Current Issue An Independent Journal Dedicated to the Advancement of Chip - Scale Electronics March 2001 Email the editor

Back

The Yield Factor

Yield produces the greatest impact on rework cost when compared to machine uptime and throughput. As throughput changes and uptime changes, rework costs change proportionately.

This is also true for yield, but with a more significant effect. Incremental yield increases, in steps of 0.5% from 99.50 to 99.95%, can mean reductions in rework costs of 10% up to as much as 50%.

A further increase in yield from 99.95% to 99.98% will account for a 60% drop in rework costs, as Figure 4 shows.

Figure 4. Percentage change in rework vs. yield

The Effect of Yield On Rework

Consider the cost of rework or the value of the parts that are potentially scrapped. Do they need to be reworked, or is it less expensive to scrap rejected parts?

Depending on a company's business model (and actual rework costs), these events may be treated in a variety of ways. One company surveyed values throughput above all other factors. This company feels that rework costs at the $0.008 per ball level are not significant, and on a social level, their business model supports employment for the repair team.

However, without fully considering the other factors in their analysis, their operation can actually generate more rework than a slower, higher yielding process (see Figure 5).

Figure 5. Potential rework costs vs. yield

For example, consider two machines producing a CSP with 48 I/Os and 21 parts per strip. One machine is 50% faster than another, but with a reduced yield and machine uptime (see the table).

Things look grim for Machine 2! Even after the penalties of reduced uptime and yield saddle Machine 1 with reduced efficiency, this machine remains 44% faster in effective throughput.

Defect Rate

The interesting part is the potential defect rate. At 48 bumps per package and 21 packages per strip, there are now a greater number of defective sites requiring repair for Machine 1 than 2:

As a result, rework costs at a rate of $0.008 per I/O (assuming that only the defective sites are repaired-which is typically not true in a practical sense), would generate repair costs of:

Another important item to note is the actual cost of the repair or rework. The number used in this discussion can be disputed as a single data point.

Nevertheless, process control extends to all aspects of an operation, including rework. As a process or a business matures, it is necessary to periodically revisit the models used to define them on a cost basis as well as their process efficiency or output.

Time To Buy Another Machine?

The repair/rework costs shown above are not significant in either case, until one considers added repair issues like separating defective parts from the product stream or total annual costs for repair or scrap and multiple machines.

The difference in projected rework costs at year-end between the two machines in the example above is roughly $77,800. We could conclude not to rework.

For a five-line operation, the number for repair alone for Machine 1 becomes $450,000. Here is the time when a company can be forced to seriously consider another line or single placement machine just to repair defects from the other machines. (This can also create a business opportunity for someone else to perform the repair.)

If two processes are equal in yield and throughput, but unequal in machine uptime, the less efficient process actually produces less defective material.

This factor clearly must be weighed against the value of good production, but it does illustrate that these factors should not be ignored in an analysis.

A 3% difference in machine uptime would allow the more often "down" machine to "save" the company money by producing roughly $1,000 less rework per year. This is hardly an impact number, and underscores the fact that yield and yield alone is truly the most significant effect on the operation.

While throughput is king in the minds of many; it has to be coupled with yield to be most effective. You could argue that machine uptime can be neglected, and that is true to a point, but likely when high throughput does not matter, as well.

'Cost Is King, But Yield Is God!'

I'd like to paraphrase Chuck Bauer of TechLead Corp., who says, "Cost is king, but yield is God" for IC package production.

Chuck has noted in the past that as the BGA market grew, emphasis lay mostly on the quality, availability and pricing of BGA substrates. This resulted from the dominance of the substrate in the manufacturing of a package.

Contributions from other than the substrate include handling (strips), materials wastage (molding compound and solder spheres used on "bad" substrates), and package subcomponent cost. (Substrates still represent nearly 40% of BGA materials costs.)

As a product matures, yield takes a front seat, either alongside throughput or actually taking the driver's seat.

Ultimately, yield, coupled with through-put and a history of efficient machine operational time (including maintenance and die changeovers), is the building block of a successful area array package assembly operation.

Acknowledgements

The author acknowledges the following for their assistance: Chuck Bauer, TechLead Corp.; Jan Vardaman, TechSearch Inter-national; John Briar, STATS; "Eddie" Moltz, Texas Instruments; Jon Greenwood, Amkor Technology; Mark Barden, Autron Tech Pte Ltd Singapore; Watson Liu, Taiwan Sigma Equipment; and Speedline associates.

References

1. T. Zizzo, "Semis: So Far, So Good," Electronic Business, December 2000, p. 46.

2. S. Berry and S. Winkler, "Area Array I/O Pitch Will Continue to Decline," Chip Scale Review, March-April 2000, p. 9.

Tom Edwards

Mr. Edwards is the business team leader for the Speedline Technologies' MATRIXX solder sphere placement equipment line. In previous posts with Cookson Electronics, he was global product and marketing manager for Cookson Semiconductor Packaging Materials. He earlier held a similar position with Alpha Metals Inc. Mr. Edwards graduated with a bachelor of engineering degree in metallurgy from The Stevens Institute of Technology and an MBA from Rutgers, The State University of New Jersey. [tedwards@speedline.cookson.com]

Back