Speaker
Description
Reservoir computing is a bio-inspired computational paradigm that exploits the intrinsic dynamics of complex systems for temporal information processing. For any physical element to perform reservoir computing, it requires to have at least features of (1) nonlinearity and (2) fading memory. Microelectromechanical resonators, owing to their inherent nonlinearity and transient states, constitute an attractive platform for reservoir computing while offering the ability to integrate of sensing and computing within a single hardware device [1-2].
Here, we present our recent work on the experimental demonstration of reservoir computing in a double-drum electromechanical system [3–5]. Within the framework of phonon–cavity electromechanics, a virtual neural network is mapped onto two capacitively coupled drum resonators by modulating the pump force to induce the required nonlinearity via sideband pumping of the phonon cavity. This pump tone generates nonlinear dynamics in the energy transfer between the two resonators, enabling reservoir computing functionality. To enhance the memory capacity of the system, we exploit not only the intrinsic transient dynamics of the mechanical resonators but also implement a time-delayed feedback loop, which establishes coupling between the current and past states for temporal information processing.
The performance of the resulting neural network is evaluated using both parity-benchmarks and normalized auto-regressive moving average (NARMA) benchmarks through training and prediction tasks. Furthermore, to demonstrate the integration of sensing and computing within this double-drum system, one of the coupled resonators is used as the input sensing channel, while the other serves as the computational node for training and readout. This new, neuron-inspired computing scheme can easily be extended to other multimode coupling platforms, such as mechanically coupled resonator arrays and optomechanical systems.
Reference:
[1] G. Dion, et al., J. App. Phys. 124,152132 (2018).
[2] X. Guo, et al. Microsyst Nanoeng 10, 84 (2024).
[3] X. Zhou, et al. Nano Letters, 21(13), 5738-5744 (2021)
[4] A. Pokharel, et al. Nano Letters 22(18), 7351–7357 (2022)
[5] T. Farah, et al, arXiv:2601.02617v3, accepted by Microsystems & Nanoengineering (2026)