Power aware Neural Networks for self-powered IoT devices with edge computing capabilities

Offentlig prosjektsammendrag

ONiO is a highly innovative SME with a vision to bring self-powered solutions to the IoT market, across a wide range of IoT market segments: healthcare, industrial, automotive, buildings, etc. Our core technology platform is based on a highly flexible, ultra low power, compact chip designed for energy harvesting - i.e. capturing the available energy from the surroundings, primarily radio frequency (RF) energy but also capable of using sources like solar/voltaic cells, ambient, temperature or kinetic energy. The chip is capable of powering multiple embedded sensors in a system, replacing the need for a battery, and is currently equipped with Bluetooth (crystal-less BLE transceiver) for wireless communication. With this project, we wish to push computational power and energy efficiency beyond the state-of-the-art by implementing neural network-based hardware accelerators which can perform logic operations based on peripheral/ sensor data without the intervention of the CPU, and which only draw power whenever a workload is present. We call this a Power Aware Neural Network (PONN). The central goal is to save power by waking up the CPU only when needed, and to make better decisions at the edge nodes in a network - resulting in less radio traffic and load on the back office/cloud solution. We plan to start marketing and commercializing the new ONiO.zero version (edge computing-enabled, using PONN) for smart buildings, smart cities and assisted living segments - where we expect the value proposition and competitive advantages to be the highest.