Neural network control of a MEMS torsion micro mirror

High-performance optical communication networks require stable switches to rapidly redirect a light beam from one port, or fiber, to another without converting the optical signal into an electrical signal. Micro-electromechanical (MEMS) optical switches are miniature devices that use tiny mirrors to alter the path of light. These switches enable all-optical capability, are small and compact, and cheap to fabricate. However, many torsion mirror optical switches are activated by electrostatic actuators that exhibit a pull-in phenomena which greatly effects system stability. The pull-in phenomena occurs when the gap between the electrodes in the actuator is reduced to less than two thirds of the original value thereby causing an uncontrolled contraction between the two sides the capacitor. This paper describes how a backpropagation neural network can be used to control an electrostatically actuated optical switch without using stiff suspension systems or mechanical stops. The two-layer network is applied to both single input and dual input MEMS torsion mirror optical switches. Simulation studies are presented to demonstrate how the proposed scheme allows the switching mechanism to operate in a stable range and avoid the effect of pull-in phenomena