Enhancing supply chain resilience and efficiency of HVAC systems in semiconductor manufacturing facilities using graph-based large multimodal models

Semiconductor manufacturing facilities (SMFs) demand ultra-precise environmental conditions maintained by specialized HVAC systems, critical for a resilient and sustainable semiconductor supply chain. While AI-driven solutions have been applied to generic supply chain optimization, they often fail in addressing the unique challenges of SMFs, where HVAC systems must maintain sub-0.1 °C temperature stability, account for 40–60 % of facility energy consumption, and comply with stringent cleanroom standards. This paper proposes an innovative framework that integrates graph-based large multimodal models (G-LMMs), enhanced by graph neural networks (GNNs), to optimize SMF HVAC supply chains across the Design, Construction, Installation, Maintenance, and Operation (DCIMO) phases. GNNs enable the capture and analysis of complex relationships within HVAC systems, facilitating real-time anomaly detection and optimized material flows. Unlike conventional AI models, G-LMMs combine GNNs with multimodal data processing to achieve three key advancements: (1) real-time anomaly detection, (2) automated compliance monitoring, and (3) circular economy integration through resource reuse. G-LMMs enhance supply chain visibility by harmonizing diverse data types while meeting SMFs' precision requirements. As the first framework to unify GNNs and multimodal AI for HVAC optimization, this approach represents a paradigm shift in sustainable semiconductor manufacturing, with broader implications for industries reliant on precision-controlled environments.