Superconducting chip processes fusion data at 1M samples/sec

A superconducting neural-network chip processed fusion-plasma data at one million samples per second in a demonstration led by Brookhaven National Laboratory, keeping pace with signals collected at the DIII-D tokamak. The team used algorithms developed at Brookhaven for anomaly classification and disruption prediction. ^[1]

The researchers also tested the chip on simulated tokamak data, using it to predict the onset of sawtooth instabilities—rapid changes in plasma behavior. Brookhaven says the results demonstrate potential for real-time monitoring and control, but do not establish that the chip is ready for operational use. The lab said the team is preparing a paper and expects to test a larger chip at Brookhaven before a planned DIII-D deployment. ^[1]