This integrated project aims to advance the application of cross-domain multi-modal artificial intelligence (AI) technologies in smart manufacturing. Focusing on key capabilities such as generative model, semantic understanding, digital twin simulation, anomaly prediction, and trustworthy decision-making, the project seeks to develop a highly integrated and modular AI-driven manufacturing system. The project consists of five tightly coupled sub-projects, covering defect image generation and diagnosis, event-based visual anomaly prediction, digital twin platform construction, IC manufacturing anomaly detection with trustworthy reasoning, and explainable AI–driven intelligent scheduling. By integrating state-of-the-art technologies—including vision-language models, multi-modal generative models, event cameras, 3D/4D Gaussian Splatting, diffusion models, graph neural networks, and reinforcement learning—the proposed system enables data sharing and modular collaboration, significantly enhancing system robustness and practical deploy-ability. From a societal and environmental perspective, the project is expected to substantially improve defect detection accuracy and early anomaly warning capabilities in manufacturing lines, thereby reducing defect rates, yield fluctuations, and material and energy waste. These advances directly support green manufacturing and sustainable development. Through close collaboration with semiconductor and advanced electronics manufacturers, the project will conduct multi-site real-world validation, strengthen industry–academia partnerships, and cultivate professional talents with expertise in AI research, manufacturing process optimization, and cross-modal system integration, thereby enhancing the competitiveness of the domestic workforce. From an economic perspective, the project will develop multiple critical AI modules to improve production efficiency, yield stability, and flexible scheduling capabilities in the semiconductor and electronics industries, while reducing reliance on manual operations. In collaboration with industrial partners, the project will accelerate the industrialization of AI technologies for process monitoring, risk assessment, and intelligent scheduling, promoting supply-chain-level innovation and strengthening international competitiveness. In addition, the project proposes a novel Sim-to-Real-to-Sim closed-loop framework centered on event-based vision and digital twin technologies to address key challenges of data scarcity, confidentiality, and high acquisition costs in manufacturing environments. By leveraging high-quality multi-modal data generation and advanced generative models, the framework enhances factory-level perception accuracy and anomaly prediction while reducing dependence on real-world data and model update costs. The integration of large language models for multi-task reasoning and explanation further improves human–AI trust and decision transparency. Overall, this project aims to drive a comprehensive upgrade of Taiwan’s smart manufacturing industry, enhancing technological autonomy and global competitiveness in high-tech manufacturing.