A superconducting neuromorphic chip that borrows techniques from quantum computing has passed a battery of tests, including processing fusion-related data.

A superconducting neural network chip that borrows techniques from quantum computing has passed a series of tests that involved processing data from a fusion reactor. The work, part of the Department of Energy's Genesis Mission, paves the way for a larger chip that will be tested on fusion data next year. It also points to a new type of silicon that could outperform GPUs in supercomputers on both energy use and performance.

Great Sky is a chip startup based in Boulder, Colorado. CEO Jeff Shainline and his co-founders spun it out of the National Institute of Standards and Technology in 2024, after nearly a decade of development at NIST. The company emerged from stealth in March. It is building a chip that uses technologies common in quantum computers, such as superconducting circuits and optical connectivity, to implement an analog neural network in hardware rather than emulating one digitally.

Great Sky calls its approach superconducting optoelectronic network (SOEN). The company says SOEN can process data at significantly higher rates than traditional chips such as GPUs while using the same amount of energy. According to the company, the performance comes from combining superconducting Josephson junction circuits, which offer the highest speed-to-energy ratio of any known circuit element, with integrated photonic waveguides that enable long-range, single-photon optical communication.

"We use many of the same devices as quantum technologies—superconducting circuits, single-photon detectors—but we use them for classical, analog computation in a parallel, distributed architecture, which is ideal for AI," Shainline said in a blog post earlier this year.

"By constructing new hardware that enables more sophisticated architectures and algorithms while performing operations near the physical limits of speed and energy efficiency, we can transition to a completely different roadmap for scaling that doesn't require hundreds of billions in capex and gigawatt data centers," he added.

The claims were tested in a series of Genesis Mission challenges organized by Great Sky, Brookhaven National Laboratory, the Princeton Plasma Physics Laboratory (PPPL), and the DIII-D National Fusion Facility, which General Atomics operates in San Diego, California. Brookhaven led the development of a specialized chip based on Great Sky's technology, which was used to manage the tokamak at the heart of DIII-D, the largest operating tokamak in North America.

Brookhaven said the chip had "no problem" processing data from the tokamak at rates of up to one million data samples per second.

"This chip took data straight from a tokamak at millions of samples every second and kept up," said Steven Cowley, director of PPPL, which took part in the testing. "That's a real step toward practical fusion power, and I can't wait to see where we take it next."

The chip also passed a simulated test using TORAX, a tokamak simulator developed by the Google DeepMind fusion team. The test evaluates how well a system responds to rapidly changing plasma behavior during a fusion reaction.

Great Sky has completed four chip tapeouts for under $4 million and plans to build a neural network chip that is 10 times larger. The company aims to finish it by the end of the year, test it at Brookhaven in early 2027, and deploy it at the DIII-D tokamak later that year.

"It's incredible to see the team processing data from our tokamak," said Richard Buttery, director of DIII-D. "The folks here at DIII-D are very excited to see how this chip can impact diagnostics and control for the fusion industry."

Managing fusion reactors is one potential application of Great Sky's superconducting neural network chip, but not the only one. Great Sky, which received a phase one contract under the Genesis Mission, ultimately aims to build supercomputers with more neurons and synapses than the human brain, operating a million times faster.

"This is our largest chip to date, and it's great to see it processing high-speed data directly measured from a tokamak," Shainline said. "But this is just the beginning. This approach to AI hardware has so much to offer with regard to fusion, as well as other applications at the forefront of technology. Control and error correction of quantum systems, advanced cybersecurity, video intelligence with continual learning—these are all incredible opportunities as we scale up our systems."

Generating tokens for AI is another use case. Great Sky says its neuromorphic chip could generate 1 million output tokens for one penny, which it claims is 300 times less expensive than the best AI hardware available today, or $3 per 1 million tokens.