Results from last month’s study on registration using closed-loop control.

Ed.: Last month’s column reviewed a design of experiments (DoE) of print registration control. This month we discuss the results.

Evaluation tests. Figures 6 and 7 show results from the evaluation tests. Results from the rear to front direction are shown with two different correction factors. The graph shows that for 50% correction factor, it takes four boards to reach the target registration. On the other hand, for 25% correction factor, it takes about eight boards to reach the target registration. Predictably, smaller correction factor holds the offset value at a much tighter range. Based on these results, a 50% correction factor was considered adequate and was used for the remainder of the tests.

Baseline test. Figures 8 and 9 show results from the baseline test for F2R stroke direction only. R2F stroke direction showed similar behavior. As is clear from the plots, both X and Y offset fluctuate around a fixed spot that which is not zero. Additional analysis comparing the baseline results to the long-run result will be presented later.

Long-run test. Results from this study are presented in Figures 10 to 13. From these results, the closed loop control algorithm is capable of reaching the target value rather quickly. Additionally, it is observed that the average registration value can be maintained close to the target for the entire duration of the test.

Print performance comparison between with/without closed loop control gives a measure of process improvements produced by the closed-loop control system. Comparison of the process capability index, Cpk, for X and Y offset measurements for the cellphone board is shown in Figure 13. The increase in the Cpk for both X and Y offset with control is due to the centering and tightening of the print process. Additional statistical analysis confirms the improvement of print performance by employing the closed-loop process control. This analysis is presented in Figure 14. Figure 15 shows the individual moving chart for Y offset with both with/without the control. It is clear from this analysis that both mean and control limit improve for process with the closed-loop control.

Closed-loop controls have been implemented at many stages along printed circuit board manufacturing lines, including within reflow ovens and component placement.

Miniaturization will drive closed loop-process controls for printing slowly but surely into assembly lines. Closed-loop controllers, when implemented correctly, have advantages of keeping complex processes within control limits, even when small external perturbations affect the product line. In addition, closed-loop control minimizes operator intervention and has self-tuning properties.

Limited controlled experiments in a laboratory environment show an improvement in the print process capability with the closed-loop control in place. The full extent of the benefit can only be accessed by employing such a system in a true high-volume production environment.  CA

Acknowledgment
The authors would like to express their sincerest gratitude to Paul Haugen from CyberOptics for continued support in operating the SPI system and insight into inspection system in general. Without his support, this project wouldn’t have been successful.  

References
1. Rita Mohanty, “In the Loop,” Circuits Assembly, March 2009.

Rita Mohanty, Ph.D., is director advanced development at Speedline Technologies (speedlinetech.com); rmohanty@speedlinetech.com.

New measurement technology enables high degrees of verification and confidence.

Since the dawn of SMT, the solder paste printing process has been considered the primary source of assembly defects. The process itself is an enigmatic combination of art and science, and is not easy to control. As surface mount technology evolved, a variety of solder paste inspection tools progressed, often employing the latest advancements in optics, lasers and machine vision systems. But when miniaturization efforts produced 0.5 mm area array devices, the incumbent inspection technologies all fell short, either on accuracy or cycle time, or both. The ability to read 0.010˝ deposits well enough to be trusted and quickly enough to meet production requirements became the Holy Grail in automated paste inspection. That’s when phase shift interferometry, a light and vision-based technology, was introduced as an inspection method, and changed the way we look at solder paste.

How does it work? Simple physics. Imagine sun shining through window blinds. The sunbeams form a pattern of stripes, alternating light and dark. These stripes appear straight on planar surfaces such as floors or tables, but seem to jog back and forth as they cross over furnishings that alter the surfaces’ flat topographies. The amount of perceived zigging and zagging depends on the height of the object and the relative angles of the observation and source light. By taking several measurements at known intervals, the height of the surfaces can be calculated mathematically.

The alternating light and dark stripes are known as a Moiré pattern (Figure 1), and the method of measuring them at known intervals is called phase shift interferometry (Figure 2). The Koh Young solder paste inspection system utilizes a patented, sophisticated application of phase shift interferometry to measure solder paste deposits. It essentially divides each paste deposit into tiny segments as small as 10 µm, and measures each segment’s location and height (Figure 3). It then calculates the individual segments’ volumes, and constructs a 3-D model of the entire solder paste deposit (Figure 4). The measured model’s volume, height and/or location are then compared to a theoretically perfect model. The modeling is unprecedentedly accurate, demonstrating a gage repeatability and reproducibility (GR&R) of less than 10%.

