NSF CoDec will develop a new technical approach for computational decarbonization problems through a novel class of computational techniques, algorithms, systems, and AI methods designed to sense, optimize, and reduce the operational, embodied, and lifecycle GHG emissions of societal infrastructure over intermediate time scales of minutes-to-years and spatial scales of communities-to-countries.
Computational decarbonization recognizes the unique role computing will play in decarbonizing society both as a “means” to automate and coordinate carbon-efficiency optimizations across time, space, and sectors, and as a “medium” that consumes increasingly significant amounts of energy but also has substantial temporal, spatial, and performance flexibility.
NSF CoDec includes two foundational research thrusts and three use-inspired research thrusts that are tightly integrated with each other.
FOUNDATIONAL
Theory & AI
This thrust designs general theory and AI approaches for optimizing carbon-efficiency at mesoscales, including learning-driven online optimization, optimization-in-the-loop learning, and multi-agent learning, while also considering economic incentives.
FOUNDATIONAL
Systems
This thrust designs general software platforms and carbon services for improving distributed infrastructure systems’ visibility, flexibility, and programmability to monitor and respond to changes in both their operational and embodied carbon emissions.
Use-Inspired
Computing & AI
This thrust adapts our foundational thrusts to optimize lifecycle carbon—both embodied and operational carbon—for AI applications, large-scale cloud platforms, edge networks, and client devices, while also extending equipment lifetimes.
Use-Inspired
Societal Infrastructure
This thrust applies our foundational thrusts to optimize lifecycle carbon for other critical large-scale societal infrastructure, including the built environment, electric transportation networks, and human-in-the-loop systems.
Use-Inspired
Coupled Infrastructure
This thrust leverages both our foundational thrusts and our insights above to develop new approaches for cross-domain decarbonization between the grid, computing, and domains such as buildings and transportation.
