Sr. Data Scientist

Micron Technology, 2016/12 - Current 

As a technique lead, leading projects and different teams to develop AI/ML solutions to accelerate yield ramping and virtual metrology, design system architecture for analysis/inference at scale.

Projects 

• Chip image pattern recognition.
• Apply various Deep/Machine Learning technique to cluster sub-die issue pattern for users.
• Implement supervised VAE and Xception model for classification with ~90% accuracy.
• Interactive with yield auto diagnostic system, provide root cause in minutes.
• Yield Auto Diagnostic 
• Automatically analysis/search the yield opportunity and/or the root cause of issues. 
• Hundreds of yield toggle have been found via this system, saving millions of dollars. 
• Implement and fusion various machine learning and statistical theory for modeling. 
• A particular architecture across scala/spark/python/sparkR/php/angular/spray with celery distributed task queueing system and spark daemon for serving 7 fabs globally. 
• Internal Data-as-a-Service (DaaS)
• A easy/reuse-able single source API via scala/spark, able to reach more than 20 internal data source. 
• High performance and saving human hour of developing data process > 500 hours. 
• Image Virtual Metrology
• Fusion the CMOS image and FDC to build up virtual metrology on CMP/DE. 
• Estimated benefit on improving CPK and increase inline inspection to 100%. 
• Implement via Machine Learning/Deep Learning with spark and MLflow for serving model evaluation and tracking. 
• Best Window of Inline
• Define the critical process and also suggest the best target/window.
• Shorten the New Product Introduction(NPI) cycle, saving $1.3M.
  
• Implement by Pyspark with own-design algorithm based on biserial rank and moving LOWESS.

Data Science Engineer

Projects

• Inline auto-analysis system
• Automatically looking for the tuning knob and source of variation for SPC control charts.
• Improve 10% CPK for critical SPC charts.
• Implement by hierarchical structure random effect model with R and Teradata.
• Predictive Maintenance of furnace tool
• Predict the remain lifetime of furnace parts to prevent wafer scrapped.
• Saving hundreds of wafer before scrapped.
• Implement by gaussian process and random process to predict remain lifetime with python and tableau.
• Wafer Accept Test(WAT) diagnostic system
• Find out the root cause of WAT issue from incoming SPC charts automatically.
• Saving ~20 labor hours per week for PI team.
• Implement by Pyspark/hive. 

• Host work shop and roll out the SoV system to whole fab, which improving 10% CPK on all critical charts.
• Main developer in hbase workshop to solve major hadoop issue and redesign hbase tables for improving performance ~10x.

Big Data Analysis

• Analysis the data from Facebook ads. 
• Tableau Dashboard for FB ads about retain/revenue rate. 
• Building up Hadoop and spark system via EC2 on AWS Build up recommend algorithm. 
• ETL programing with spark. 
• Graph algorithm for integrating the membership data. 
• DataBase design for PostgreSQL.

Big Data Analysis

• Develop FDC (Faults Detect and Classification) System and Low Yield Analysis. 
• Using parameter of Machine to develop Virtual Metrology system via machine learning for measurement cost reduction 
• Statistical method and algorithm development. 
• Proposed a PCA-based abnormal detection algorithm