This paper addresses collision-free path planning for Multi-AGV clusters in semiconductor intelligent warehouses, where nanometer-level vibration control and ultra-high-frequency material handling require ultra-smooth trajectories and flawless coordination. Traditional Conflict-Based Search (CBS) algorithms struggle with computational complexity and excessive path conflicts in dense AGV environments, risking throughput and fragile cargo integrity. To overcome these issues, we propose an enhanced CBS framework with three innovations: 1) Allowing unloaded AGVs to traverse beneath storage racks, optimizing traffic flow; 2) Introducing a turns&waits heuristic in A* to reduce sharp turns and unnecessary stops for smoother paths and 3) Incorporating a spatiotemporal heatmap to predict congestion and avoid conflicts. Experiments in a \(61 \times 29\) grid show a 94.2% faster computation time and 71.5% fewer path search collisions during planning iterations, with 71.7% fewer aggressive maneuvers. The proposed method ensures vibration-minimized, stable transport crucial for semiconductor logistics while improving computational efficiency.