A Mems Accelerometer with in-Sensor Neuromorphic Computing Capability

This paper introduces a neuromorphic accelerometer prototype designed specifically for “In-Sensor Computing,” leveraging reservoir computing for local analog signal processing. The prototype excels in classifying analog signals, achieving an impressive 97.3% accuracy in distinguishing six motion postures detected by the accelerometer. Moreover, this accelerometer has the ability to operate independently, capable of completing real-time classification of motion signals locally without relying on external physically separated sensors or digital processing units. The remarkable performance demonstrate its great potential in the energy efficiency of information processing, which facilitates the new future possibilities for exploring the emerging applications using smart sensors with edge computing in the future AI era.