Solving the DLD boundary problem using iterative CFD

Deterministic Lateral Displacement (DLD) is a size-based sorting method relying on precise flow patterns. Those flow patterns are disrupted near boundaries, resulting in poor separation performance. In this work, we use a gradient descent method similar to that used in machine learning to evolve the physical geometry for a target flow pattern in 3D. The method is computationally expensive and difficult to implement for large arrays with low row shift fractions. To develop a design rule that can be used for any DLD device, we solve the boundary problem with row shift fractions (ϵ) of 1/6, 1/8, and 1/10, and various depths. All devices have small Re, circular posts, and a gap that is equal to the post diameter. We verify the design rule by using it to draw and simulate a larger array with a row shift fraction of (ϵ) of 1/20. The resulting array has a much more uniform critical particle size than all prior work.