Remaining Useful Life Prediction through Gradient Boosting Transfer Learning for Semiconductor Digital Twins

Digital twins for predictive maintenance can significantly optimize operational effectiveness and enhance productivity in semiconductor manufacturing environments. This work develops a Remaining Useful Life prediction model compatible with digital twins, utilizing a weak parameter-based transfer learning approach. By balancing the strengths of individualized and generalized models, the transfer learning approach provides accurate, scalable predictions for every tool, even those with limited data. This comparative analysis highlights the effectiveness of transfer learning for remaining useful life prediction and in addressing data challenges inherent in semiconductor manufacturing. The transfer learning approach enhances the digital twin's capability to provide realtime analytics to optimize maintenance scheduling and extending tool lifetimes.