Technology

WHAT IS STREAM.AI™

  • STREAM.AI, StreamMosaic’s powerful and scalable engine at the core of all of our products, transforms data into accurate and actionable predictions.
  • STREAM.AI was built by semiconductor veterans to meet the complex and unique needs of semiconductor and electronics manufacturing.
  • STREAM.AI tackles semiconductor fabrication and post-fabrication by analyzing and identifying the most important features to predict failures at the wafer, die, and system level.
  • Leveraging data from multiple sources (in-line process tools, e-test, wafer sort, final test, and system level test), our machine learning engine automatically adjusts features and chooses the right model to ensure prediction accuracy and production

stream-ai

 

STREAM.AI™ – PAIN POINTS ADDRESSED

REDUCED COMPLEXITY

  • Edge Analytics – model deployment for data scoring at the edge
    • Model goes to where the data is residing for real-time scoring

MAINTAINABILITY & DEPLOYABILITY

  • Excursion Handling
  • Advanced monitoring capabilities
  • Responds to missing data issues
  • Automates model development and validation
    • Constantly adjusting to shifting data
    • Deploys the best scoring model on a user-defined basis
    • Drastically reduces engineering resources to develop and maintain models

HIGH AVAILABILITY & SCALABILITY

  • Redundancy and Load balancing between multiple nodes ensures nearly 100% availability
  • Scalable to multiple nodes as more computing resources needed

EASE OF USE & IMPROVED PERFORMANCE

  • Designed for Specific Use Case
    • GUI and User Experience focused on specific use case
    • Advanced outlier (anomaly) detection
    • Retrain existing models on new data only
    • State-of-the-Art ML models deployed on an ongoing basis

STREAM.AI™ KEY FEATURES

EASE OF USE

  • Automated selection of best algorithms, feature engineering, and hyper parameters

SCALABILITY

  • Horizontal scaling to ensure real-time predictions on large-scale production deployments

DESIGNED FOR SEMICONDUCTOR & ELECTRONIC INDUSTRIES

  • Proprietary approaches tailored specifically for the nuances of the semiconductor industry

PRODUCTION-WORTHINESS

  • Failure prevention, excursion handling, and automated model adjustment

REAL-TIME EDGE ANALYTICS

  • Client-Server Architecture allows models to be deployed to the edge

YIELD PREDICTION

  • Predict yield:
    • Using upstream test data
    • Far in advance for better supply forecasting and inventory reduction
  • Reduced engineer time in predicting yield
  • Identify yield excursions faster
    • Help identify root causes of yield excursions
  • Shorten the yield learning curve
  • Uncover design flaws that are not easily detectable
  • Improve overall yield by identifying correlations not found by other means

SMART TESTING

  • Predict if a chip will pass at a test station using upstream data
  • Focus test resources on product that are at the highest risk for failure
  • Reduce testing costs by diverting a significant portion of low-risk product volume away from expensive tests
  • Smarter product binning by quality

*Smart Testing can be applied to any test


VIRTUAL METROLOGY

  • Predict Critical Dimensions (CD) for specific process steps using sensor and inline data
  • Improve wafer-to-wafer and within-wafer variation
  • Immediately identify process excursions
  • Consumable life extension

FAILURE PREDICTION

  • Predict wafer failure at a process or test station using upstream sensor data obtained from manufacturing tools
  • Reduce scrap by identifying and correcting faulty behavior immediately
  • Increased yield as less product is exposed to faulty process behavior
  • Immediately remove faulty product from the process flow
  • Reduce operating costs by eliminating unnecessary processing of faulty products downstream
  • Reduction of tool downtime by quickly identifying the root cause of faulty process behavior