Multi-line production alters test economics

The need to adopt flexible production strategies makes it necessary to modify traditional simplifying assumptions when justifying test-equipment deployment.

Steve Scheiber, Contributing Technical Editor -- Test & Measurement World, 9/1/2001

In evaluating test-strategy alternatives, you must include economic costs and benefits among decision criteria. In addition, once you decide on a strategy, company management will demand an economic justification before giving the go-ahead to proceed. If your strategy involves deploying a single line to test products, the calculations required to prepare the justification can be straightforward. Increasingly complex manufacturing operations, however, are making the calculations much more difficult.

The most fundamental change in the electronics industry over the past decade has been the reliance on contract manufacturers and contract-manufacturer-like in-house processes. (To improve efficiency, many companies set up their production facilities to behave as though they were contract manufacturers.) Contract manufacturers must maintain high utilization levels (the percent of available capacity currently running) in their factories to protect razor-thin profit margins. Similarly, in-house production operations must perform well enough to justify their continued existence.

As a result, production lines dedicated to a particular product, as in Figure 1a, have become less common in favor of a more flexible approach that accommodates fluctuating market conditions, product enhancements, new-product delays, and other challenges. Complicating the planning process, companies need to match production volumes to demand—such as by balancing loads among active production lines—to avoid building inventory that quickly becomes obsolete.

For example, in Figure 1b, Product 1 runs on Line A into Day 2, when the line switches to Product 5. On Day 3, the manufacturer needs to return to Product 1, but Line A is unavailable, so it is assigned to Line B. Line C is dedicated full-time to Product 3. Line D can provide sufficient quantities of Product 4 in two days. Therefore, when additional Product 3 capacity is needed on Day 3, it is assigned to now-idle Line D. 

Figure 1 (a) Simple models assume either that each product runs on a single manufacturing line or that all lines for that product are identical. (b) A more realistic model includes load-balancing within a single facility.

One consequence of transferring production to another facility is that the expertise that created the original manufacturing and test strategies does not generally go with the product. Responsible engineers, although available for consultation, remain in the original location to begin work on the next products coming out of development.

If allegedly identical manufacturing lines—either in one or more than one facility—do not behave identically, additional costs accrue. An in-circuit tester’s digital timing threshold voltages may vary slightly from one machine to the next. As a result, even if the production process is identical, one tester may pass some boards that another tester would fail. Additional false failures place more boards in the repair and retest loop, where they likely will cycle several times before someone decides to pass them to the next step. Widening test tolerances to prevent false failures means that some faulty boards will pass in-circuit test and will have to be detected at a later test step at a higher cost.

Figure 2 Here, process development and initial manufacturing take place in one facility; then, the product transfers to another facility for ramp-up.

To alleviate such problems, a test engineer tweaks the program after transfer so the results match. Such adjustments may prove trivial in some cases, or they may take hrs of engineering time.

Suppose that the situation involves a contract manufacturer running close to full capacity. The transfer from Line A to Line B in Figure 1b incurs a cost. Sending the product back to Line A at a later time will involve some of the same verification and adjustment, incurring those costs again.

In addition, bed-of-nails fixture performance can vary, depending on wire lengths, integrity of wire connections, and similar factors. Again, a program that runs reliably with one fixture may not immediately run correctly with a new fixture. If the second line represents additional capacity (as with lines C and D on Day 3 in Figure 1b), then the test engineer will have to match the new fixture/program combination with the previous line. The only subsequent headache is the need to maintain two sets of test programs and two sets of fixtures, which requires more storage space and higher costs for program maintenance.

Changes of circumstance

An engineer or manager who wants to purchase a tester or other capital equipment must create an economic justification in addition to a technical one. Economically justifying a purchase means showing that the benefits of that purchase outweigh its costs and that another alternative is not more cost-effective. To do so requires modeling and comparing the costs and benefits of each option under consideration. Constructing a model can seem a daunting task. Many engineers find the number of possible cost and benefit factors intimidating, at best.

