Reinforcement Learning Double DQN for Chip-Level Synthesis of Paper-Based Digital Microfluidic Biochips

Digital microfluidic biochips (DMFBs) can effectively reduce the cost of biochemical analysis and improve experimental efficiency, as they are easy to carry, use fewer reagent samples and have high precision. Paper-based DMFBs (PB-DMFBs) are a branch of microfluidic biochips. This technology prints ink containing carbon nanotubes on special paper to form electrodes and control wire, so the manufacturing cost and time required are far less than the traditional digital microfluidic chip, in which droplets move between two control layers. However, the chip-level synthesis of PB-DMFBs becomes more challenging because all circuits of PB-DMFBs are printed on a single paper layer. Furthermore, current PB-DMFB designs must address various issues, including fabrication cost, reliability, and safety. Therefore, a more flexible method for the chip-level synthesis of PB-DMFBs is needed. In this article, we propose a chip-level synthesis method of PB-DMFBs based on reinforcement learning. Double deep
$Q$
-learning networks (Double DQN) are suitable for agents to select actions and estimate actions, and then obtain optimized comprehensive results. Experimental results demonstrate that the proposed method is not only effective and efficient for chip-level synthesis but also scalable to applications with high reliability and safety requirements.