DPMO: A Tool for Achieving World-class Process Quality

Estimated reading time: 5 minutes

When used properly, DPMO is a more effective quality measurement tool than first-pass yield.

By Charles-Henri Mangin

Measuring and comparing the quality of assembly processes has been an elusive endeavor for some time. The number of defects per million opportunities (DPMO) is the metric used by best-in-class assembly shops. It is also the foundation for a methodology that permits computing the quality achievements at each step in the process, from screen printing to wave solder and manual insertion. This information is needed to identify which process steps and board design features need to improve. DPMO data facilitate the computation of the "pristine" yield for a printed circuit assembly (PCA), i.e., the yield that would have been reached had no online touch-up taken place.

Traditional First-pass Yield

First-pass yield (FPY) has been the traditional yardstick for evaluating assembly quality. FPY is plagued by three capital sins:

1. The worse the original yield, the more potentially misleading the future improvements. Going from 50 to 90 percent FPY can mean the original performance was dreadful and the "impressive" 90 percent achievement does not deserve praise.
2. It does not provide a valid comparison between different products or competitors because it does not factor in the impact of board design complexity.
3. It disregards online inspection and touch-up, which enhance FPY numbers and add to cost.

The DPMO methodology counts all defects generated during the assembly process and identified with manual and automated inspection and test. By counting all defects, it is possible to compute a pristine yield equivalent to FPY, assuming no touch-up has taken place prior to testing. The pristine yield is not a fully satisfactory metric for benchmarking purposes because it inherently varies with the board design complexity, but at least it is a truer representation of the process quality than FPY.

Defining DPMO

When DPMO measurements began in the early 1990s, there were no standard computation methods and, therefore, results were not portable and did not allow for performance comparison of different products or facilities. For example, DPMO could incorporate solely the defects counted at in-circuit, ignoring those corrected during the assembly process as part of touch-up, as well as those identified at functional or system test. Also, the total number of OFDs could be based on the number of parts, leads or a combination of both. When solder joints were factored into the equation, some facilities included vias while others included only the number of leads or terminations. By undercounting defects and overcounting OFDs, the value of DPMO is enhanced artificially. This is because the formula for DPMO is:

null

The number of OFDs on a given board part number is defined as the addition of the number of components plus the number of leads (Table 1).1 This is a standard that hopefully will become the industry norm. For an analysis of the DPMO at each process step, it may be necessary to add the number of glue dots to the defect opportunity count as defined above. This applies only to the computation of the DPMO at individual process steps, not the total process DPMO. For the total process DPMO, the number of OFDs remains equal to the number of components plus the number of solder joints.

Figure 1. Percentage defect distribution for world-class process quality.

Counting the defects should be as straightforward as counting the number of OFDs. Whether one defect is referenced "bent lead," "open solder" or "missing solder," it is still only one defect, not three. Common sense is recommended for defect counting; for example, follow the rules in Table 2. Just as the OFD count should not unduly inflate DPMO performance, the defect count should not underrate the achievement.For the case study selected as an example of best-in-class performance, 294 defects were counted from a sample of 1,080 boards (Table 3). In this instance, the number of boards included in the sample is the same from screen print to in-circuit test (ICT) because the board is assembled continuously on a double SMT line. When the process is split, i.e., bottom, top and test, the number of boards processed at each of these steps frequently differs.

Defects by Process Step

There are three steps to attribute defects to the process step where they originated:

1. List all defect types and indicate at which step they were identified.
2. Allocate each defect to its origin using the reference designators. Many cases are simple, but others are far more complex, especially for defects identified at test. Although this research can be tedious, it is necessary for a true quality measure of each step.
3. Eliminate from the list of defects those which are not process related. No defect found, defective components and bare boards are easy to identify. It can be more complex for items such as bent leads, which could be attributed to the part supplier or part handling during assembly.

Correlating defects with the actual process origin still is not as complex as it might seem common sense and some experience reduce this exercise to a few hours at most for complex cases. The results for the selected board and the sample previously defined are presented in Table 3.

When process quality is at a world-class level, screen printing and component placement most likely are the two largest causes of defects (Figure 1). In the case presented here, manual assembly is not world-class (best-in-class value should be a DPMO below 1,300, not above 5,000), which explains its 22 percent contribution to the total defect distribution.

For all processes except manual assembly, the DPMO values in Table 3 are world-class and generate high pristine yields considering the board complexity (5,314 OFDs and an average of 4.4 leads per component). The pristine yield is computed using the following formula:

Yield = [1 - (DPMO/106)]nwith n = number of OFDs per board

The data in Table 3 prove that FPY is not an accurate quality measurement: the somewhat unflattering process yield of 76.2 percent distorts the high level of quality a 51 DPMO reports. For a PCA with close to 1,000 components, a 28 DPMO for screen print, 82 for pick-and-place and 55 for wave soldering are undoubtedly world-class in 2000.

Conclusion

Process DPMO analyses remain the most potent tool for comparing the quality performances of different processes and assembly lines. The variation in DPMO levels for a given process and assembly line for different board part numbers facilitates the comparison of the manufacturability of different PCA layouts. This is an invaluable reference for enhancing the design for manufacturability (DFM) process. Robust process DPMO data are necessary to predict the yield of new PCAs at their design stage. This can influence the layout and component selection before production release and provide an estimated production cost, which impacts selling price and market competitiveness of the new product.

REFERENCE

1. "Best in Class Process Quality Benchmarks," report DPMO 2000, CEERIS International Inc.

CHARLES-HENRI MANGIN is president of CEERIS International Inc., P.O. Box 939, Old Lyme, CT 06371-0939; (860) 434-8740; Fax: (860) 434-8742; E-mail: ceeris@aol.com; Web site: www.ceeris.com.