Nowadays, IoT sensor devices use conventional processor chips, with low-power ARM architectures to enable long battery lifetime. However, traditional microcontrollers are not well-suited to all the new algorithms that sensors need to perform on the edge. The performance of deep neural networks (DNNs) is often limited by memory bandwidth, rather than processing power. In the upcoming years, it is expected that new application-specific integrated circiuts (ASICs) will reduce the power consumption required to run a DNN, enabling new edge architectures and embedded DNN functions in low-power IoT sensors. This will support new capabilities such as data analytics integrated with sensors, and image/speech recognition included in low cost battery-powered devices.
The sensor market driven by Internet-of-Things applications will evolve continuously through 2023. New sensors will enable more audio-visual events to be detected, while current sensors will fall in price to allow business cases to become more affordable, and new algorithms based on neural networks will emerge to deduce more information from current sensor technologies.