Plasma inside a fusion reactor changes in an instant, so control systems must keep up with its speed. Great Sky, together with Brookhaven National Laboratory, Princeton Plasma Physics Laboratory, and the DIII-D National Fusion Facility, has demonstrated a new AI chip built for exactly that challenge.
The chip belongs to the SOEN (superconducting optoelectronic network) family. Instead of following the conventional silicon-transistor route of digital computing, it implements analog neural networks with superconducting circuitry plus optical communication, breaking past the speed and energy-efficiency bottlenecks of traditional chips.
In live tests, the team fed real data from the DIII-D tokamak directly into the chip: at a sampling rate of a million samples per second, the chip completed anomaly classification and disruption prediction with ease. A third task used simulated data from TORAX, the tokamak simulator built by the Google DeepMind Fusion team; the chip successfully predicted sawtooth instabilities, which are critical to tokamak operation, confirming the feasibility of real-time plasma-state simulation and control in fusion reactors.
Great Sky says this is its largest chip to date, and in December it will demonstrate a next-generation chip with a 10x larger neural network. Applications could extend to quantum system control and error correction, advanced cybersecurity, and more.
Source: Read the original. This is a summarized digest for reference only.
