Harnessing Vibro-Impact Dynamics for High-Performance MEMS-Based Reservoir Computing
Part Of
Proceedings of the IEEE International Conference on Micro Electro Mechanical Systems (MEMS)
Start Page
632
End Page
635
ISSN
10846999
ISBN (of the container)
9798331572518
ISBN
9798331572518
Date Issued
2026
Author(s)
Abstract
This work demonstrates the first physical reservoir computer based on a MEMS vibro-impact resonator. This approach harnesses the rich, nonlinear dynamics generated by periodic mechanical contact, a regime fundamentally different from the weak, continuous nonlinearity of conventional Duffing-based physical reservoir computers. Vibro-impact systems access this strong nonlinear regime with minimal physical displacement, offering a significant advantage in device scaling and complexity generation. This work uses CMOS-MEMS resoswitch, a device proven for enabling ultra-low-power operation in wireless communication systems. Benchmarked against standard tasks, including NARMA, XOR, and MNIST image classification, the vibro-impact-based system achieves comparable or superior performance to state-of-the-art Duffing-based counterparts. These results are achieved with smaller training datasets, suggesting the generation of a more complex, high-dimensional state space that is highly effective for nonlinear and classification tasks.
Event(s)
39th IEEE International Conference on Micro Electro Mechanical Systems, MEMS 2026, 25 January 2026 - 29 January 2026, Salzburg
Subjects
CMOS-MEMS resoswitch
neuromorphic computing
nonlinear system
reservoir computing
Vibro-impact resonators
Publisher
Institute of Electrical and Electronics Engineers Inc.
Type
conference paper