A powerful tool for improving yields, statistical process control (SPC) is highly effective for products such as PCBs that have close tolerances. The two keys to using SPC to improve yields are: 1) the effective application of statistical techniques to analyze process variation and 2) the successful use of that analysis to effect changes in the manufacturing process.

How SPC Improves Yields

Three Considerations of SPC Software (below)

Yield problems with boards tend to be process-related, such as machine maintenance, temperature settings, poor circulation, or too much solder. Simple solutions such as buying new machines or better training of operators may not solve the problem.

Using SPC to improve yields involves reducing variation in the process and/or inputs to the process. Assuming that the process is capable of meeting customer specifications, the less variation that occurs, the fewer defects in your finished product.

While SPC tools, such as control charts, can be used to track and benchmark actual yields, this method will reveal very little information about why the process is achieving these results. Gathering data on key attributes of the process that affect yields, such as solder temperature, is far more useful.

Whenever possible, variable data should be collected directly by process software, which eliminates data measurement and entry errors from operators. When this collection is not feasible, establishing and following collection procedures for consistency and integrity is critical.

Is the Process in Control?

Once the data has been collected, the next step is to find out whether the process is predictable, not whether the process is producing acceptable product. If the process is producing a high percentage of defects predictably, changes can be made to reduce defects. Process changes to an unpredictable process are much harder to evaluate.

SPC analysis is used to evaluate the process. A predictable process is “in control,” meaning in statistical control. Likewise, an unpredictable process is “out of control.” Note that “in control” does not mean the process is producing product within specification.

The workhorse of SPC is the control chart, which graphically shows the extent to which various measurable characteristics, such as solder height, relevant to the process are in control or out of control. Each chart includes upper and lower control limits, which are calculated to distinguish between in-control points—those inside the limits—and out-of-control points—those outside the limits.

A common control chart is the x-bar and range. This chart shows the central tendency of the process as an average of each sample (x-bar) and the short-term variation of the process as the difference between the largest and smallest measurement in each sample (range).

If the process is statistically predictable, initial control charts will show almost all of the data points falling between the upper and lower control limits; only three out of 1,000 are expected to fall outside of the limits due to chance. If nearly all of the data points do fall between these control limits, then the process is being influenced by external or special causes such as differing raw material supplies, varying temperatures or different operators. If the process is not predictable, investigate the external causes and correct them before making process changes that might impact yields.

Examining Solder Height Defects

If initial control charts show that the process is predictable, use additional charts to analyze the process further and help determine how yield might be improved. The following example shows how SPC analysis is used to improve yields through increasing solder height consistency.

A PCB assembler tracks solder height to make sure it is within control limits, which is critical. Inadequate solder height can indicate many problems, including wear on the solder stencil, a calibration problem with the machine applying the solder, inadequate stencil cleaning or a raw material problem. Any of these problems can lead to waste or rework.

In this example, the specification for the solder height is 0.25 to 0.55 mm. Samples of five units are taken regularly and analyzed with an x-bar/range chart. The average of each five-unit sample is plotted on the x-bar chart, while the difference between the largest and smallest values of each sample is plotted on the range chart.

In Figure 1, data from a full day’s run is charted in an x-bar/range chart. This chart indicates that the process is out of control on the x-bar chart—note the two data points that fall below the lower control limit—but is in control on the range chart. This result indicates that the process is being influenced by something external to the process.

Figure 1

Results from a full day's run.

ENLARGE IMAGE

In this case, the process engineers learned, by investigating the records, that the inadequate solder heights came from lots using raw materials from a single vendor. Further analysis of lots from that vendor revealed the vendor's inability to produce consistent material. As a result, the manufacturer could either work with the vendor to improve consistency or drop the vendor as a supplier.

The process is run and charted for another day using only raw materials from reliable vendors. The process is now in control (Figure 2). Note that all data points in the figure are between the upper and lower control limits on the x-bar/range chart.

Figure 2

An in-control process.

ENLARGE IMAGE

Now that the process is predictable, the next step is to determine how capable it is in meeting specifications. The correct analysis is a process capability histogram that shows the distribution of solder heights and some statistics to help analyze the process’s ability to meet specifications.

As the histogram in Figure 2 shows, very little room exists between the specifications and the spread of the process, as shown by the ±3 standard deviation lines. The most commonly used statistic to evaluate the relationship between the variation in the process and the specifications is Cpk, which looks at the ratio of the spread of the specifications and the spread of the process, or the distance between the ±3 standard deviation lines. A Cpk of 1.3 is usually considered adequate. This process shows a Cpk of 1.05, indicating that it will not be capable of consistently producing within these specifications. To make the process more capable and increase yields, the inherent process variation must be reduced.

