Auto Teaching for Component Programming
December 31, 1969 |Estimated reading time: 7 minutes
As component technology and surface mount equipment have advanced, the challenge remains: how to handle the latest and most complex components quickly and effectively.
By Niklas Andersson
While complex components generally only make up a small part of the board population, they take up most of the setup time because the operator or production engineer must program component data into the machine.
Manufacturers in markets such as prototyping or small series productions, in particular for large OEMs, find that highly complex components with dense or complicated interconnections (such as for flip chips) are becoming more common. To remain competitive in the industry, it is essential for production economies that these complex components are mounted accurately. These components are not only the most expensive by far, but also the most time-consuming if rework is required. Rework also can compromise quality.
Alignment and placement accuracy, as well as quick and accurate setup and changeover, are paramount. Complications associated with handling even one complex, non-standard component on a board accurately can literally bring production to a standstill.
Challenges
An increasingly common type of complex component for high-reliability applications such as space, military and telecommunications, which require hermetic packages and high-density interconnections and higher board-level reliability is the ceramic column grid array (CCGA). CCGA packages use high-temperature solder columns instead of balls to create a higher standoff for more flexible interconnection, as well as significantly increase thermal fatigue life of the package's solder joints. Today, CCGA components represent a common programming challenge for machine operators, as do highly complex, asymmetrical BGAs; CSPs with very fine pitch; large MCMs; large QFPs; and flip chips. These components are not only complex and expensive, but in many cases also extremely fragile and easily damaged if not handled properly (Figure 1).
Figure 1. CCGA packages use high-temperature solder columns instead of balls to create a higher standoff for more flexible interconnection.
While QFPs only have leads on their sides, BGAs and CCGAs have interconnections in their arrays, thereby using less space on the board. Demand for these complex components will continue to increase, since applications are getting smaller and as a result, less space is available for devices.
In current production environments, these types of components often are delivered to production without CAD data. Where no package data is available, it must be created quickly at the machine. However, asymmetric BGAs are difficult and time-consuming to program manually, since each ball's coordinates in the array must be entered. Small components might even require a microscope to see details of the balls/leads. Larger components also can be challenging because they are difficult to illuminate effectively, often resulting in shadowing effects along the perimeter. Low contrast between leads and the background make false detection of errors by the vision system common. This is particularly true of CCGAs with their almost mirror-like undersides, as well as the "busy" images of microBGAs with traces of their interconnections showing underneath (Figure 2).
Figure 2. While complex components generally only make up a small part of the board population, they take up most of the setup time because the operator or production engineer must program component data into the machine.
An experienced operator or production engineer needs anywhere from 30 minutes to a few hours to program a new component, depending on complexity. Normally, this is done when setting up the machine for a new job, resulting in long setup times and machine downtime. While this costs time and therefore money, the costs of faulty placement, which jeopardizes the quality of the entire board, potentially are even higher.
Auto-teach Function
Reducing both the time and risks involved in handling delicate, expensive and complex components can be accomplished using a so-called "auto-teach" function that dramatically speeds up the programming of non-standard or complex components, such as those described above. This provides a new and efficient way of creating package data for the mounting process. If necessary, this solution even can be used to "teach" the machine standard components.
Figure 3. Auto-teach software uses its 'snap-to-grid' function to match the measured data against a standard grid, after which adjustments can be made. The image result is presented to the operator on the screen, with each detected lead/ball shown and localized with red marks.
The process starts with taking a grayscale image of the component using the machine's vision camera. The computer then analyzes this image to detect the BGA or leads. The software then uses its "snap-to-grid" function to match the measured data against a standard grid, after which adjustments can be made. The image result is presented to the operator on the screen, with each detected lead/ball shown and localized with red marks. If there is a ball missing, a bent lead on the component from the manufacturer or a fiducial mark that has been mistaken for a ball, to name a few defect examples, the operator can use the software to quickly and easily add and/or remove leads/balls directly on the screen using the mouse (Figure 3).
The result is perfect CAD data for the component based on standard pitches. This is important, since slight variations in the individual component used to "teach" the machine do not affect the resultant CAD data for the component.
Step two in the auto-teach function uses the vision software to set up illumination and vision tolerance parameters for the component in what essentially is an optimization stage. The component is illuminated from different angles and the camera images are collated. The best illumination then is selected, automatically eliminating potential sources of error such as shadowing, and the function automatically determines quality parameters for the component. The resultant package data then can be shared with other machines in the facility, eliminating the need for the procedure to be duplicated at each machine. This saves time, and sharing data reduces the risk of inaccuracies. The entire auto-teach procedure itself takes only a few minutes.
Figure 4. Using the auto-teach function reduces both the time and risks involved in handling delicate, expensive and complex non-standard components. The BGA shown is programmed in a few minutes and presented to the operator on the screen.
For example, an operator is faced with a new asymmetric BGA never before mount-ed by the machine. Rather than manually entering the array, the auto-teach function can be used to rapidly "teach" the machine the BGA — how to align and place it, and what illumination and tolerances to apply. In fact, the finer the pitch and smaller the component, the bigger the benefit, since fine pitch is exceptionally difficult to measure manually (Figure 4).
When each component is picked for mounting, it is inspected by the vision system and compared with the package data. Defective components, or those not falling within the machine's quality parameters, can be identified and rejected before they are mounted on the board.
Reducing the Learning Curve
Since being faced with new components is nearly a daily occurrence for many operators and production engineers, auto-teach functionality provides many advantages. The cost of each component often is high, so the number of extra components supplied often is limited. Subcontractors or OEMs producing prototypes likely will get the only two ever made, making accuracy even more important.
Not only is setup time reduced, but new component types can be programmed into the machine at any time without loading a specific program or interrupting the production run. The learning curve for auto-teach is short, reducing the workload for production engineers and freeing up technical staff for more complex tasks. Auto-teach functionality also eliminates the need for component drawings, necessary for manual programming of package data. The machine also can find better min/max values for best performance than an operator could programming these manually. Finally, should a change of vendor become necessary, resulting in slight differences in the component to be used on a particular board, "re-teaching" the machine the new vendor's component is simple.
The auto-teach function also provides a higher level of quality assurance, since each complex component is inspected and compared to the tolerances set in the machine — essential for applications such as space, military and medical, where requirements are particularly demanding.
Flexibility
Since both machines and feeders today need to be flexible, why is the vision system so difficult to program? How fast is a machine that runs at top speed but sits idle for hours while operators program components?
The trend among SMT manufacturers is to move into higher-mix assembly, where responding quickly to the rapidly changing needs of the market is required. This demands the capability to assemble a wide range of products, most likely containing a wide range of components. The wider the variety of components the machine can mount, the more the machine can actually be used. With auto-teach functionality, the surface mount machine can be used to its maximum capacity for production, with minimal downtime for setup and changeover.
Niklas Andersson, M.Sc., machine vision engineer, may be contacted at MYDATA automation AB Sweden, Adolfsbergsvägen 11, SE-168 66 Bromma, +46 8 475 55 00; E-mail: niklas.andersson@mydata.se.