A U.S. company is accelerating the path to practical fusion energy by leveraging Google’s vast computing power and machine learning expertise.<\/p>\n\n\n\n
By using self-improving software, TAE Technologies has reduced tasks that once took two months to a few hours.<\/p>\n\n\n\n
Nuclear fusion has the potential to ensure an abundant supply of low-carbon energy. This works by merging two light elements to produce a heavier one.<\/p>\n\n\n\n
For nuclear fusion to be economically viable, it must first produce more energy than is invested in the quantity. Unfortunately, no one has yet achieved this.<\/p>\n\n\n\n
“I want to deliver fusion first, but anyone who does it is a hero,” said TAE chief executive Dr. Michl Binderbauer.<\/p>\n\n\n\n
California-based TAE has already raised more than $880 million in private funds, with prominent backers including Goldman Sachs, the Rockefeller family and the late Microsoft co-founder Paul Allen. Former U.S. Energy Secretary Ernest Moniz sits on the board.<\/p>\n\n\n\n
Google’s machine-learning expertise – where computer algorithms improve over time – has been utilized for “optimizing” TAE’s fusion device.<\/p>\n\n\n\n
Optimization, or tuning for best performance, happens when something on the device changes, such as new hardware being added. This used to take around two months, but with machine learning “we can now optimize in fractions of an afternoon,” said Dr. Binderbauer.<\/p>\n\n\n\n
Professor Jeremy Chittenden from Imperial College London said TAE is “doing something quite different to what everyone else is doing.” Instead of depending on the heat of the plasma to produce fast-moving particles for fusion, the device uses external particle beams fired into the hot gas.<\/p>\n\n\n\n
Dr. Binderbauer says that TAE’s approach is also less prone to turbulence. In fact, the leakage rate decreases as the temperature in the device rises. “As you get more energetic, the behavior gets more manageable, more predictable, more reliable,” he says.<\/p>\n\n\n