Newswise — Microelectromechanical systems (MEMS) electrothermal actuators are widely used in applications ranging from micro-optics and microfluidics to nanomaterials testing, thanks to their compact size and powerful actuation capabilities. However, their motion typically follows a nonlinear voltage-displacement relationship, mainly due to the combination of Joule heating and thermomechanical effects. To compensate, engineers often deploy sensors and closed-loop control systems, which increase manufacturing complexity, power consumption, and failure risk, especially at small scale. In many applications, fragile sensing elements also have difficulty withstanding harsh processing conditions. Based on these challenges, there is a strong need to develop new approaches that can achieve linear actuation without relying on sensors or complex control electronics.
Researchers from McGill University and the University of Toronto report a new design strategy for MEMS electrothermal actuators in a study published in October 2014 (DOI: 10.1038/s41378-025-01065-4). Microsystems and nanoengineering The research team demonstrated that actuator nonlinearities can be mechanically compensated for by integrating a machine learning-optimized metastructure into the actuator itself. These structures reshape the mechanical response so that displacement increases nearly linearly with applied voltage without the use of sensors, feedback loops, or additional electronics, providing a simpler and more robust route to precise microscale motion.
The central idea behind this work is to use complementary mechanical responses to counteract the actuator’s inherent secondary voltage displacement behavior. Electrothermal actuators naturally produce displacements that are proportional to the square of the applied voltage. The researchers designed a mechanical metastructure whose deformation follows the square root relationship of the input displacement. When combined, the two nonlinear behaviors effectively cancel each other out, resulting in an overall near-linear response.
To realize this concept, the team built a metastructure using a combination of straight and tilted microbeams that exhibit stiffness that hardens and softens upon deformation. The design space spans thousands of geometric configurations, making traditional trial-and-error optimization impractical. Instead, the researchers generated tens of thousands of finite element simulations to train a neural network model that can predict mechanical behavior almost instantly.
These machine learning models made inverse design possible. Starting from the desired linear response, the algorithm identified the optimal geometric parameters for the metastructure. The optimized design was fabricated using standard MEMS processes and experimentally tested within a scanning electron microscope. Measurements confirm that the integrated metastructure reduces actuator nonlinearity by approximately 85% compared to conventional designs, in close agreement with simulation predictions.
“Nonlinearity has always been treated as something to be corrected electronically,” said one of the study’s senior authors. “What we’re showing here is that the mechanical structure itself can be designed to do that job.” By moving linearization from software and sensors to geometry, this approach simplifies system architecture while maintaining accuracy. “This opens the door to more robust MEMS devices, especially in environments where sensors are fragile, power is limited, or space is limited,” the researchers said.
Sensor-free linearization has the potential to greatly expand the utility of MEMS electrothermal actuators in precision engineering. Applications such as tensile testing of 2D materials, biomedical microdevices, and remote or implantable systems can benefit from simpler and more reliable actuation. Beyond electrothermal actuators, the design framework can be adapted to other nonlinear drive technologies at both micro and macro scales. More broadly, this research focuses on how machine learning can transform machine design to enable intelligent structures that achieve complex functionality without additional electronics. As data-driven optimization becomes more integrated into engineering workflows, such mechanically “smart” devices may become the basis for the next generation of microsystems.
###
References
Toi
10.1038/s41378-025-01065-4
Original source URL
https://doi.org/10.1038/s41378-025-01065-4
Funding information
CC, LZ, and YS gratefully acknowledge financial support from the Natural Sciences and Engineering Research Council of Canada (NSERC) Ideas to Innovation Program, the McGill Innovation Fund, the McGill TechAccelR Program, the Canadian Foundation for Innovation (CFI) JELF Program, and the NSERC Discovery Program. AHA acknowledges financial support from Programmable Multifunctional Metamaterials of Canada and the Natural Sciences and Engineering Research Council’s Canada Research Council Program through the NSERC Discovery Grant (RGPIN-2022-04493). HM is supported by a Doctoral Award (B2X) from the Quebec Research Fund – Nature and Technology (FRQNT).
About Microsystems and nanoengineering
Microsystems and nanoengineering is an online-only, open-access, international journal dedicated to the publication of original research results and reviews on all aspects of micro- and nano-electromechanical systems, from basic to applied research. This journal is published by Springer Nature in partnership with the Institute of Aerospace Information, Chinese Academy of Sciences, and with support from the National Key Laboratory of Transducer Technology.
