Ricerc@Sapienza

The Internet-of-Things (IoT) relies on a large number of sensors, actuators, processing units and networking units capable of giving services that could not be implemented otherwise.
An important requirement is the dynamic extendability of the network structure (adding nodes) and of the applications offered by the network (adding services).
The particular deployment of many IoT implementations requires energy efficiency maximization, while guaranteeing the required quality of service and the dynamic extendability. In fact, the IoT relies on the possibility of demanding data processing to centralized servers of even data-centers (cloud computing); yet, in order to maximize energy efficiency, local processing close to the sensors is pursued in present implementations (edge-computing), thus requiring ad-hoc computing resources in the edge-nodes.
The proposed project aims at developing an adaptable edge-computing node hardware architecture, featuring near-sensor reconfigurable signal processing acceleration units and memory units, along with the possibility of sharing computational workloads (i.e. program code and data) with other nodes and only ultimately with the cloud. The policy for defining the optimal microarchitecture, circuit and workload sharing parameters will be based on Artificial-Intelligence (AI) decision algorithms explored within the project, with genetic algorithms and support vector machine algorithm as first candidates.
The adaptability of the edge-node hardware architecture to the dynamic modification of the application and of the network composition is a key innovation. The innovation potential of the project is threefold: implementation of reconfigurable hardware microarchitecture in edge-computing nodes, implementation of workload sharing capability as an additional parameter in nodes' architecture reconfigurability, and the application of AI algorithms to tailor the configuration to the (dynamically changing) target application.