Acoustic field design in microfluidic geometries via Huygens-Fresnel diffraction and deep neural networks

Abstract

Creating defined nonuniform acoustic fields, with particle capture locations beyond the grids and lines that are typically generated, has a range of applications in microfluidic systems. In this work, we use a unique interaction between polydimethylsiloxane (PDMS) channel geometries and a travelling substrate wave to explore the creation of nonuniform acoustic fields for microscale pattering. Surface acoustic waves (SAWs) are ideal for generating high-frequency, microscale wavelengths, though conventionally generate particle pattering via the interference between opposing SAW transducers, creating a standing SAW. Our recent work, however, has shown that patterning can alternatively be created using only a single travelling SAW and channel boundaries via diffractive effects, where fringe spacing is a function of the channel wall orientation. Unique to SAW, this creates a condition where the channel shape directly impacts the acoustic field that results, and therefore the possibility of designing acoustic fields by defining the channel geometry. To solve this inverse problem, we implement a machine-learning approach based on a Deep Neural Networks (DNN) that can define channel geometries resulting in desired acoustic fields. This submitted work will introduce the mechanisms via which this diffractive patterning occurs and the implementation of the DNN design process.