Reinforcement Learning based Droplet Routing Algorithm for Digital Microfluidic Biochips

Digital Microfluidic Biochips (DMFBs) are part of lab-on-a-chip (LOC) devices and comes under the category of micro-electro-mechanical systems (MEMS). DMFBs are designed to be an alternative for traditional biochemical laboratories. DMFBs achieve miniaturization, automation, and programmability. DMFBs use electro-wetting-on-dielectric (EWOD) property to manipulate droplets on-chip discretely. Several computer-aided design (CAD) techniques have been designed for synthesizing DMFBs to reduce design complexity. Finding the concurrent routes between all source-target pairs of a bioassay is a challenging problem and NP-Complete. We proposed a reinforcement learning based droplet routing algorithm for DMFBs. Q-learning technique is used to determine a certain predefined number of optimal paths between a source-target pair. Q-learning is an off-policy reinforcement learning algorithm. After the paths for all the source-target pairs are determined, routes will be checked for constraint violations and collisions. If any collisions or violations are found, route compaction is done using stalling and detouring. Experimental results show that our proposed droplet routing algorithm outperformed compared algorithms.