Machine Learning Assisted Multiparameter Sensing via a Single Thermal Piezoresistive Membrane Resonator in Air

This work presents a novel machine learning assisted microelectromechanical systems multiparameter (temperature and pressure) sensor using a single thermal piezoresistive membrane resonator in air. It uses differential thermal actuation and piezoresistive sensing to achieve dual outputs, i.e., modulated output (resonant frequency modulated temperature/pressure signal) and non-modulated output (unmodulated temperature/pressure signal). Optimized location of the resistors boosts actuation efficiency and reduce crosstalk for sensing. Actuation via three resistors results in a Q factor of 457 (6th mode, 1388 kHz). Different machine learning algorithms were studied to decouple temperature and pressure signals from the dual output signals. Experiments show the proposed dual-output decoupling multi-task learning framework reduces root-mean-square errors by up to 65.7% for pressure sensing and 96.7% for temperature sensing compared with 3rd order polynomial fitting, multi-task learning, multi-task learning with efficient channel attention. The dual-output, multiphysics sensing principle can be applied to measure many other parameters under atmospheric pressure, using a single MEMS structure, e.g., mass, humidity, gas concentration, and acoustic pressure.