Vicor Corp., a manufacturer of power components, uses the Koh Young inspection equipment on its prototype and production lines. Sometimes referred to as “bricks” by assemblers, Vicor’s power management components are anything but brick-like on the inside. These packages contain some of the smallest and most complex SMT packages available: 0.5 mm area array devices, leadless packages like QFNs and SOT883s, and 0201 passives. The boards on which they are assembled are panelized in 10- to 20-up arrays, and a typical solder paste print consists of over 10,000 deposits, most of them smaller than 0.012˝.

Ten thousand paste deposits in the 0.012˝ range is not an impossible process, but it certainly is a challenging one. Ray Whittier, senior process engineer, is charged with keeping it in control. He uses the phase shift inspection technology extensively, both online and offline. Online process control includes screening prints for quality and optimizing throughput. Offline analysis includes stencil qualification and verification, paste qualification, and solder powder size selection.

The best place to begin employing the technology is in production. Implementation begins with programming a board, which takes about 10 min. The system’s computer creates a virtual “golden board” to which it compares actual print readings. It learns the theoretical stencil aperture sizes and locations from the stencil’s Gerber file; it learns the theoretical Z-height from user input, and it learns the reference plane by reading a bare board. The programmer can set a number of inspection parameters, including allowable deviations from the golden board’s stencil heights, stencil volumes, or aperture positions. Typical starting parameters are 50% minimum and 150% maximum of the stencil height and calculated aperture volume, and 50% center-to-center deviation from paste to pad.

To inspect print quality, the system scans the printed board and creates a second model in its computer. It takes less than 20 sec. for it to read over 10,000 paste deposits, construct the model, and compare the measured solder paste deposit volumes and locations to those on the golden board. If every deposit falls within the programmed specifications, the system passes the board on to the next operation. If one or more deposits do not meet the set criteria, however, the system fails the board and alerts an operator.

Whittier cautions that a failed print may not necessarily be a poor quality print. It may be an indication of variation among bare boards. “Critical features like circuit card dimensions, pad sizes and locations, and solder mask alignment each can vary substantially from lot to lot. A failed inspection may be the result of a bad print, or it may be an indication of a change in the bare boards. To distinguish between the two, operators run a bare board to reset the reference plane and scan the print again. If it passes, it is returned to production. If it fails, the print is scrapped. Many of the scan fails are the result of PWB tolerances, and are resolved by the bare board re-teach operation.”

How many actual print failures are captured? That depends on the complexity of the different products and the mix on the production line. Figure 5 shows the percentage of scrapped prints for top and bottom sides of the boards. The less complex bottom-side averages around a 1 to 2% failure rate, while the more challenging topside runs around 6 to 7% failures for the current product mix, with one monthly spike topping 12%. Had all these bad prints continued down the production line, they would likely have created expensive defects. Reworking BGA and QFN components is costly and risky for most assemblers, but for Vicor, rework is not an option. Failed assemblies are scrapped. With no chance for after-the-fact recovery, it is absolutely essential to catch defects at their point of origin.

How is overall process control instituted? The operating procedures are very specific regarding failures. If three true print failures occur sequentially, the process is paused and support is called out to the line to investigate the source of the failures. Interestingly, nowhere do the procedures call for visual confirmation of the machine-identified defect. Pass/fail decisions are based purely on the output of the machine. Whittier explains that “with a GR&R consistently in the 7 to 8% range, the machine is far more accurate than the human eye. It makes absolutely no sense to override a superior system with an inferior one.”

Boosting Yields and Cutting Costs
What is the end of line impact? For Vicor, overall yields climbed 3% upon implementation of the systems – a considerable and noteworthy boost. The bottom line also was improved by cost reductions and utilization improvements. When Vicor recently changed solder paste formulations, Whittier used the bridge detection feature to dial in the wipe frequency in production. He found he was able to double the wipe interval from every print to every other, halving his cost of wiper paper and the down time associated with changing it. He was also able to identify the pause time at which paste kneading is necessary to ensure good prints, eliminating unnecessary print rejects, kneading operations and board cleaning.

What is Whittier’s favorite feature of the phase shift interferometry equipment? “The combination of speed and accuracy. Every process engineer knows the tradeoffs between the two. In this machine, there are none.” His favorite unadvertised application? “Stencil verification. We now approve stencils for production faster and more accurately than ever before, as we can catch missing or improperly cut apertures in one reading” (Figure 6). His next future app? “DfM feedback to the design teams. There are no off-the-shelf DfM guidelines for the type of product we make. Using this inspection technology on our prototype lines alerts us to potential stencil printing issues while there is still time to design them out of the product.”