By making certain simplifying assumptions, however, you can drastically reduce the model’s complexity without significantly compromising its accuracy or usefulness (Refs. 1 and 2). One common simplification breaks costs into three categories: one-time costs; recurring costs, such as fixture construction and program generation; and ongoing costs of labor, facilities, and so on.

Most models implicitly assume that the investigation involves either a single manufacturing line or several identical lines. They generally include tester pricing as a one-time cost and assume that the
manufacturer buys one or more testers all at the same time, so the number of testers is irrelevant.

Unfortunately, lines may not perform identically, which complicates the analysis. In the multiple-line/multiple-site case, a manufacturer may buy one or more testers initially, then phase in others over time as production ramps up. Remote sites may have to buy testers as production transfers to them as well. Nevertheless, the effect on single-line or identical-line models is insignificant. As long as the pattern of tester purchases does not change, the initial purchase price and timing for all testers and associated infrastructure remains the same. Only expenditures that relate to a lack of consistent performance come into play here.

To ease transitions, contract manufacturers (CMs) stress the need for commonality of configuration from tester to tester. They equip a new line with the same number of test drivers, the same power supplies, and the same optional equipment as the original line had. Therefore, if programs and fixtures are developed on a fully loaded system, commonality of configuration would dictate that other machines in the pipeline be fully loaded as well—even if they would not otherwise need all of the features—which can add considerably to the cost of the multi-line approach. I will call this new recurring cost commonality cost. Note that pay-per-use arrangements can reduce this cost considerably, since manufacturers can bring testers up to code for compatibility without paying for features that they seldom, if ever, actually use.

If the appropriate equipment is already available, as when adding a new product to an existing line, the capital acquisition cost for the initial purchase is actually zero, although the commonality cost may not be. On the other hand, if to achieve line-balancing and other benefits a manufacturer must transfer the product to a different tester type or other platform, that step introduces another recurring cost (which may only happen once) that I will call the platform transfer cost.

Other affected recurring costs

The most obvious new cost covers additional work on fixtures and programs, which incurs costs for labor and materials. This cost is higher if tester configurations are not identical, so it trades off against commonality cost. Also in this category is additional startup time.

I assume most companies will account for these costs together as transfer startup costs, which manifest primarily as production delays. One prominent CM in the US estimated that these delays, including associated labor and materials, cost $1700 per hr per line. A manufacturer in Mexico quoted $650 per hr. Other contractors quoted costs both lower and considerably higher. I will therefore use these numbers as a good compromise for economic analysis (see Table 1).

Ramp-up to an additional line or transfer to another facility may require additional training to bring engineers, supervisors, repair technicians, and equipment operators up to speed. This factor did not come up much in discussions with OEMs and CMs.

Transferring a product from one line to another within the same facility may not incur much in the way of training costs because the relevant personnel would transfer as well. I also could consider the cost of training additional operators to handle expansion to additional lines within the same facility. Training costs at a new facility may prove even more significant, especially for an unfamiliar product arriving at the facility for the first time. In the example here, I assume they are small enough to ignore.

I assume identical end-product quality between lines or facilities. Lower product quality would incur additional repair and retest costs, possibly test and monitoring costs, and scrap costs. Ignoring these costs keeps calculations conservative, as well as simpler. Also, maintenance, spares, and downtime will vary proportionately with the number of systems purchased. Therefore, the analysis for multiple lines will not change from the single-line case unless they result directly from matching line-to-line configurations.

Ongoing costs

Most ongoing costs remain unaffected from single-line or identical-line scenarios to scenarios involving multiple, independent lines, with two prominent exceptions:

•    False failures—If tester-to-tester transfer is not stable, then false failures become more likely. The magnitude of this rise will depend on the degree to which the test program is tuned to correspond with the idiosyncrasies of the new production line. Generally speaking, the more time spent on (and the higher the cost of) the transfer itself, the lower the increase in false failures.

