High-precision whispering gallery microsensors with ergodic spectra empowered by machine learning

Whispering gallery mode (WGM) microcavities provide increasing opportunities for precision measurement due to their ultrahigh sensitivity, compact size, and fast response. However, the conventional WGM sensors rely on monitoring the changes of a single mode, and the abundant sensing information in WGM transmission spectra has not been fully utilized. Here, empowered by machine learning (ML), we propose and demonstrate an ergodic spectra sensing method in an optofluidic microcavity for high-precision pressure measurement. The developed ML method realizes the analysis of the full features of optical spectra. The prediction accuracy of 99.97% is obtained with the average error as low as 0.32 kPa in the pressure range of 100 kPa via the training and testing stages. We further achieve the real-time readout of arbitrary unknown pressure within the range of measurement, and a prediction accuracy of 99.51% is obtained. Moreover, we demonstrate that the ergodic spectra sensing accuracy is ∼11.5% higher than that of simply extracting resonating modes’ wavelength. With the high sensitivity and prediction accuracy, this work opens up a new avenue for integrated intelligent optical sensing.