Smart Semiconductor Production Optimization Using AI-Assisted Hybrid Analytics and Edge Computing
Keywords:
AI-enhanced semiconductor manufacturing,Semiconductor yield optimization,Hybrid deep learning models,Edge data analytics,Smart wafer fabrication,Machine learning for yield prediction,Fault detection in semiconductor processes,Real-time process monitoring,Predictive analytics for fabs,Edge AI in manufacturing,Advanced process control (APC),Data-driven yield improvement,Defect classification using deep learning,Cyber-physical semiconductor systems,Intelligent manufacturing systems.Abstract
With the electric vehicle and consumer electronics industries projected to drive high demand for their products, semiconductor manufacturers are focusing on further improving their yield rates and reducing production costs. To achieve this goal, more effective yield prediction methods capable of deriving actionable recommendations to mitigate yield loss are required. While deep learning has shown promise for yield prediction, the training of such models usually requires a large number of label documents, which may not be readily available.
A hybrid modeling framework is proposed that combines cost aspects with a deep-learning-based yield-prediction model. The new approach is particularly well suited to settings in which only a limited number of label documents are available. A semiconductor manufacturer facing this scenario is considered. The results demonstrate that combining deep learning with cost information enhances the assistance provided to engineers, leading to information that is more useful for reducing production costs while maintaining an acceptable yield level. A further objective is to provide a deep-learning yield-prediction model with the lowest possible latency and bandwidth requirements while satisfying the accuracy needs of users.
