On October 19, the U.S. Department of Energy’s Brookhaven National Laboratory announced
that it has demonstrated the analysis of fusion plasmas
using a neural network chip built with superconducting circuits.
The chip handled data collected at 1 million samples per second,
and the team aims to apply it to future real-time control.
The Demonstration: Three Tasks Tested with Experimental and Simulated Data
The research was led by Brookhaven National Laboratory,
with participation from Princeton Plasma Physics Laboratory,
the DIII-D National Fusion Facility,
and Great Sky, which manufactured the chip.
Using a chip that mimics the brain’s networks with superconducting circuitry,
the team tested three tasks related to plasma control.
In two of the tasks, data collected at DIII-D was fed into the chip,
and algorithms provided by Brookhaven’s AI department
were used to classify anomalies and predict plasma disruptions.
In the third, data from TORAX, a simulator developed
by Google DeepMind’s fusion team,
was used to predict the onset of sawtooth instabilities in the plasma.
A paper describing the results is currently in preparation.
What Changes: Toward Analyzing Rapidly Changing Plasmas
According to the announcement, the chip was able to process
experimental data at 1 million samples per second.
The team says it has shown the potential to use the chip
for computations that monitor and control rapidly changing plasmas.
This is a demonstration of data analysis,
not an announcement that the chip has achieved plasma control on an actual device.
LC Signal’s Perspective: Next, Testing Whether It Works on Experimental Devices
The team plans to demonstrate a next-generation chip
with a neural network 10 times larger in early 2027,
and to deploy it on site at DIII-D later that year.
DIII-D has also indicated plans to integrate the chip
into its next experimental campaign
to study its use in diagnostics and control.
The next demonstration should show how the speed of analysis
can be put to use in operating actual devices.
Key Sources
-
Brookhaven National Laboratory official announcement, October 19, 2026:
First-of-a-Kind Super Intelligence Chips Successfully Analyze Fusion Plasmas
This article was prepared with AI assistance using publicly available information
as of October 19, 2026.
Announcement dates follow the dates shown on the official pages.
Demonstrated results are distinguished from future deployment plans.
“LC Signal’s Perspective” reflects this publication’s interpretation
of the available information.