A U.S. company is speeding up the path to practical fusion energy by using Google’s vast computing power.
By applying software that can improve on its own, TAE Technologies has cut down tasks that once took two months to just a few hours.
Google has lent the firm its expertise in “machine learning” in order to help accelerate the timeline for fusion. Nuclear fusion promises a plentiful supply of low-carbon energy, using the same process that powers the Sun.
Existing nuclear power is based on fission, where a heavy chemical element is split to provide a lighter one. Nuclear fusion works by combining two light elements to make a heavier one.
For fusion to become economically viable, it must first generate more energy than the amount being put in. But no one has reached this point, despite an eight-decade effort to “build a star on Earth”. The challenges are immense, but some in the fusion community hope that new thinking and disruptive technologies could help shatter this paradigm.
“I want to deliver fusion first, but anyone who does it is a hero,” TAE’s chief executive Dr Michl Binderbauer told BBC News.
TAE, located in leafy Foothill Ranch, southeast of Los Angeles, has raised over $880 million in private funding – more than any other fusion company. High-profile backing has come from Goldman Sachs, the Rockefeller family, and the late Paul Allen, co-founder of Microsoft.
The company’s 30-metre-long fusion cylinder – called C2W “Norman” after TAE’s founder, physicist Norman Rostoker, who died in 2014 – represents a different approach to the doughnut-shaped “tokamak” to be used for the world’s biggest fusion experiment, the multi-billion-dollar ITER project.
Controlling plasma at tens of millions of degrees requires a finely tuned system. Google’s expertise in machine learning – where computer algorithms improve with experience – has been used for “optimizing” TAE’s fusion device.
Optimization, or tuning for best performance, is carried out when something on the device changes, such as new hardware being added. This process once took around two months, but with machine learning, “we can now optimize in fractions of an afternoon,” Binderbauer explained.
Machine learning is also used to reconstruct what’s going on during a fusion experiment, or “shot”. Multiple strands of data can be pulled together for a deeper understanding of the process.
“That’s incredibly computer-dense and is a problem heretofore fairly few people even attempted to attack,” the CEO explains. He says the results of the partnership with Google could shave a year from the company’s longer-term schedule, which envisages a commercial fusion test device by 2030.
The company has already come a long way: Rostoker, a professor at University of California Irvine, founded it as Tri-Alpha Energy in 1998. Austrian-born Binderbauer was one of Rostoker’s PhD students, and became the company’s CEO four years ago. The two physicists chose TAE’s approach by starting with the requirements for a fusion power plant and working backwards.
By using a hot, electrically-charged gas called plasma, fast-moving particles can fuse, releasing energy. The $150-million “Norman” device crashes together two balls of plasma at supersonic speeds inside the tube. Magnetic fields, in what’s known as a field-reversed configuration (FRC), are used to control the process, which happens in 40 millionths of a second.
According to Prof Jeremy Chittenden, of Imperial College London, TAE is “doing something quite different to what everyone else is doing.” Rather than relying on the heat of the plasma to generate fast-moving particles for fusion, the device uses external particle beams which are fired into the hot gas, similar to what happens in a particle accelerator. “That’s your fusion source,” he explains.
Fusion efforts such as ITER will use fuel consisting of deuterium and tritium – two heavy versions of the element hydrogen. This produces energy from fusion at tens of millions of degrees Celsius, which is still at a lower temperature than other options. However, there are downsides: tritium is radioactive, wears down the insides of fusion reactors, and has a finite supply.
The “Norman” device powers its reactions with “regular” hydrogen and deuterium – a more benign, if less potent choice. But these are good surrogates for the fuel TAE eventually wants to move on to – hydrogen and boron. This produces no neutron particles and therefore little radioactivity, making machines straightforward to service and maintain. But this fuel also requires extremely high temperatures.
C2W “Norman” operates at around 70 million degrees C, but hydrogen-boron fuel requires temperatures to rise by a factor of 20-30, to several billion degrees C. It’s a major challenge: “Powers of ten are a big deal in science,” says Binderbauer, “Can we get to hydrogen-boron? I’m very convinced we can.”
Professor Roddy Vann, a plasma physicist at the University of York, UK, who works on fusion with tokamaks, said: “Whilst you’ve got to get the temperature right, the temperature and the density and the energy confinement time all have to be sufficiently high, simultaneously.”