Economies of scale

Seyed Paransun, VP of Amkor Technology's test business, commands more than 800 test systems.

Read other articles from this issue: Table of contents, February 2005 Economies of scale, Cover story Test-system development:    Do everything first Vision meets motion Simulate voice networks

The answer to the first question is easy, according to Seyed Paransun, VP of the company's test business: "Essentially, it's less expensive to outsource than internalize test because of the cost and utilization of test equipment. If our customers buy testers and dedicate them to specific product lines, they are essentially slaves to market conditions and the success of their products. When they outsource test, they only pay for what they use, and I, as an outsourced provider, can mitigate the risk across multiple markets as well as multiple customers."

To gain insights to the second question, I visited the engineering team at this Chandler facility, which focuses on RF assembly and test and is on the road to becoming the company's center for RF test excellence, according to Sriphani Vissamraju, senior RF test engineer. To that end, he said, the team relies on RF-capable proprietary as well as commercial ATE while providing expertise in areas ranging from the solid modeling necessary for RF package, load board, and contactor design to RF-test software development.

With 5 million ft² of manufacturing space, Amkor, Paransun said, employs 22,000 people in 16 locations, with test systems available at each of its production facilities, located in China, Japan, Korea, the Philippines, Singapore, Taiwan, and the US. The company fields 2045 people worldwide who focus on test, and it deploys more than 845 testers at its production facilities as well as at its engineering operations here and in Santa Clara, CA, and Wichita, KS. The 845 figure includes nearly 60 distinct ATE models, ranging from venerable Sentry systems from Schlumberger (now part of Credence Systems) to state-of-the-art equipment including Advantest T6675, Agilent Technologies 93000, Credence Sapphire and ASL 3000RF, LTX Fusion CX, Nextest Maverick, and Teradyne Flex systems.

Paransun places the 2003 OSAT market at $6 billion, or about a quarter of the overall $26 billion assembly and test market. (The remaining 75% represents traditional IDMs' portion.) He expects the OSAT market to grow to $13 billion by 2007, representing 30% of the total.

Jeff Luth, Amkor VP of corporate communications, noted that the test portion is the fastest growing OSAT segment, and he commented that IDMs and even fabless companies have been possessive about their test, even when outsourcing assembly. Now, such firms are more comfortable taking advantage of outsourced test, he said, and are more than willing to benefit from the reduced cycle times, and hence lower costs, that turnkey assembly and test services can provide. Paransun added that Amkor doesn't offer test services alone but only as part of its turnkey assembly and test services.

Amkor's RFt handles signals to 6 GHz.

Paransun also explained that an OSAT is likely to have production facilities located near its customers' customers, which can help control costs by reducing cycle times. "Most of my customers' customers are in Asia, where my factories are, so immediately after test we can drop ship directly to their customers to help with cycle time."

In addition to location as one issue in choosing a subcontractor, Paransun said, the subcontractor's equipment line-up is also critical—mainline equipment that's well supported by the subcontractor will provide the best costs and cycle times. He also recommended that a prospective OSAT customer evaluate a subcontractor's test-engineering expertise. "The depth and the breadth of engineering expertise for a subcontractor is a very significant piece of the equation even though a customer may not outsource test development. It is key that a customer engage a subcon with strong engineering, because inherently in the test process there are issues such as product sensitivity to fab processes that have to be addressed quickly so the production doesn't come to a screeching halt."

Daniel Adney, test integration engineer at Amkor, ensures proper docking between an ATE system, such as this Credence Systems ASL 3000RF, and a handler.

He said it is of key importance for a customer's equipment to match an OSAT's equipment roadmap with respect to test systems as well as handlers. "You'd be surprised how often the link is broken here," he said, "where the customer develops a test program based on a tester, a handler, or a load-board design that can't be supported in high-volume manufacturing. All of a sudden, the customer has to go through a re-spin process. That's quite a costly endeavor that can take up to three months."

Of course, the use of an OSAT requires that the customer and OSAT exchange a great deal of data—ranging from new test-program releases to customer access to production data. "We provide access so that our customers in the US can actually log on to a tester that we are running production on in Korea or the Philippines, and they can actually monitor yield and—more important—debug problems online in their offices. So, IT is a big deal for us. We have a program set up so if the yield falls below a certain target, a computer automatically calls an engineer here in the US and leaves a message: 'this lot experienced low yield and is on hold,' for example."

