A deep learning approach for designed diffraction-based acoustic patterning in microchannels

Abstract

Acoustic waves can be used to accurately position cells and particles and are appropriate for this activity owing to their biocompatibility and ability to generate microscale force gradients. Such fields, however, typically take the form of only periodic one or two-dimensional grids, limiting the scope of patterning activities that can be performed. Recent work has demonstrated that the interaction between microfluidic channel walls and travelling surface acoustic waves can generate spatially variable acoustic fields, opening the possibility that the channel geometry can be used to control the pressure field that develops. In this work we utilize this approach to create novel acoustic fields. Designing the channel that results in a desired acoustic field, however, is a non-trivial task. To rapidly generate designed acoustic fields from microchannel elements we utilize a deep learning approach based on a deep neural network (DNN) that is trained on images of pre-solved acoustic fields. We use then this trained DNN to create novel microchannel architectures for designed microparticle patterning.