With the increasing intelligence of the shop floor, the manufacturing cell, as the basic organizational element of the shop floor, is becoming capable of autonomous sensing, learning, and decision-making. Leveraging the above capabilities of the manufacturing cell to respond in real-time to the states of the cell production local to the shop floor can improve shop floor flexibility, decision-making efficiency, and scheduling performance. Based on this, this paper proposes a smart shop floor distributed scheduling method based on the multi-agent deep reinforcement learning algorithm QMIX, which can take advantage of the intelligence of the manufacturing cells to obtain effective cell decisions and optimize the overall scheduling performance with the assistance of shop floor collaboration. This method is divided into two stages: firstly, a cell agent is built for each manufacturing cell, which is able to make autonomous decisions based on the cell state; then, a MIX network is built on the shop floor, which uses the shop floor state to modify the structure of each cell agent so that the decisions it makes can meet the cell and shop floor scheduling objectives. The proposed method was validated on the MiniFab semiconductor production shop floor, and the results showed that the method was effective in achieving collaboration between cell and shop floor objectives.