A Small MEMS Neural Network to Classify Human Sitting and Standing Activities

The next frontier of MEMS applications is their use as computing units to harness data at the sensor level. In particular, MEMS intelligent computing unit has the unique promise of significantly increasing energy efficiency while simultaneously increasing data processing speeds and eliminating data latency in many applications such as wearable devices. Along this line, this paper presents a demonstration of the first simulated microelectromechanical (MEMS) network capable of classifying real-life experimental data based on acceleration measurement to distinguish between sitting and standing behaviors without the need for any computing or processing unit. The MEMS network is made of four MEMS; two MEMS in the input layer and two in the output layer. The first MEMS in the input layer is responsible for detecting a rising edge acceleration and allows the first MEMS in the output layer to be triggered at a following falling edge of the acceleration signal. This signature corresponds to the sitting activity. On the other hand, the second MEMS in the input layer is responsible for detecting the falling edge of the acceleration signal and allows the second output MEMS to declare a standing activity if it detects a following falling edge signal. This work demonstrates the potential of distinguishing between the sit and stand activities without any computing unit.