In this case, the process engineers discovered that the machine needed certain components replaced more frequently. After replacement, the x-bar/range chart indicated a process in control, and the histogram showed it was capable of producing output well within specifications. Note the Cpk of 1.43 in Figure 3.

Figure 3

A process with an acceptable Cpk.

ENLARGE IMAGE

Defect Tracking With Pareto Diagrams

After the board has been assembled, the PCB manufacturer uses Pareto analysis to examine the relative contribution of different defects that lead to PCB rejects. Pareto diagrams are commonly used to rank the relative frequency of different categories of defects. With a Pareto diagram, the quality control staff can assess the relative contribution of different defects to PCB rejects and assign priorities for addressing their causes.

Figure 4 illustrates how Pareto analysis can summarize a large number of defect categories; in this case, the figure shows the top 10 categories. The next step is to create SPC charts for the major defects. Figure 4 shows charts illustrating the percent defective. The interpretation of these charts is essentially the same as for the x-bar/ range charts: Points outside the control limits indicate an out-of-control process.

Figure 4

Pareto analysis of a number of defect categories.

ENLARGE IMAGE

In this case, the most common defect, pinholes, is out of control, while the next two defects, open joints and insufficient solder, exhibit perfect control. The strategy for dealing with these problems is, therefore, quite different. The pinhole problem is severely affected by external influences/special causes, while the open joint and insufficient solder problems are built into the process. In the former, process engineers will look for the special causes, while, in the latter two, they will need to change the process itself.

A process's control status—in control or out of control—does not indicate whether the defect levels are acceptable or not. It indicates whether the process is predictable and what the most fruitful line of investigation might be. Another important concern is defect cost. Again, Pareto analysis can be used to re-rank defects by cost rather than frequency of occurrence. A defect category may be frequent but carry a low cost, while another, although rare, is very costly.

Using SPC To Effect Change

The most stringent and sophisticated SPC analysis will not improve the process if the information gained is not used to make changes. The real challenge of SPC arises in problem management, such as when data clearly show that a supplier's solder is causing yield problems and the process engineer must convince purchasing that a favorite vendor must be dropped.

In this situation, specialized SPC software and the clarity of graphic communication can make a difference. With good quality graphics, even a nontechnical staff person can easily see the process going out of control and producing defects when a particular vendor's product is used.

For example, a sample situation in which simple, high-quality SPC graphics can play a strong role in improving yields may be the following. SPC analysis reveals that a supplier consistently ships material outside the specifications for the PCB manufacturer's process. The purchasing agent is reluctant to drop the supplier because the company receives favorable purchasing terms. By showing control charts that clearly reveal the defective materials, the process engineer is able to show how the supplier’s impact on the bottom line is more significant than the cost savings from the favorable terms.

Conclusion

From control charts to Pareto analysis, SPC offers an entire workbench of power tools that can show you how to improve yields. However, the most stringent statistical techniques will not help if they do not lead to process changes. In the end, your ability to improve yields with SPC will rest on how well you can convince others to implement what you have learned.

About the Author

Jeffery L. Cawley is vice-president of Northwest Analytical, Inc., Portland, OR; e-mail: jcawley@nwasoft.com.

Three Considerations of SPC Software 1 How Easy Is It To Learn and Use? Ease of use is key because employees with different skill levels will use the software. Software should have a familiar look and procedures similar to other common applications at your company. Commonly used tasks should be obvious, and users of all levels should be able to tell where they are in the program at all times. 2 Can Routine Tasks Be Automated? One of SPC software’s great advantages is automating the laborious task of producing routine charts and reports. Will the software allow you to automate routine functions? If so, how easy is the automation? Before deciding, ask for an evaluation copy of the software and try automating a task you are continually doing, such as generating a weekly report. How easy is automating the task? Will you have to use a file that records keystrokes or does a wizard make the process easier? Can you reduce the entire task to an icon on your desktop that, when clicked, will perform the task? 3 Can It Be Linked To Your Databases and/or Manufacturing Execution System (MES)? Companies are increasingly collecting process data in company databases. These databases can be mined for SPC analysis and eliminate the problems of duplicate data entry. Entering the data twice, once in a corporate database and then again in the SPC software, is wasteful and increases the likelihood of entry errors. Also, with twice-entered data, SPC analysis may not be based on the most current production data. SPC software that connects with your corporate databases eliminates these problems and provides much greater flexibility in analyzing data.

back to top