< retour aux publications

Event-based processing: a new paradigm for low-power

Auteur(s) : L. Fesquet, S. Germain, J. Simatic, A. Cherkaoui, T. Le Pelleter, S. Engels

Doc. Source: 19th IEEE Mediterranean Electrotechnical Conference (MELECON'2018)

Publisher : ACM IEEE

Today, our digital society exchanges large data flows. The amount of data is incredibly huge and the future promises new data communications between humans and technological equipment such as robots, cars, planes, etc. This is known as the Internet of Things (IoT) and is probably part of the humanity progress. Nevertheless, this data orgy wastes a lot of energy. Indeed, these “communicating things” already consume about 10% of the electrical power produced in the world. Even if there already exist design solutions to enhance the energetic performances of the electronic systems, it appears more relevant to reduce the amount of data to go further. Rather than sampling at a fixed rate, non-uniform and sparse sampling appear as a relevant solution to drastically reduce the amount of useless data produced by the analog-to- digital conversion. Indeed, an excess of data produces more computation, more storage, more communications and also more power consumption. It is time to invent new analog-to digital conversions and sampling processing techniques. Moreover, such sampling techniques produce events that are perfectly compliant with asynchronous circuits, which natively process data on events. The talk will present a set of ideas exploring the event-based approach, covering event-based detection, non-uniform sampling and event-based processing techniques as well. This can be applied to many smart sensors, IoT platforms or image sensors.