Signal Classification Using a Mechanically Coupled MEMS Neural Network

This paper reports on the implementation and operation of the first micro-electromechanical (MEMS) network to perform basic neural computing. The device is comprised of a mechanically coupled network of three electrostatically controlled micro-structures with two of the coupled structures acting as the input layer and the third as the output (computing) layer. It has been shown that such device can be programed by application of appropriate bias voltages to the electrostatic control electrodes so that it can distinguish between a ramp (gradually increasing) input signal and a step (abruptly rising) input signal. The results serve as the proof of concept and a pre-cursor to performing more complex computational tasks using coupled micro-structures acting as a network of interacting neurons.