Helena Hedson

Categories

Tags

Archives

Blogs » Technology » Worth of Domain Expertise in Managing Semiconductor Fab Data

Worth of Domain Expertise in Managing Semiconductor Fab Data

  • Posted by Helena Hedson May 10, 2023 Filed in Technology #Data Analysis  #3d emblems manufacturing technology in India  #semiconductor  #manufacturing process  #failure analysis  #correlations  #semiconductor fabrication  #yield  412 views
    click to rate

    In the world of semiconductor fabrication, there is an abundance of data generated daily. This data comes from a variety of sources such as manufacturing, design, test, assembly, and in-field, and can include information on yield, performance, quality and reliability. However, without the proper expertise and understanding of the domain, the data collected can be meaningless and lead to wasted time and resources. In this article, we will discuss the importance of domain expertise in managing and effectively utilizing the tsunami of data collected in semiconductor fabs.

     

    The Risks of Naive Data Analysis

    Naive data analysis is the practice of applying simplistic techniques to data sets without understanding the underlying domain. This approach can lead to uninteresting or even spurious correlations that do not contribute to the end goal. In the case of semiconductor fabrication, naive data analysis can result in wasted resources and a failure to identify potential yield excursions or quality mishaps.

     

    The Importance of Domain Experts

    To avoid the pitfalls of naive data analysis, a team of experts is required to encode and store the data, build the underlying data structures, and create models that predict yield excursions and identify potential quality mishaps. A single test equipment engineer cannot possess all the necessary domain knowledge, but a team comprising end-users, data scientists, and IT experts can set up the data analytics platform for success.

     

    Data Ingestion and Database Modeling

    Multiple people from each domain, such as manufacturing, design, testing, assembly, and in-field, are needed to perform data ingestion and database modelling efficiently and effectively. Each domain expert brings unique knowledge and perspective to the table, which is crucial in developing an accurate and efficient data model. The end-users knowledge of the manufacturing process and the product design is critical in determining the data sets to be collected, while data scientists can help to identify patterns and correlations that may not be apparent to end users.

    IT experts are responsible for building the infrastructure to store and process the data effectively. This infrastructure must be scalable and able to handle the growing volume of semiconductor data that is generated daily. They are also responsible for creating the interfaces and tools necessary for end-users to access and manipulate the data. By bringing together the various domains and their respective experts, the data analytics platform can be optimized for success.

     

    Building Accurate Models 

    One of the primary benefits of domain expertise is the ability to build accurate models that predict yield excursions and identify potential quality mishaps. By leveraging their knowledge of the manufacturing process and product design, end-users can help data scientists identify the critical parameters that affect yield and quality. This information can be used to build accurate models that can predict the yield and quality of a product under different conditions.

     

    Conclusion 

    In conclusion, the importance of domain expertise in managing and utilizing data in semiconductor fabs cannot be overstated. Naive data analysis can lead to uninteresting or even spurious correlations, wasting time and resources. A team of experts comprising end-users, data scientists, and IT experts is required to set up the data analytics platform for success. By bringing together the various domains and their respective experts, accurate models can be built that predict yield excursions and identify potential quality mishaps.

     

    References:

    C. Fornaciari, M. Pelcat, L. Benini, E. H. Hoummady, and F. Vincenzi, "Domain-specific modelling for designing MPSoCs: a review," ACM Trans. Embed. Comput. Syst., vol. 16, no. 4, p. 93, 2017.

    N. Asgarzadeh, A. H. Soheily-Khah, and M. B. Menhaj, “A Data Analytics Approach for Yield Improvement in Semiconductor Manufacturing,” IEEE Transactions on Semiconductor Manufacturing, vol. 30, no. 3, pp. 276-283, Aug. 2017.

    S. S. Kodakkal and R. K. Mittal, “Machine Learning Approach for Yield Improvement in Semiconductor Manufacturing,” IEEE Transactions on Semiconductor Manufacturing, vol. 33, no. 2, pp. 147-156, May 2020.

    M. A. Lesczynski, M. J. Stuecheli, and J. A. Torres, “Yield Learning and Improvement at an Integrated Device Manufacturer,” Journal of Manufacturing Science and Engineering, vol. 136, no. 2, p. 021018, Apr. 2014.