Semiconductor Manufacturing Data Synthesis through GANs

This paper introduces a novel application of Generative Adversarial Networks (GANs) in the creation of a digital twin of a semiconductor manufacturing plant, as well as the investigation of new evaluation metrics in synthetic data quality. Digital twins are the next technological step for long-term semiconductor manufacturing capacity planning, performance optimization, and job scheduling. However, real-time predictive analytics for this industry demands substantial historical data about equipment, operations, and performance metrics like Overall Equipment Effectiveness (OEE). In cases where such data is unavailable, GANs can fill these data gaps by providing sufficient synthetic data to enable predictive modeling and digital twin creation. This investigation successfully generated synthetic semiconductor manufacturing data through Conditional, Wasserstein, and Copula GANs, enabling the creation of the digital twin for predictive modeling and further research of manufacturing improvements. A new method of evaluating synthetic data through a metric based on the distance between distributions scaled by similarity is also investigated.