Droplet Sorting Computer

We develop a simple combined design framework to design a microfluidic droplet sorting computer. The framework uses suitable design parameters and Multi-Agent Reinforcement Learning (MARL) to optimize the microfluidic device. The sorting computer, which constitutes an optimized network topology and minimized inter-droplet spacing, routes all the drops toward one of the outlets through simple hydrodynamic interactions. The design has sufficient robustness and stability for various operating parameters, such as inlet spacing and droplet properties. Its optimized hydrodynamic interactions between the drops facilitate the device's scalability to a large number of drops. We envision that the framework and the device design would yield faster and more efficient design of lab-on-chip devices and facilitate microfluidic adoption in commercial processes.