Using Data Analytics More Effectively

The semiconductor industry is under a lot of pressure from their customers nowadays. They’re expected to keep up with consumer expectations for shorter electronic product life cycles, without compromising on the reliability and quality of the components and products coming off the line. A recent article from McKinsey & Company, however, describes how quality procedures have become a bottleneck in the drive to reduce time-to-market:

“Lead times for bringing integrated circuits to market have been gradually rising with each node. New design and manufacturing techniques account for some of the increase, but more complex inspection, testing and validation procedures also create delays.”

The McKinsey piece also goes on to discuss how advanced analytics and machine learning are critical to streamlining and optimizing manufacturing and testing processes in the semiconductor industry. In this blog, we focus on the challenges of test data analytics in the semiconductor industry, and we introduce OptimalPlus’ new Test Data Analysis service that can empower in-house teams.

Test data analytics: the challenges
Inspection, testing, and validation procedures in the semiconductor industry have become more complex because they have to meet the ever-increasing requirements and complexities of the products being delivered. In order to open up design and manufacturing bottlenecks, foundries and IDMs as well as fabless companies are all seeking to automate these procedures to the greatest extent possible.

Semiconductor manufacturers invest in equipment and platforms to collect data at scale and at velocity across the production, test and assembly floors. However, analyzing this big data in order to extract value in terms of improved yield, efficiency, and quality involves a number of challenges:

Data analytics and machine learning: their critical role in semiconductor manufacturing and testing.

Despite the challenges, data analytics and machine learning are being embraced across the semiconductor industry in order to achieve levels of prescriptive and predictive analytics that can reduce manufacturing costs and accelerate revenue growth. In light of this, IDMs and fabless companies are harnessing analytics and machine learning to:

Extending in-house teams
Semiconductor companies should consider extending their data analytics capacity and capabilities by consuming Test Data Analysis as a Service—a new approach that augments and extends in-house quality, product engineering, and data science teams. The service can enrich existing internal knowledge and methodologies with new best practices.

How does the Test Data Analysis service work?
The customer uploads a specific set of product test data into a secured folder. Several weeks later he will receive a detailed report identifying areas of potential improvement in yield, efficiency, and quality. The report also provides practical recommendations for addressing the issues identified, including:

• Yield: Detection of multiple yield issues and how to improve them, such as setting optimal test limits, using wafer sort data to predict edge fallout at Final Test results, and optimizing retest policies.
• Efficiency: Analysis of test equipment performance (pause time, site-to-site, tester-to-tester, etc.) and overall throughput across test fleet and global operations, with suggestions of how to streamline and improve testing processes and products.
• Quality: Leveraging a powerful combination of deterministic and statistical analytic methods developed through running analytics for 70+ billion devices each year, Test Data Analysis as a Service identifies quality issues such as tester freeze, test escapes, over-probing and outliers. The service also suggests ways to improve the identified quality issues.

Fabless and IDM companies that do not have dedicated tools or currently use a variety of tools can benefit from the incremental knowledge and expertise introduced through this new service that is based on advanced data analysis and machine learning.