Reproduce “A Benchmark Time Series Dataset for Semiconductor Fabrication Manufacturing Constructed using Component-based Discrete-Event Simulation Models” on OpenResearch

About this paper

Vamsi Krishna Pendyala, Hessam S. Sarjoughian, Bala Potineni, Edward J. Yellig

Researchers from Arizona State University and Intel Corporation developed a publicly available benchmark time series dataset for semiconductor fabrication manufacturing. This dataset, constructed from high-fidelity Discrete-Event Simulation models of an Intel MiniFab, facilitates the development and evaluation of machine learning models for factory optimization and enables broader research into smart manufacturing.

Key insights

• PDEVS-based simulations can effectively generate high-fidelity, dynamic data reflecting complex semiconductor fabrication processes, including realistic repair states and varying input loads.
• Principal Component Analysis (PCA) of extracted features revealed distinct data patterns and dominant characteristics for different factory configurations, highlighting the impact of repair states and wafer generation on process variability.
• Integrating static covariates (e.g., repair state, lot size) within machine learning models, such as Temporal Fusion Transformers, is crucial for effectively distinguishing and predicting behavior across varied factory configurations.

Summary by alphaXiv.