As for offline applications, Whittier was able to quickly characterize the transfer efficiencies of solder pastes under multiple print scenarios. The data he’s generated have allowed him to understand the impact of stencil suppliers and manufacturing processes on paste printing, resulting in simultaneous quality improvements and cost reductions. It’s provided him with a solid comparison of current and new solder paste formulations to upgrade his materials set and reduce variation in print quality. And it’s given him guidelines as to which feature sizes will require a switch to Type 4 solder paste so he can plan the timing of its introduction to meet Vicor’s product designs.

Imagine increasing an end-of-line yield by three percentage points. That alone is worth the price of admission, as it directly improves profitability. Cutting costs, improving throughput, qualifying suppliers, preparing for next-generation technology – they all improve profitability also, but less visibly. And, unfortunately, they’re still sometimes forgotten.   CA

Chrys Shea is founder of Shea Engineering Services (sheaengineering.com); chrys@sheaengineering.com.

Legacy products are prone to fake parts. How to mitigate.

Ensuring component integrity is an increasing challenge for electronics manufacturing service providers focused on industrial products. Short consumer product lifecycles, the electronics industry’s conversion to lead-free components, and general weakness in the global market have increased the speed at which components are becoming obsolete. At the same time, industrial products often mature into legacy status due to the cost of redesign and qualification that drive very long life-spans.

In some cases, OEMs assume their best option is to avoid the cost of redesign and task their EMS providers to take reactive measures to source through independent brokers as components become unavailable. Ultimately, this can be false economy, because when a component goes obsolete, many hidden and unexpected costs creep in over time. These risks can include:

• Inability to support market demand if components are not available.

• Purchasing counterfeit parts through unknown independent distributors if component supplies are extremely limited.

• Higher fallout from using legitimate parts with older date codes.

• Higher fallout from parts that have been reworked and sold as new through the gray market.

• Disappointed end-customers if parts pass in manufacturing test, but fail prematurely in the field.

Best Practices
What are the alternatives to reactive procurement practices in supporting legacy product? First, establish a plan and team with both the EMS provider and key manufacturers within the component supply chain to address obsolescence as early as possible. Second, since material costs will increase over time on a long lifecycle product, periodic redesigns should be budgeted and planned. Finally, since some purchasing of components through independent distributors is likely, it is important to establish relationships with the best suppliers – ones that will provide some level of guarantee that the source and integrity of the parts provided are reliable.

For example, one of our customers scheduled a revision spin of a complex industrial process control board. Before the redesign was finalized, we performed a lifecycle analysis of the components and identified nearly 20 parts at risk for obsolescence. The customer analyzed the alternate component suggestions and adopted about half. The remainder was left unchanged because substitutions would have required a complete re-layout of the printed circuit assembly. While this type of team effort and planning added some time and cost in the short run, the end-result is much more time and money saved over the life of the product.

The Real World
Even when lifecycle analysis is performed and regular redesigns are scheduled, legacy products can still experience unanticipated obsolescence. Component manufacturers give good advance notice prior to ceasing production on their components to companies with scheduled demand. However, with low-volume legacy products, components may be purchased as infrequently as every 12 to 18 months. In those cases, the first notice that the part is obsolete may be received only when the order is placed.

If known independent distributors don’t have a critical part, the next step becomes searching lesser-known independent distributors, and this requires a strong teaming focus. The customer may have suggestions on distributors they have used successfully, and the EMS provider is likely to have a network of companies that are potential suppliers. Regardless of how potential suppliers are identified, the customer should maintain awareness of any untried independent distributors being used. Counterfeit parts do exist, and some of these parts are so well packaged that only destructive test sampling or high failure rates in the product will lead to their discovery.

Whenever unknown suppliers are used, the original source of components should always be ascertained. Components also should go through a more rigorous incoming inspection. If either of those activities generates concerns, sending a component to its original manufacturer may offer an additional measure of safety, because the manufacturer can easily compare part numbers, lot codes and date codes with its database.

When parts from an unknown source must be used and can’t be tested prior to assembly, producing and testing a small sample run may be a good way to uncover problems quickly. Even if parts are legitimate, issues such as age, improper handling or reworked parts sold as new may drive higher-than-normal failure rates. The sample run also can help set cost expectations if the higher failure rates are deemed an acceptable tradeoff.

