In the semiconductor industry, tool owners face growing challenges when managing the various data streams, volumes, and contexts produced by materials, processes and equipment. The adoption of scalable Equipment Digital Twins (DT) with integrated Machine Learning (ML) capabilities and monitoring applications presents a compelling solution to address specific quality, cost and delivery needs of tool owners as well as Fab automation teams alike. This co-authored talk by Tignis and onsemi will discuss the need and benefits from using a scalable ‘Designed for Semiconductor Equipment’ DT with fully integrated ML capabilities that can be deployed quickly in Fabs.
Fabs struggle with data silos and a lack of stored relevant contextualized information from tool sensors, alarms and metrology for signal feature extraction. This legacy problem has always made troubleshooting time consuming as engineers continuously drive for root cause analysis of process & yield excursions. This data quality problem also adds significant engineering cycle time to data science teams as they currently must spend more time understanding, iterating, and replicating datasets before meaningful insights can be gleaned from advanced algorithms.
Now, with the advent of tailored DT automation systems, companies like onsemi can have the ideal streamlined data repository that is designed with raw sensor data quality in mind for equipment Reliability Centered Maintenance (RCM) part lifetimes. This, as well as having built-in data science (DS) automation, with best practices and scalable deployment standards in mind. This new system-level design approach for Fab Engineers is based around the principles of the Ishikawa (or fishbone) diagram and is a big step forward away from legacy retrospective engineering approaches, where once the data was stored, engineers then scrambled to handle data quality gaps and DS skill gaps.
Until now, legacy data engineering created uncertainty with users of analytical and controls systems due to lower ML predictability scores and a distinct lack of ML value-add (and credibility) at the tool owner level. The legacy approach economizes storage and compute costs over ML readiness and equipment reliability lifetimes. Now, Fabs can store more of the relevant raw sensor and metadata with fully integrated DS and factory automation needed for smarter signal extraction and controls. The adoption of a purpose-built DT for materials, equipment, and processes in Fabs is now enabling a new era of smarter system engineering, allowing for faster automation of key signals to drive earlier detection of unwanted variability issues.
Furthermore, scalable Digital Twins with fully integrated ML applications deployed in production is now enabling companies like onsemi to achieve enhanced value in shorter timeframes. Ultimately, this talk will paint a compelling picture for other companies who are wondering what a digital twin is, why we need them, and what they can now deliver versus legacy factory systems. The Tignis and onsemi authors will also outline a digital twin roadmap for equipment that will further inform IT and Fab strategies toward real business value.
