Brookhaven unveils superconducting AI chip to monitor tokamak plasma in real time
Brookhaven National Laboratory and Great Sky demonstrated a superconducting neuromorphic AI chip capable of processing fusion-plasma data at megasample rates.
On Monday, Brookhaven National Laboratory, in partnership with other DOE facilities and AI hardware firm Great Sky, unveiled a superconducting neuromorphic chip designed to analyze fusion-plasma signals in real time. The prototype processed a million-sample-per-second data stream from the DIII-D tokamak in San Diego, executing algorithms that detect anomalies and forecast disruptive events. Leveraging superconducting circuits, the chip runs its AI model at gigahertz frequencies with on-chip memory, offering potential edge-computing advantages over conventional computers.
Researchers emphasize the system is still at a demonstration stage and has not been connected to a live tokamak, leaving its effect on fusion reliability uncertain. Project leaders aim to build a successor chip with a neural network ten times larger, planning a debut at Brookhaven early next year and eventual installation at DIII-D. The effort highlights growing interest in applying AI to both plasma control and materials discovery.
Why it matters
Fast, on-chip AI could improve real-time control of fusion reactors, a key hurdle for practical fusion energy.
In this story
