Idle Vehicle Rebalancing in Semiconductor Fabrication Using Factorized Graph Neural Network Reinforcement Learning

With the advancement in the semiconductor industry, the size of fab becomes larger and thus more overhead hoist transportation (OHT) vehicles need to be operated, which necessitate efficient operation strategies for a large number of OHTs. In this study, we propose a cooperative rebalancing strategy of OHTs to increase the overall productivity of the material handling process in the fab. We discretize the fab into a number of zones and derives decentralized rebalancing strategies for each zone by applying a graph neural network (GNN) based multi-agent reinforcement learning (MARL). The proposed algorithm first represents the overall state of the fab into a directed graph and uses the graph representation to construct embedding values for each zone. The node embedding values are then used to determine the rebalancing action from each zone in a decentralized manner but to induce cooperation among zones. Simulation studies have shown that the proposed algorithm is effective in increasing various system-level key performance metrics compared to other heuristic and learning-based rebalancing strategies.