Controlled Manipulation and Active Sorting of Particles Inside Microfluidic Chips Using Bulk Acoustic Waves and Machine Learning.

Manipulation of cells, droplets, and particles via ultrasound within microfluidic chips is a rapidly growing field, with applications in cell and particle sorting, blood fractionation, droplet transport, and enrichment of rare or cancerous cells, among others. However, current methods with a single ultrasonic transducer offer limited control of the position of single particles. In this paper, we demonstrate closed-loop two-dimensional manipulation of particles inside closed-channel microfluidic chips, by controlling the frequency of a single ultrasound transducer, based on machine-vision-measured positions of the particles. For the control task, we propose using algorithms derived from the family of multi-armed bandit algorithms. We show that these algorithms can achieve controlled manipulation with no prior information on the acoustic field shapes. The method learns as it goes: there is no need to restart the experiment at any point. Starting with no knowledge of the field shapes, the algorithms can (eventually) move a particle from one position inside the chamber to another. This makes the method very robust to changes in chip and particle properties. We demonstrate that the method can be used to manipulate a single particle, three particles simultaneously, and also a single particle in the presence of a bubble in the chip. Finally, we demonstrate the practical applications of this method in active sorting of particles, by guiding each particle to exit the chip through one of three different outlets at will. Because the method requires no model or calibration, the work paves the way toward the acoustic manipulation of microparticles inside unstructured environments.