Diffractive Neural Networks for High‐Throughput Classification of Objects in Microfluidic Systems

The integration of optical imaging and machine learning on microfluidic platforms makes it possible to achieve high‐content, minimally invasive characterization of a population of samples on a single chip. As the analysis of this high‐content information is typically conducted in the electronic domain, optoelectronic conversion speeds and the bandwidth available for data processing put a limitation on the throughput of these methods. In this work, we present an analysis system based on diffractive neural networks with the potential for integration in microfluidic systems for high‐content classification of objects with high sampling rates. We show that such a system can distinguish objects by size through passive optical inference with a numerical test accuracy of 98.2% and an experimental test accuracy of 83.4%, in an environment compatible with a microfluidic chamber. This development paves the way for novel approaches in high‐speed phenotyping of large cell populations based on all‐optical or hybrid optoelectronic neuromorphic information processing.