AI-Empowered Digital Twin for Multi-FAB Simulation

Semiconductor manufacturing systems are extremely large-scale and complex, which makes their analysis inherently challenging. In advanced semiconductor production, Send-FABrefers to an operational strategy in which wafer lots are processed across multiple fabrication facilities (FABs) during their manufacturing flow. By enabling Send-FAB operations, manufacturing capacity across heterogeneous FABs can be more effectively utilized. To optimize Send-FAB operations, multi-FAB simulation is an essential enabling technology. However, conventional multi-FAB simulation models are often excessively large, requiring substantial computational resources and resulting in prohibitively slow simulation performance. To address these limitations, this study proposes an AI-empowered Digital Twin framework for multi-FAB semiconductor manufacturing systems. The proposed approach accelerates simulation by partially transforming the conventional mechanism-driven models into data-driven models, while preserving essential system dynamics. By selectively embedding AI-based surrogate models within the multi-FAB simulation, the computational burden is significantly reduced without sacrificing modeling fidelity. This paper presents the concept of the AI-empowered Digital Twin and describes a systematic methodology for its construction and application to Send-FAB optimization.