A Three Degree of Freedom Model Approach to Enable a MEMS-Based Neural Computing Unit

With enormous amounts of data being generated every day from countless sensors and sensor networks, the need for intelligent devices to process and make use of this data continues to grow and is only projected to increase. The advent of wearable technologies has exacerbated this problem, and with researchers struggling to process data locally with small power budgets, it is clear a solution is needed. Micro-electromechanical (MEMS)-based innovation will have high impact on these issues. MEMS devices can process computing taskes in the hardware level which consumes almost no power (nW). They are very small in size and do the classification without the need of storing the data which boosts up the power saving. Toward this goal, simulation results for a MEMS network to perform basic neural computing is shown in this paper. The network is made up of a mechanically connected network of three electrostatically controlled microstructures, two of which serve as input layers and the third as output (computing) layers. The mechanical coupling was achieved through stiffnesses connecting the masses of the MEMS. It has been demonstrated that such a device may be programmed to distinguish between a ramp (gradually growing) input signal and a step (abruptly rising) by applying suitable bias voltages to the electrostatic control electrodes. The findings serve as a proof of concept and founding to completing more sophisticated computational tasks using MEMS and opening a new direction for alternative efficient computing technologies compared to current digital computing.