Superconducting neural chip analyzes fusion plasma in real time

The US Department of Energy's Brookhaven National Laboratory, Princeton Plasma Physics Laboratory, DIII-D National Fusion Facility and Great Sky jointly developed a first-of-a-kind Super Intelligence chip that mimics neural networks with superconducting circuitry and demonstrated it on three tokamak-control tasks. Two tasks fed real DIII-D data into the chip at one million samples per second for anomaly classification and disruption prediction. A third used TORAX simulated data from the Google DeepMind Fusion team to detect sawtooth instabilities, showing feasibility for real-time plasma-state simulation. A detailed paper is still pending and the device has not yet performed closed-loop control inside a reactor. The team plans a next-generation chip with a ten-times-larger neural network in early 2027, followed by deployment at DIII-D.