When supporting legacy products, it is important to realize component-sourcing problems due to obsolescence will occur at some point. Similarly, counterfeiting is prevalent enough in the electronics industry that it is no longer a matter of “if” illegitimate parts will end up in your legacy product, but “when.” Robust incoming inspection, sample builds and testing can help when tradeoffs to standard procurement practices must be made. More important, establishing a plan with your EMS partner, good supply chain qualification and management practices, proactive product lifecycle analysis, and regular redesigns can help mitigate the risks to customers and the bottom line.  CA

Dennis Gradler is director of business development for Industrial Solutions at Kimball Electronics Group; dennis.gradler@kimball.com.

A trying year was dramatized by two unfolding stories: the economy and Elcoteq.

When it comes time to write the story of the EMS industry in 2009, foremost will be the effects of the recession. There is another story, however: the dramatic fall of Elcoteq, the former cellphone wonder whose sales have roller-coastered from a high of 4.28 billion euros (about $6.17 billion at today’s conversion rates) to an estimated 1.54 billion euros ($2.2 billion) at the end of 2009.

Last year felt in many ways like the nightmare of 2001-02 all over again. Factory shutdowns among Tier 1 and II EMS companies were the norm, with Flextronics, Celestica, Sanmina-SCI, Plexus, Benchmark, Jabil and Elcoteq all cutting plants and people. Most of the closures came in Europe or the US, where even mighty Foxconn shuttered a 400,000 sq. ft. PC assembly and enclosures plant in Fullerton, CA, relocating or letting go some 600 workers. (Most but not all: Flextronics, for example, shuttered its factory in Kuala Lumpur, the former Casio Computer site it acquired in 2002.) The differences between last year and the tech recession of 2001 were telling, however, in that EMS companies this time around acted with accordance with clear and aggressive strategies. It was as if, after the last go-around, the majors decided that if and when another industry recession hit, they were not going to react passively.  

Key moves during the year include Asustek’s decision at year-end to split off its Pegatron manufacturing group, creating overnight one of the world’s largest ODM companies, and Venture’s decision late in the year to push harder into ODM work, which many see as more profitable than traditional EMS services. But while analysts and pundits continue to predict (guess?) that one or more of the top-tier EMS companies would be bought or closed, all – with the notable exception of one (see sidebar) – appear to have learned from past mistakes. They kept their inventories in line, their balance sheets clean, hoarded cash and refinanced near-term debt in order to ensure survival. They also cut fast and deep.

Almost every major EMS company saw revenues fall last year. For most companies, the first or second quarter was the trough, and sequential gains were seen through the rest of the year. In all probability, a return to 2008’s revenues will be a few years away. Among the top 20 companies, No. 1 Foxconn was an exception – barely – with sales growing less than 1%. Another outlier, No. 7 Cal-Comp Electronics, actually grew year-over-year the first two quarters, then fell behind by 8% year-over-year during the third, despite turning in its best revenue period of the calendar year. No. 13 Beyonics, Singapore’s largest EMS company, was up big and will be banging on the door of the Top 10 if it can keep up the pace.

With no major acquisitions or mergers during the year, there were few big changes to the Circuits Assembly Top 50 list. Falling off this year’s Circuits Assembly Top 50 is one company that, in hindsight, never belonged. Viasystems, which is relisting its shares, and as such began to break out its revenues, was in retrospect nowhere near the $504 million in sales we estimated last year. Hitachi Computer Products America announced in late fall it would exit the electronics manufacturing services business (albeit much of its EMS work was likely for its parent company). The biggest shock, however, was Jurong Hi-Tech, which came in at 26th last year. The Singaporean EMS company imploded under accounting irregularities and sold off its assets. (Needless to say, it did not make this year’s list.) And our apologies to Team Precision, Eolane, EPIQ and Electronic Network, all of which should have made last year’s rankings.

Besides Jurong, the two companies that appeared hardest hit in 2009 were Elcoteq and Surface Mount Technology (Holdings) Ltd. The former took a body blow from Nokia, which pulled all its assembly work in-house, taking an estimated $5 billion worth of EMS work offline. As a result, the company’s revenues fell more than 50% last year. Unlike Elcoteq, whose fate is heavily tied to the telecom and handheld sectors, Hong Kong’s SMT Holdings operates in more balanced markets. The company’s revenue comes primarily from industrial controls, computer peripherals, consumer and automotive, each of which makes up 15 to 35% of the firm’s revenues. Since the company does not disclose its major customers, it’s hard to discern why its sales drop (roughly 45%) was decidedly worse than the industry average.

