Researchers demonstrate reservoir computing with coupled microelectromechanical drum resonators
Updated
Updated · Nature.com · May 7
Researchers demonstrate reservoir computing with coupled microelectromechanical drum resonators
1 articles · Updated · Nature.com · May 7
The team used two capacitively coupled MHz drum resonators, achieving parity and NARMA benchmark performance with sideband-pumped phonon-cavity electromechanics and time-delay feedback.
The compact MEMS system integrates sensing and computing on one platform, using pump-amplitude modulation to create nonlinear dynamics and coherent energy transfer between distinct vibrational modes.
Researchers said the approach could extend single-resonator MEMS reservoir computing to multimode arrays and optomechanical systems without requiring closely matched resonance frequencies.
As sensing and computing merge, can tiny mechanical drums outperform silicon chips in future smart devices?
This new AI hardware is ultra-efficient but processes at just 25 Hz. Can it overcome this speed limit?
Sideband-Pumped Phonon-Cavity MEMS Reservoir Computing: Overcoming Frequency Constraints for Scalable Neuromorphic Hardware
Overview
In 2026, researchers developed a breakthrough MEMS reservoir computing platform using capacitively coupled aluminum and silicon nitride drum resonators. This system employed a pump-probe scheme and a time-delay feedback architecture to create 100 virtual nodes from a single device, overcoming traditional frequency-matching constraints through sideband pumping. This innovation enabled efficient energy transfer between mismatched resonators, reduced fabrication costs by 40%, and lowered power consumption to just 3.8 mW. Validated by benchmark tasks, the compact, scalable design supports real-time edge computing for applications like biomedical monitoring and industrial IoT. Despite challenges in fabrication and integration, pathways such as in-sensor computing and heterogeneous integration promise to unlock its full potential in neuromorphic hardware.