Federal Project Grant Award Summary

Georgia TECH Research Corp has received a $386,963 Project Grant from the National Science Foundation's Division of Information and Intelligent Systems under the Computer and Information Science and Engineering program (CFDA 47.070), awarded July 15, 2026, with completion targeted for June 30, 2031. This CAREER award supports research establishing a new investigative area that treats Artificial Intelligence (AI) infrastructure as an integrated cyber-physical system. The project addresses the critical gap between current AI training assumptions and hardware realities by developing frameworks and runtime algorithms that account for thermal hotspots, frequency throttling, power spikes, and energy constraints in large-scale distributed training environments.

The research deliverables are organized into three integrated thrusts: (1) a cross-platform framework capturing how thermal and power limitations constrain distributed training performance; (2) thermally-aware runtime algorithms incorporating real-time hardware telemetry into training decisions, including device placement optimization, adaptive batch scaling, and dynamic kernel scheduling across heterogeneous accelerators; and (3) a co-design methodology jointly optimizing accelerator architecture and memory hierarchies in relation to thermal and energy constraints. These products will advance energy efficiency and thermal management in data centers supporting large-scale AI model training, addressing the long-term research challenge posed by continued growth in AI model and infrastructure scale.