Standards for Test

Most test data standards that exist today either comply with or are derived from the binary format "STDF" introduced by Teradyne. Several automatic test equipment (ATE) companies, including Verigy, LTX, Credence and others, support this format, and a small industry of analysis software tools accept STDF format. STDF is a market-driven standard, in that Teradyne introduced it to form a cohesive output format for their UNIX-based test equipment. However, its use is not universal by all test equipment manufacturers, and there is enough "flexibility" in the definition to create situations where STDF from one platform is different from STDF from another platform.

Historically, semiconductor test has been just one part of the flow of an integrated device manufacturer (IDM), so most of the large consumers of ATE have developed their own in-house proprietary data flows and analysis methods/tools. Now that the industry is "disintegrating," a lack of standards hinders yield learning, design debug and so forth. The IC design comes from one company, another fabricates the wafers, and the test can be outsourced to any one of a number of vendors. Making sense of the data that comes out of test is difficult enough with such a diversity of players, let alone a lack of agreement of what the data should "look like" and how the data should flow. How does a fabless semiconductor company learn about its systematic circuit design issues when its silicon is fabricated at a foundry, its test is done at two different test houses, and the data systems for all of the above do not "understand" each other? How does that foundry learn yields with this product?

The most difficult problem to solve involves diagnosis of functional failures. From the standpoint of the test equipment, the failure consists of a failing pin, logic state and clock cycle (vector). Getting that information to the point where failure analysis is possible is a monumental task. Somehow, the functional fail data needs to be "understood" by the tools that performed the test generation in the first place. At present, most systems that flow these data and perform these types of diagnoses are highly customized and non-transferable.

Besides these basic questions, there is growing interest in software systems for the disposition of devices after test or adaptation during test. In this paradigm, the test equipment becomes more of a "collector of information" rather than just a red light/green light system. Interesting, new ideas for both during and post-test decision-making (statistical post-processing, adaptive test, etc.) are hampered by a lack of a standard for data reporting. Any software tools that implement these types of innovations can become necessarily custom and difficult to transfer to other test IT domains. Integrated device manufacturers have successfully implemented such systems in their own IT infrastructures, but the rest of the industry lags behind.

In a vacuum of standards definition, de facto standards emerge through market forces. An ATE company publishes a standard and sells a number of tools, so that data format takes hold. A "test data analysis" company makes popular software that interfaces to that format, and some facet of the IC manufacturing community "endorses" this tool through sales. A large test house installs a base of the combination of equipment and analysis software, or a large IDM player forces the matter through its purchasing power.

The trouble with this type of development is that it is parochial in some sense; no single ATE company, IDM or test house has all of the "answers" about what might be needed in a test data standard. Standards making is a difficult process because it is hard to come to agreement on low-level details no matter what the topic. However, innovation in the domain of IC test will constantly be held back in the absence of such standards. And in the present economic climate, where the pieces of IC design, manufacture and test are becoming more separate and less integrated, standards for data are the essential glue for making it all work.