Biological systems seamlessly combine multiple functions in lightweight and energy-efficient structures. Such capability in synthetic structures would be desirable in numerous engineering applications such as aerospace, robotics and wearable devices. Here we report an integrated silicon-based structure configured to sense, perform different classification algorithms, and produce an action signal within the same physical layer. The algorithms are coded in the mechanical responses of the sensing elements of multiple coupled micro-electro-mechanical systems (MEMS), simultaneously capturing acceleration measurements to produce an actuated signal. This all-in-one structure operates with zero circuitry and low power consumption. As a demonstration, we designed and fabricated a network of three MEMS neurons to successfully perform both simple signal classification and activity recognition problems (standing and sitting) with only 9.92 × 10−17 kWh and 17.79 × 10−19 kWh energy consumption per operation, respectively. Our approach will enable emergent technologies, such as wearable devices, to perform complex computations with power from a single battery charge.
This paper demonstrates the use of micro-electromechanical systems to perform an efficient classification problem. Classification using digital computers is inefficient due to its power requirement and the separation between memory and processing units. As a proof of concept, basic signal classification has been performed using a network of only three mechanically coupled devices. The classification occurred at the hardware level rather than using a digital computer. The classification problem involves distinguishing between a ramp signal versus a step input signal. In this demonstration, a three-degree of freedom model was built for the proposed network in MATLAB. Detailed simulation results were extracted to validate the network operation and sensitivity to operating conditions. Next, the device network hardware was fabricated and tested to perform the required classification problem. Good agreement between simulation and experiment is demonstrated.
Microelectromechanical systems (MEMS) provide a platform where sensing, memory, and computation can co-exist within the same physical substrate. Here, we present a hysteresis-aware MEMS-based Continuous-Time Recurrent Neural Network (MEMS-CTRNN) that transforms nonlinear bistability from a device limitation into a computational resource. By embedding hysteresis into the training process, the MEMS-CTRNN achieves temporal memory and noise robustness with threefold fewer parameters than digital Long Short-Term Memory (LSTM) models, while maintaining comparable accuracy. Using insect-scale robotic flight data for validation, we demonstrate collision detection at 93% accuracy, establishing MEMS devices as viable analog recurrent processors. This framework suggests a pathway for MEMS-based neuromorphic architectures that unify sensing and computing for applications spanning microrobotics and in situ biomedical monitoring.