Design automation for deterministic lateral displacement by leveraging deep Q-network.

Despite the widespread application of microfluidic chips in research fields, such as cell biology, molecular biology, chemistry, and life sciences, the process of designing new chips for specific applications remains complex and time-consuming, often relying on experts. To accelerate the development of high-performance and high-throughput microfluidic chips, this paper proposes an automated Deterministic Lateral Displacement (DLD) chip design algorithm based on reinforcement learning. The design algorithm proposed in this paper treats the throughput and sorting efficiency of DLD chips as key optimization objectives, achieving multi-objective optimization. The algorithm integrates existing research results from our team, enabling rapid evaluation and scoring of DLD chip design parameters. Using this comprehensive performance evaluation system and deep Q-network technology, our algorithm can balance optimal separation efficiency and high throughput in the automated design process of DLD chips. Additionally, the quick execution capability of this algorithm effectively guides engineers in developing high-performance and high-throughput chips during the design phase.