Communication Dans Un Congrès Année : 2026

Modeling of an all-mechanical in-sensor vibration classifier compatible with MEMS implementation

Résumé

This work explores the feasibility of an electromechanical in-sensor intelligent timeseries classification system implemented using MEMS technology. The proposed architecture integrates, on a single chip, an acceleration sensing layer and a mechanical MEMS neuron for vibration features classification, eliminating the need for any digital interfacing or processing. The computing principles relies on the direct electromechanical coupling between the sensing mass and the MEMS neurons, so that capacitive transduction enables computation within the sensor itself. A neuron performs a mechanical summation of input forces followed by an electrostatically governed nonlinear activation, analogous to neural processing. Simulations of the system's constituent elements, on a toy example of classification task, demonstrate the compatibility of MEMS-based neuromorphic architectures with MEMS sensing units, paving the way for truly integrated mechanical intelligence at the edge.

Fichier principal
Vignette du fichier
neuromems_iscas2026.pdf (2.53 Mo) Télécharger le fichier
Origine Fichiers éditeurs autorisés sur une archive ouverte
Licence

Dates et versions

hal-05562234 , version 1 (22-03-2026)

Licence

Identifiants

  • HAL Id : hal-05562234 , version 1

Citer

Aleksandra Marković, Armine Karami, Hervé Fanet, Dimitri Galayko, Bernard Legrand, et al.. Modeling of an all-mechanical in-sensor vibration classifier compatible with MEMS implementation. ISCAS 2026 - 2026 IEEE International Symposium on Circuits and Systems, IEEE, May 2026, Shanghai, China. ⟨hal-05562234⟩
529 Consultations
71 Téléchargements

Partager

  • More