A Generative AI-Assisted Piezo-MEMS Ultrasound Device for Plant Dehydration Monitoring

Plant health, closely tied to hydration, has a direct impact on agricultural productivity, making the monitoring of leaf water content essential. Current devices, however, are often invasive, bulky, slow, power-inefficient, Complementary Metal-Oxide-Semiconductor (CMOS)-incompatible, and unsuitable for large-scale, re-usable outdoor sensor networks. Utilizing micro-electromechanical systems (MEMS) fabrication enables wafer-scale miniaturization and precise control of ultrasound transducers, thereby enhancing sensitivity while significantly reducing power and cost. This work introduces the CMOS-compatible, plant-leaf attachable piezo-MEMS ultrasound device (PMUT-Leaf-PMUT, PLP) for real-time dynamic moisture monitoring and rapid one-shot measurement of relative water content (RWC). Notably, the PLP is reattachable to pre-calibrated plant leaves, enhancing reusability and reducing electronic waste. Employing piezoelectric micromachined ultrasound transducers (PMUTs) fabricated via piezoelectric over silicon-on-nothing (PSON), the device non-invasively monitors hydration across diverse cultivars with a 70% relative water content (RWC) detection range. Generative deep learning using a conditional variational autoencoder (CVAE) translates electrical signals to precise hydration measurements, achieving an RWC root-mean-square error of 1.25%. The deployment of this generative AI-assisted PLP system directly links plant responses to environmental shifts, representing a significant advancement in precision plant health management and irrigation practices, thereby substantially improving agricultural efficiency and promoting environmental conservation.