Networks-on-Chip-based Manycores for Distributed Intelligence in Cyber-Physical Systems

The development of the interactive computation paradigm, either under the internet-of-things (IoT) or cyber-physical systems (CPS) incarnations, offers the promise of unlocking crucial solutions in science, healthcare, transportation, energy, or business, yet poses significant challenges in terms of network infrastructure design and optimization. For example, while current search engines provide computational results on query processing within fractions of a second, in the future, we aim to be able to design systems that support complex information-exploration, -recognition, -mining and -synthesis processing that may even sustain brain-to-brain communication triggered by simple intentions. Such opportunity does not only calls for algorithmic strategies for dealing with cyber components running at different rates, but also requires efficient computing platforms that provide real-time sense- and decision-making. From a healthcare perspective, there is a need for architectures that can integrate sensed physical processes, analyze risk indices, and determine therapeutic-based control strategies.

Meeting the real-time requirements of these emerging applications calls for new design and optimization methodologies of CPS infrastructures with ultra-low power and thermal dissipation profiles, as well as ultra-low latency. To address these challenges, we develop algorithms for profiling applications from the CPS domain (e.g., biochemical stochastic simulation for viral detection studies, model predictive control for homeostasis management), identifying the computational and communication requirements and designing NoC-based accelerators. For instance, we proposed algorithms for designing efficient NoC-based multicores capable of solving large-scale nonlinear model predictive control (NMPC) problems. Relying on small form factor and energy efficient manycores, in the healthcare domain, the CPS can not only monitor the physiology and quality-of-life of individuals, but can also find signs of abnormality (e.g., bacterial/viral infection), and trigger control decisions over multiple space and time scales.