This site is fictional demo content. It is not real news or affiliated with any real organization. Do not treat it as fact or professional advice.

NextPaper Logo
nextpaper.ccCYBER CHRONICLE
RSS
EN中文

Full article

FULL TEXT

View this issue
Deep diveENERGY

FusionMind Uses AI to Tame Plasma — Extending Fusion Stability From 3 Seconds to 45

TerraFusion releases FusionMind, an AI plasma control system that uses deep reinforcement learning to adjust magnetic confinement parameters in real time. In tokamak experiments, it extended stable plasma duration from 3 seconds to 45 seconds — a 15x improvement.

Taming a Miniature Sun With AI — A Breakthrough in Fusion Plasma Control

Nuclear fusion — the process of fusing light atomic nuclei into heavier ones while releasing energy — is the ultimate energy prize. But achieving controlled fusion faces a central challenge: plasma (ionized gas heated to 100 million degrees) is wildly unstable. The slightest perturbation causes it to escape magnetic confinement and cool rapidly.

TerraFusion's FusionMind system attacks this plasma-control problem with AI. Released on April 8, the control system uses deep reinforcement learning to achieve an unprecedented stable plasma duration in a tokamak device — extending it from 3 seconds to 45 seconds.

FusionMind's control loop runs at 10,000 Hz — adjusting magnetic-field coil currents 10,000 times per second. The system ingests real-time temperature, density, and magnetic-field data from over 500 sensors distributed around the plasma, then the AI computes an optimal field-adjustment strategy within 100 microseconds and sends it to the actuators.

"Plasma instabilities unfold on millisecond timescales," explained Dr. James Chen, TerraFusion's chief scientist. "A human operator simply can't react fast enough. Only AI can make effective control decisions at that speed."

The AI model was trained on more than one million simulated plasma instability events. In live experiments, the system successfully suppressed tearing-mode instabilities, edge-localized modes, and neoclassical tearing modes — among the most troublesome forms of plasma disruption.

While 45 seconds of stability is still far short of the hours-to-days continuous operation required for a commercial fusion power plant, it represents a 15-fold improvement over the previous best. More importantly, FusionMind's control strategy shows strong scalability — TerraFusion believes it will perform even better on larger devices.

TerraFusion has signed a cooperation agreement with ITER (the International Thermonuclear Experimental Reactor) and plans to integrate FusionMind into ITER's control system by 2031.

The company has closed a $500 million Series C at a $3 billion valuation. Investors include Breakthrough Energy Ventures (Bill Gates's fund) and Temasek.