A Deep Topological and Self-Supervised Learning Approach
In semiconductor manufacturing, the ability to analyze vast amounts of high-dimensional data is critical for ensuring product quality and optimizing wafer yield. Traditional supervised machine learning approaches struggle with data sparsity, imbalance, and the need for extensive labeled datasets, particularly during new product introduction (NPI). To address these challenges, this webinar introduces an explainable AI framework that integrates deep topological data analysis (DTDA) and self-supervised learning (SSL). DTDA, an unsupervised machine learning method, transforms complex datasets into two-dimensional networks, where nodes represent clusters of similar samples. This enables the identification of key features impacting yield and quality, providing actionable insights for corrective measures. SSL complements this by leveraging large amounts of unlabeled data to detect patterns and anomalies without explicit supervision. The framework is further enhanced through transfer learning, enabling rapid adaptation to new datasets while reducing computational overhead. Validated using public open-source datasets, this approach demonstrates effectiveness in unsupervised image segmentation, defect detection, and classification. By automating clustering and grouping, it enhances process monitoring, defect identification, and decision-making, ultimately improving yield and reducing costs. Beyond semiconductor manufacturing, this framework holds potential for broader applications in industrial settings requiring analyzing and extracting insights from large-scale, high-dimensional datasets.
About the Speaker
Janhavi Giri, PhD
Intel Corporation
Dr. Giri is a data scientist with more than a decade of experience in the semiconductor industry, currently at Intel Corporation. She specializes in applying advanced machine learning techniques, including topological data analysis (TDA) and causal AI, to solve high-dimensional data challenges in semiconductor manufacturing. Her expertise focuses on yield optimization, equipment productivity, root cause analysis, and high-dimensional data mining. Dr. Giri actively contributes to the data science community through technical publications, conference talks, and thought leadership. She holds advanced degrees in Physics and Applied Mathematics and is passionate about bridging AI innovations with manufacturing to drive smarter, more efficient solutions. She was recently recognized by SPIE for her scientific contributions and received media coverage for her talk at SEMICON West 2024, solidifying her impact in AI-driven semiconductor manufacturing.
Future of Electronics Series
This webinar is the first in a year-long series on the intersection of smart manufacturing, sustainability and systems integration. Watch the INEMI calendar for additional webinars in this series.
Registration
This webinar is open to industry; advance registration is required. If you have any questions or need additional information, please contact Mark Schaffer ([email protected]).
Thursday, May 6, 2025
11:00 a.m. – 12:00 p.m. EDT (North America)
5:00-6:00 p.m. CEST (Europe)
Please note: You will need to log into your web account (free to members and non-members) to register. If you do not have a current web account, please create one and set up your profile.