Federal Grant Award Summary

Purdue University received a $427,000 Project Grant from the National Science Foundation (NSF) Division of Information and Intelligent Systems under the Computer and Information Science and Engineering program (CFDA 47.070), awarded July 1, 2026, with completion targeted for June 30, 2029. The project, titled "NETS: Accelerating AI Workloads with Programmable Photonic Interconnects," develops networking algorithms and programmable photonic interconnects designed to enhance communication efficiency for artificial intelligence (AI) training and inference workloads. The research addresses current constraints in electrical interconnects by investigating how reconfigurable photonic circuit switching can dynamically adapt network connectivity to match the communication patterns of AI systems, with emphasis on collective communication operations such as Allreduce and All-to-All that are foundational to distributed training architectures.

The project is structured around three technical thrusts: optimization of collective communication algorithms for static photonic topologies; development of topology synthesis techniques and switching schedules for programmable photonic interconnects while accounting for realistic photonic constraints including reconfiguration delay and insertion loss; and analysis of workload-level implications for mixture-of-experts models under photonic interconnect architectures. Deliverables include theoretical analysis, packet-level simulation using NS3 and ASTRA-SIM simulation tools, and validation through hardware experimentation on an eight-GPU testbed connected to programmable photonic interconnects. The research directly addresses scalability challenges in modern AI systems as graphical processing units (GPUs) continue to scale and data exchange requirements expand beyond the capabilities of conventional electrical networking infrastructure.