Repeatability Is AOI Watchword for Volume SMT Production

Dominic Haigh, Teradyne -- Test & Measurement World, 12/1/1999

For automated optical inspection (AOI) to establish itself as a reliable method of process inspection, systems need to overcome the perception that they can produce unpredictable or inconsistent results. Nowhere is this requirement more acute than where manufacturers use AOI to inspect similar SMT products built on several lines, in several locations, and even in several continents. Products must perform equally well at any time around the world, which demands a high level of stability and repeatability in the AOI systems themselves. This requirement is especially true in the increasingly common model of contract manufacturing, where the OEMs rely on manufacturing partners to minimise costs or serve regional markets.

The most practical and acid test you can apply to any AOI system is to check its repeatability of performance on a real board. You should expect the system to report the same results — with and without the conveyor turned on — today, and over the weeks and months ahead. As a further requirement, expect a test program developed and run on one system to produce the same defect coverage, inspection time, and false flag rate on a second or third model of that type of system.

AOI is a complex technology that involves several engineering disciplines, including mechanics, optics, and electronics (see Figure 1). Each one of these disciplines can have an impact on the stability of an AOI system.

Figure 1. AOI systems combine mechanics, optics, and electronics, and elements of any one can influence the overall stability of its performance.

An AOI system typically scans a camera/lighting head over the board under inspection. This method is the preferred solution for two reasons.

First, moving the head rather than the board minimises system footprint for a given area of inspection, and allows vendors to design smaller and more rigid frameworks. (Moving boards under a fixed head requires a system at least four times the area of the largest board).

Second, moving the head allows more precise control of mechanical variation, because the moving object is totally under the design control of the system vendor. An AOI system vendor can design the head to be rigid and remain in specification over the acceleration and deceleration of the X-Y table. In contrast, systems that scan by moving user’s boards introduce unknown variables. The rapid acceleration required for best system throughput can cause board flex, or could even cause components to shift position in a pre-solder inspection application.

AOI systems must also provide the required accuracy of movement and registration between the board and the cameras. In practice, inspection of fine pitch components requires system accuracy of ±0.025 mm or better. Any flexing of the system will cause variation in position and, thus, variation in results. You should aim to use rugged mechanical systems with welded frames and substantial internal mechanical components.

Throw On Some Light

The light source in an AOI system is just as much a test stimulus as a voltage or current source is to in-circuit test (ICT). As with ICT, it’s important to question the stability, repeatability, and calibration of that stimulus.

First, using illumination sources that age, such as flash tubes, means that an inspection program will not produce the same performance today as it will in the weeks or months ahead. You will either have to perform frequent re-calibration, or regularly tune programs to accommodate AOI light variation.

Second, for high-volume test, it’s important that light level is consistent from system to system. If there is significant variation from one system to another, you’ll have to develop separately tuned programs for each line, which means you’ll also end up applying different inspection criteria from one line to another as the inspection programs diverge.

Third, look for a light level calibration procedure that is straightforward, and preferably automatic. At least ensure that you can perform calibration without vendor support.

Lastly, you should check on the mean-time-between-failure (MTBF) of the system’s illumination source. You don’t want to contemplate the frequent changing of such a critical system element. Again, high system availability precludes using illumination sources that fail, age, or require maintenance. Solid-state sources, such as LEDs, satisfy the requirements for a high-volume application because they are reliable and output stable levels of illumination. For example, typical MTBF on an LED-based lighting system is in excess of 50,000 hours, compared to the 2,000 hour MTBF of flash tubes. LEDs don’t age appreciably and they require no on-going calibration. Flash tube sources require monthly, or even weekly calibration.

Figure 2. An overall stability in the mechanical framework, cameras, and lighting is essential to minimise AOI performance variations system-to-system, site-to-site, and over time.

Look Closely at Optical System

The levels of repeatability required in high-volume manufacture demand that AOI vendors understand and control even apparently stable elements such as optics. For example, camera lenses are far from ideal. They introduce non-linearity across the field of view, variations in optical axis, and other variations such as chromatic aberration. You should check whether an AOI vendor has analysed, understood, and controlled these types of variation. For example, the system should correct for lens aberrations as part of its calibration procedure. Also, vendors should be able to describe how the optical assembly is aligned to eliminate variation.

For consistent registration across the field of view, look for automatic tools — preferably built into the system hardware and software — for performing an initial camera alignment and field of view corrections. If a system has no means for ensuring that a given object at a certain location will appear at the same location to the analysis software, there will be variations from system to system.

Make sure that optical elements — lenses and cameras — align optically as well as physically. Consistency of the optical path through the assembly (not just the mechanical registration of the lens and camera bodies) determines repeatability across systems.

Also make sure that the camera/ lighting head does not flex under the stresses of movement, and that the lens iris remains fixed under acceleration and vibration. Most lenses are designed for video applications where small variations of a few percent in the iris and, thus, in the intensity of light received are unimportant. In SMT inspection, these small variations frustrate the requirement for consistency without some form of control.

Analysis Also Needs Stability

It’s important that your AOI system applies inspection criteria consistently over time. Yield and cycle times are key metrics for staying in business, and the means of increasing yields is the control and reduction of variation. If your AOI system applies inconsistent criteria over time and from system-to-system, it is working directly against these goals.

For example, systems that continuously “learn” on good and bad production boards do not satisfy this requirement. The net result on such systems is that inspection standards vary with time, and that leads to system-to-system divergence between systems inspecting similar board types (simply because they learn on a different population of boards and sample defects). This feature is unacceptable in volume production.

Instead, you should look for systems that, first, set up rules for every defect instance of a particular package and, second, allow you to globally modify these rules to match your specific requirements. You should then be able to apply these refined rules every time, on every instance of the part, and on every board.

Colour Can Confuse

Today’s SMT boards are visually complex objects that consist of many shapes, colours, and dimensions. An AOI system has to sort through complex image data and extract any defects. One way to simplify the system’s task is to process black and white, in preference to colour, images. In practice, most colour information in a board’s image is “noise” rather than data. For example, a change in hue of a board substrate tells you nothing about defects, but merely adds a layer of data variation that the system can ignore. For this reason, AOI systems that analyse monochrome images prove better suited to SMT inspection applications. The use of red illumination (preferably LEDs for stability) provides further improvements because any green substrate then appears black to a monochrome camera and totally eliminates the “noise” element from the data.

Occasionally, colours of an image (for example, a change in component body colour) may indicate genuine defects, but this benefit comes at the cost of increased noise in the images. In general, the result of using colour analysis is susceptibility to meaningless variation, which gives a system the potential to fail good boards or miss defects. T&ME

Dominic Haigh is a business unit manager with Teradyne. He was previously European marketing manager and has been with Teradyne for 9 years.