At in-circuit test, after a specified number of cycles through the loop the board generally passes to the next test step. This strategy assumes that either the failure wasn’t there at all or—if it is a real failure—that a subsequent test will find it. The additional false-failure cost for the unnecessary repairs covers the cost for repair labor and for any devices replaced during the repair step.

Also, inventory levels rise because the falsely failing product remains in the repair-and-retest loop. Holding inventory incurs costs in higher asset value and more necessary warehouse space. In most cases, however, this additional cost is small enough to ignore.

•    False passes (escapes)—If instability from the migration forces the engineer to widen test tolerances to avoid false failures, some bad boards will escape. Escapes require fault detection later in the process, increasing repair and retest costs.

Some available software tools can provide information about the anticipated impact of variations among test heads. They can look at variations in test speeds, analog and shorts thresholds, and timing differences, suggesting wider tolerances to reduce the number of false failures, and thereby increasing escapes. The test engineer must draw the appropriate line.

Note that I’m assuming that differences between testers of the same make and model are sufficiently small to leave board scrap unaffected. This assumption is reasonable in most cases. Higher board scrap will skew the results of the economic analysis in the direction of the more stable tester. Therefore, this assumption is either reasonable (as in most circumstances) or conservative.

Cost Factors

In applying these ideas and assumptions to a quantitative model, you must assess the following factors (since most companies budget and plan on an annual basis, you can consider activity over a one-year period):

•    number of programs and fixtures affected,
•    number of changes to those programs and fixtures (product updates, debugs, etc.),
•    frequency of transfer and expansion to other lines within that facility,
•    cost per line transfer,
•    cost per line expansion,
•    frequency of transfer and expansion to other lines in other facilities,
•    number of lines that must run a particular fixture or program,
•    cost per facility transfer,
•    cost per facility expansion,
•    available fixturing (Does fixturing have to be modified each time the product moves?),
•    cost per engineering change order and number per year, and
•    cost for response time.

I spoke with one CM who said a calculation of downtime must include not only the time needed to implement the transfer but also the time spent waiting for an expert to arrive to fix problems. This cost is aggravated when the expert is far away, as is often the case with manufacturing operations in Latin America or East Asia. In these cases, the delay can take days.

Some factors that may affect transfer are difficult to quantify. One CM talked about a reluctance on the part of some OEMs to reveal sensitive product information to the CM for fear of losing jobs. In other cases, the CM becomes part of the design team. Where the OEM does not provide all critical proprietary information about the product, the transfer penalty will be higher. That is, the more test engineers know about how the product is supposed to perform, the more easily they can adjust to whatever peculiarities a tester-to-tester transfer uncovers.

Running the numbers

One CM estimated line-to-line transfer delays at 4 to 5 hrs on one tester type as against 1 hr on another. Site-to-site delays totaled 1 week and 1 day, respectively. Based on the quoted $1700 per hr for downtime, line-to-line transfer startup delays cost $8500 and $1700, respectively, for the two tester types (Table 1).

If a single tester and its associated infrastructure costs $200,000, the higher line-to-line delay cost represents a difference of $6800 or 3.4% of the purchase price. Each site-to-site transfer costs $54,400 or 27.2% more in the high-delay case. At the lower level, the numbers would be $2600 (1.3%) and $20,800 (10.4%), respectively. Clearly, when comparing alternatives, even a modest number of anticipated changes can significantly alter the decision point. Therefore, in analyzing costs, ease of such transitions must be considered. T&MW

References
1. Scheiber, Stephen F., A Six-Step Economic-Justification Process for Tester Selection, Quale Press, Florence, MA, 1997. ISBN: 0-9656161-0-X.
2. Scheiber, Stephen F., Economically Justifying Functional Test, Quale Press, Florence, MA, 1999. ISBN: 0-9656161-6-9.