMD-placement guidance application:

Since reliable location of PCB fiducial marks is the first step in any SMD-placement guidance application, the vision system must tolerate a variety of degradations that can make a fiducial virtually unrecognizable. Variations caused by oxidation, tinning, and wave solder from the manufacturing process, for example, can introduce specular reflections and surface inconsistencies that dramatically alter a mark's appearance (Figure 2). Other problems that can affect the appearance of a fiducial include board warpage, artwork stretch, excess solder buildup, and board color changes. Vision systems that tolerate such conditions help increase alignment yield while reducing the need for operator intervention.

A machine-vision system should reliably locate any type of nonstandard device, no matter how unusual its shape. Placement guidance software is available with built-in geometric pattern-finding tools that can learn or "understand" the geometric attributes of a device, thus allowing the system to locate the device despite its peculiar shape.

SMD mounters typically use front-light illumination, backlight illumination, or both. Often, backlighting is used to create a silhouette of the device. The resulting binary-like image makes it easier for the vision system to locate the device, but backlighting can also present challenges for the vision system: The silhouette of the pickup nozzle will often protrude from behind the device or partially occlude small devices such as chips. While front-lighting techniques may prevent this phenomenon, the pixel-gray values of the nozzle itself may prevent the vision system from reliably distinguishing between nozzle and device. Look for a vision system that can recognize shape differences between devices and the pickup nozzles that hold them. By tolerating partial nozzle occlusion, the system will increase device alignment accuracy and prevent incorrect device placement due to vision error.

To accurately locate and check lead tolerances on devices as varied as full-size BGAs, random-patterned flip-chips, and chip-scale packages, a vision system must be able to find each component in a visually confusing scene. The vision system should also reliably locate devices with white ceramic surfaces, whose low-contrast reflective nature can overwhelm traditional vision technologies. To verify these capabilities, you should construct such a scene in the machine-vision system's field of view and test whether the software can distinguish the individual objects.

New vision software tools automate the process of creating device descriptions for even the most unusual components. Instead of manually entering parameters into the system to create device descriptions from scratch, users can simply present a device to the vision camera and take a picture. The system will automatically create a synthetic, CAD-like description.

This technique can improve the accuracy of device descriptions and greatly speed up their creation and testing—especially in situations where new devices are frequently introduced or that use uniquely shaped devices. Automating the device description process may reduce operator error, and thus improve alignment yield and system throughput.

Historically, processing images has taken much longer than acquiring them, but recent advances in CPU technology have accelerated vision processing to the point where image-acquisition speed may have become the limiting factor. To increase system throughput and minimize the time spent acquiring an image, look for a vision system that can support advanced line-scan, large-format (1-kpixel x 1-kpixel resolution), progressive-scan, and high-speed digital cameras.

Many SMD mounter OEMs have unique vision requirements based on the types of placement heads they use, the types of devices being mounted, lighting and optical configurations, and specific throughput and accuracy requirements. Vision systems that include an open software development environment can provide great flexibility and control in SMD placement guidance applications. Such systems may allow users to develop applications using turnkey inspection tools, C++ programming tools, or some combination of both. Some systems will even allow users to incorporate custom-developed vision algorithms into the programs.

When evaluating machine-vision systems for SMD placement guidance, look for a system that can increase alignment yield, throughput, and productivity, and is flexible enough to meet the demands of the surface-mount technology assembly market. By selecting a vision system that achieves these objectives, you can handle new component types and device variability across manufacturers, while simplifying the tasks for vision system users.