A Full-Stack Neuromorphic Prototype Architecture for Low-Power Wireless Sensors

One emerging trend in the evolution of sensor technology is to perform Artificial Intelligence (AI) based data processing already on the sensor. The benefits are in terms of reduced upstream communication load or reduced reaction times when the sensor is applied for example, in some time critical control loop. Such developments, however, may easily get in conflict with the equally important requirement of energy efficiency and low power operation. The brain inspired neuromorphic technology may resolve this conflict, if we can put all computation tasks of the sensor on neuromorphic principles, including the communication related processing as well. For this purpose, we need to rebuild the entire communication stack, including information encoding, protocols, radio algorithms on neuromorphic basis. In this paper we show a neuromorphic native design of an end-to-end sensor application and we exemplify such a communication stack with a prototype implementation, where the radio algorithm and protocol components have been replaced by neuromorphic algorithms implemented on a real neuromorphic chip combined with Software Defined Radio (SDR) hardware.