In addition, Paransun said, a subcontractor has to be able to offer full service at each location—from qualification to failure analysis—so material doesn't have to move between organizations and locations.

"In terms of ATE selection," he said, "our recommendation for customers is to select mainstream platforms. Unique platforms are not supported well—I don't have critical mass around them, so I won't be able to provide upside capacity and manage the nonlinearity" of the customer's fluctuating production rates. Furthermore, he said, a customer "should select platforms and equipment with the turnkey price in mind. Sometimes, customers lose sight of the overall cost and err in the direction of specifying a cheaper tester, and they end up paying more for it in the long run. For instance, I have several testers that are superset testers—multiple pin count, multiple resources—and that allows me to do multisite testing. So, even though the hourly rate for the tester is higher, the eventual cost per unit is significantly lower. We want to make sure the customers keep an open mind about that." Also, a mainstream handler with a very fast index time may look expensive at first, he said, but can again result in a lower price per unit because of lower per-unit test costs.

Similar factors apply to load boards and contactors, he said. "It's important that the test-head configuration be considered during the layout of the load board so there's no need for editing when the load board goes into production. Contacting is a very big deal for us when we go to higher speeds and higher ball counts. This year, we are going to test about 4 billion units. If you think a contactor is good for about 400,000 contacts, you can do the math and figure out how many contactors I have to buy." It represents a significant cost, he said, and is one area where a mainstream component might not be the best choice. "If my customer wants to specify ECT, Johnstech, or Synergetix, that's no problem, but we may be able to offer a more attractive alternative."

Of course, sometimes customers choose to provide load boards and test programs that they have developed themselves. In that case, said Paransun, a transfer process takes place to ensure that what the customer provides is production worthy. "Nowadays, test programs are quite complex," he said, and often his engineers can identify code segments that could cause production to be slower than necessary or to completely grind to a halt. "I want to know whether any preproduction data they provide was collected from 40,000 or 50,000 units or only a few. We want to make sure the software is optimized for very fast test times. Usually in the engineering environment, programmers are lenient—they allow ample settling time for DC tests, or they add a lot of tests geared toward silicon verification and characterization, so we want to go through the process with them and really clean up the test program" to optimize throughput and cost.

Similarly, Paransun added, "We also want to look at customer-supplied load boards, probe cards, and contactors to make sure they are robust as well. I can't afford to have a $2 million tester sit idle while someone chases a $1000 socket."

Vissamraju, the senior RF test engineer, explained, "The solutions we offer here include test-program development, contactor design, and load-board design as well as tuning and debug. We also develop a correlation procedure based on what the customer is looking for—for example, 50 devices run 100 times each or 20 devices run 10 times each at different temperatures. In addition, we need to know exactly what degree of correlation they expect between their bench measurements and our measurements here." Next, come verification and low-volume characterization test, followed by factory handoff.

Figure 1.  Development engineering efforts at Amkor extend from the assignment of resources through the transfer of hardware and software to production facilities. Software documentation packages include customer specs, test listings, and sample data logs; hardware transfer involves items such as load boards, probe cards, and handler kits. After release to production, production engineers typically take responsibility for any additional test optimization.

Figure 2.  A typical project at Amkor might require about 11 weeks from customer input to release to manufacturing, as hardware and software development overlap.

Figure 3.  Solid-modeling tools such as Pro/Engineer software help Amkor engineers develop load boards, test sockets, and device packages.

Paransun commented that "In RF, quality of mechanical contacting is really key. That's why design of the load-board stiffener is so pivotal to successful load-board operation. Minor fluctuations can make a big difference. The value here is ownership. When customers engage us, we take ownership of the whole process. They know the buck stops here. Problem resolution—that's what we bring to the party."

As an example, he described one effort at RF design that involved four test sites right on the load board, but that caused some matching problems across sites because of different electrical lengths. To eliminate the trace-length problems, he said, the team turned to cabling and daughter-card designs. He noted that a load-board vendor may have been able to solve the trace-length problem, but that vendor may have developed a load-board solution that was not optimal from an overall cost perspective; similarly, a docking specialist can tweak only the docking hardware. Paransun concluded, "There is value in our ownership of the whole process."

* On semiconductor test, visit www.tmworld.com/ic

* About the test vendors mentioned in this article, visit our Buyer's Guide, www.tmworld.com/BG