What is always interesting about tallying up the revenues of EMS companies is how inflated the sums are. Macroeconomists might disagree with me, but the amount of double-, triple- and quadruple-counting is staggering. The rule of thumb is about 80% of the assembly’s cost is for materials, and most EMS companies no longer buy on consignment, which means the value of single chip gets highly distorted as it makes its way through the supply chain.

Take your average PC. Intel makes the microprocessor and sells it for $250 to Foxconn. Foxconn builds the board and box and ships it to Dell. Dell sells the finished PC to the customer. Each company adds that $250 (or whatever markup it uses) to its revenues, which means that original chip has now accounted for $750 worth of sales. Throw in a retailer like Best Buy, and perhaps a distributor like Arrow and now that same chip is worth $1,250 in accounting terms by the time it reaches your desk. In doing so, the size of the industry becomes highly inflated. Not that the same scenario doesn’t play out in just about every sector imaginable, but it is something to keep in mind when considering the “real” size of “S” (services) in EMS.

[Ed.: For the pdf of the Top 50, click here.]

To sum up 2009, we learned that putting all your eggs in the basket of a consumer electronics customer could reap great dividends, or great hardships. Whether Elcoteq can return to glory or slips further into the abyss will likely be the story of 2010 too.  CA

[Sidebar] Elcoteq’s Miserable Year

No company took it on the chin worse last year than Elcoteq. Sales plunged about 55% year-over-year to an estimated 1,537 euros ($2.46 billion) and are now about 65% off their 2006 high. Global headcount was reduced about 50%, to 10,770. Factories in the US, Russia and China were either mothballed or shuttered. As if the economy itself wasn’t bad enough, its once largest customer, Nokia, pulled all its assembly in-house. As the year drew to a close, Elcoteq was scrambling to finalize a deal with Videocon to invest a reported 50 million euros to help shore up its balance sheet.

Circuits Assembly spoke with Elcoteq director, sales and product marketing Petra Ebner in November. Excerpts:

CA: You are in talks with Videocon to invest. What would this mean for Elcoteq?

PE: Videocon is a potential investor. We are currently in due diligence. The scheduled closing is year-end. It is not clear how this will look once a deal is done, but we’re not talking about them taking over. Elcoteq also is in talks with Kaifa [its largest shareholder].

We also have set certain goals: 1) Find an investor to increase our capitalization; 2) restructure debt – we are offering 15% of the dollar to our unsecured creditors; and 3) focus on more bottom-line customers. We are extending the market scope to higher-tech customers. Those end-products would be filters, RFID, industrial – companies that can benefit from our experience in communications.

CA: When your restructuring is finished, what will Elcoteq look like?

PE: We are now more focused on consumer electronics and systems solutions. We now have two factories in China; we closed Shenzhen. We have plants in Bangalore; Pécs, Hungary; Tallinn, Estonia; Monterrey, Mexico; Brazil. Russia has been mothballed. Pécs is the largest, with about 3,500 staff.

CA: Nokia decided to pull its assembly operations in-house. Any change in its plans?

PE: There are no signs from Nokia of business coming back.

CA: Given the product mix and choice of locations, it almost seems Foxconn has put a bulls-eye on Elcoteq’s back.

PE: Competing with Foxconn is a mission impossible. Customers are under so much cost pressure. Our focus is now the top line, not the bottom line. We want to grow with our existing customers, but we also want to see some different customers. Our focus is on lower volume and higher mix. When I came to Elcoteq, we had communications and industrial customers. I think we will return to that: more mid-sized customers; more box-build and tested finished product. Our target customer is $5 million to $10 million [in annual services provided].

CA: High-mix lines typically use different platforms than high-volume lines. Will your capitalization issues constrict your ability to invest in the right machines?

PE: We won’t need new equipment. In many cases, customers have supplied [it] as consigned equipment. We work with partners on plastics, etc.

CA: Is there an EMS company that Elcoteq is modeling its new strategy after?

PE: We haven’t picked anyone as a model.

CA: Is Elcoteq targeting alternative energy as a possible market?

PE: It’s not off our radar, but we are not thinking much about it. We are discussing with some solar companies about the equipment behind the panels – inverters, electronics, etc.

Elcoteq’s Sales, 2002-09 ($ millions)
2002    1840.2
2003    2235.7
2004    2,953.7
2005    4,169.0
2006    4,284.3
2007    4,042.9
2008    3443.2
2009    1537 (estimated)

Mike Buetow is editor-in-chief of Circuits Assembly (circuitsassembly.com); mbuetow@upmediagroup.com.