Semiconductor Vendors Mine Big Data for Quality

David Park, Vice President of Worldwide Marketing, OptimalTest

The semiconductor manufacturing industry is continually driving toward higher quality, said David Park, vice president of worldwide marketing for OptimalTest. Historically, quality has been important for applications in the medical and defense and aerospace industries, he said, but now the drive for quality is penetrating other areas—automotive, for example, where cars are becoming rolling supercomputers.

Further, he said, “The perception of what is needed from a quality perspective now is reaching down to the consumer space, where the expectations of flawless execution of your smartphone, your tablet, your wired TV set are reaching significant heights. The user experience and functionality are expected to be exceptional.”

In pursuit of quality, semiconductor companies perform extensive tests and collect terabytes of data. Nevertheless, they face significant levels of return-materials authorizations. But it turns out, Park said, that 32% of “RMAs” are not bad. Only 28% are bad due to fab process problems. The remaining RMA sources include test equipment (26%), test programs (10%), and test operation problems (4%).

“Basically,” Park said, “there are a lot of different ways parts can come back to customers, which obviously impacts the perception of quality in the marketplace, and it also affects vendors’ profitability because it’s extremely expensive to process RMAs.”

Vendors can address RMA problems by focusing on four areas, Park said. First is big data. “Obviously there are just enormous amounts of parametric data that are available for companies to analyze, but it comes in so quickly and the volume is so high, and a lot of companies simply don’t know how to deal with the sheer volume of information.”

That leads to the second area—expertise or having people who cannot just find test program errors or test site issues but can perform the complex analysis necessary to optimize production and test operations. “That obviously takes a more sophisticated level of analysis than has been historically applied to this problem,” Park said, adding, “That’s not just because they didn’t want to; I think a lot of people didn’t realize it was even possible.”

The third area of focus is time, Park said. “If it takes you a year to do this analysis, that doesn’t really help you a heck of a lot.” The year might be an exaggeration, he said, but days or weeks of data mining to identify problems are common. And the fourth area of focus is cost related to the manpower that has to be applied in finding potential problems in the aggregate data. “It’s a pretty big challenge for a lot of the semiconductor companies to find that hidden gem in all the test data they’ve collected.”

OptimalTest, said Park, helps companies by automating the process of finding test-related issues to enhance product-test yield recovery, increase throughput, optimize test-cell efficiency, and reduce RMAs. With OptimalTest tools, he said, engineering teams can focus not on finding problems but on resolving them and improving the overall product output.

“The solution that we are bringing to bear for this marketplace,” Park said, “is basically a three-pillared approach. First, we bring in complete and accurate data sets from global manufacturing and test operations across the globe for both fabless companies as well as IDMs.”

He continued, “Layered on top of that, we provide the tools that actually can go through and proactively mine these very large data sets to find problems.” The first two pillars, he said, “combine with the expertise of our field team in not only knowing how to find the problems that are fairly common across all semiconductor vendors, but also in helping tune the rules based on specific customer needs. Part of our solution is out of the box in terms of the rules we use to look for issues in the manufacturing data, but we also can customize those rules specifically for customers to do things that are unique to their operation.”

In the future, customers will be able to do their own customization. As this article went to press, OptimalTest announced Release 5.5 of its OptimalEnterprise solution. This release not only incorporates new capabilities for superior outlier detection, automated ATE configuration, and enhanced part traceability, but also includes a new data platform, called Sequoia, which enables test and product engineers to tailor their environment based on specific manufacturing requirements. Visit www.evaluationengineering.com for more.

More in Instrumentation