ML Model Usage For Various Life Stages Of Semiconductor Test

By Shinji Hioki and Ken Butler

From development through high volume manufacturing (HVM), semiconductor manufacturers’ pain points change based on the life stages. Each stage requires different types of applications to help with business needs. At the early stage, where the design and process are still immature, understanding the root causes of maverick material and implementing fixes is the focus to accelerate time-to-market. Delaying qualification can result in lost business opportunities, especially in competitive business situations. Once product qualification is done, production usually starts with lower yields and unoptimized test flow/test contents. By running enough fab lots, test flow and test contents can be optimized, so cost reductions can be achieved without sacrificing quality. More efficiency gain is achieved through continuous improvement with stringent change control. After yield reaches the target level, production runs in a stable manner with occasional excursions. The focus in this stage becomes excursion management and quick feedback to the fab. A combination of necessary applications can change based on the life stages of the product. Understanding these semiconductor product life cycle gives us more insights to produce meaningful ML applications in the future.

Another primary reason why AI/ML is needed is to catch reliability failures during production tests. However, prior to utilizing AI/ML, the data must be trustworthy before being used for ML model training. Before working on test data analysis, one also needs to ensure the repeatability and reproducibility of the test data by going through MSA (Measurement System Analysis) process (= Gage R&R).

As the industry continues to innovate, we are finding that customer’s pain points are constantly evolving. Each life cycle is different, thus one model does not fit all. Understanding each product life stage and its inherent challenges is important to select the most effective applications. The combination of improved measurement and data analytics will advance the semiconductor test industry to assure better quality and reliability, faster time-to-market, and improved resource utilization.

Effectively utilizing ML is one area where Advantest is revolutionizing the test industry. Advantest’s ACS Real-Time Data Infrastructure (ACS RTDI) is a solution that offers advanced analytics, including machine learning capabilities and future-proof, real-time, automated production control. The Advantest ACS ecosystem integrates all data sources across the entire IC manufacturing supply chain while employing low-latency edge computing and analytics in a secure True Zero Trust environment. This innovative infrastructure minimizes the need for human intervention, streamlining overall data utilization across multiple insertions and supporting customers’ databases. Because security remains a top concern among customers, the ACS RTDI platform has been architected to be reliable and safe, ensuring hassle-free OS revisions, while protecting data from unauthorized access or loss. This is accomplished by leveraging True Zero Trust. Overall, the new ACS ecosystem is designed to enable customers to boost quality, yield, and operational efficiencies, and to accelerate product development and new product introductions for years to come.

Ken Butler is senior director of business development, ACS, at Advantest.