Artificial‐Intelligence‐Enhanced Mid‐infrared Lab‐on‐a‐Chip for Mixture Spectroscopy Analysis

Abstract

Bio/chemical mixture sensing in a water environment is of great importance in sensing applications. Relying on plentiful molecular fingerprints in mid‐infrared (MIR) and high integration potential, nanophotonic waveguide‐based MIR lab‐on‐a‐chip (LoC) provides a miniaturized and versatile solution for specific and label‐free bio/chemical detection. However, it is still challenging to implement an MIR LoC with on‐chip photodetection for chemical sensing in water, due to the strong MIR water absorption and limited MIR on‐chip photodetector scheme, let alone the spectral overlap issue in mixture analysis. Here, a MIR LoC integrating zero‐bias graphene photodetector is reported and the real‐time monitoring of three analytes in water leveraging the MIR LoC is demonstrated. Besides, using machine learning, the on‐chip collected spectra of the ternary mixture in water with 27 mixing ratios are successfully classified with an accuracy of 95.77%. Moreover, concentration prediction of individual analytes in a mixture is performed by developing a convolution regression network for mixture spectrum decomposition: 83.33% of the single‐component concentration predictions are within the 1 vol% error range, and an average root‐mean‐squared error of 1 vol% for mixture concentration predictions is achieved. The MIR LoC offers new opportunities for highly integrated intelligent sensing systems in various sensing scenarios in the Internet of Things era.