DeepMind Uses AI To Control Plasma In Nuclear Fusion Reactor


Jack Dunhill


Jack Dunhill

Social Media Coordinator and Staff Writer

Jack is a Social Media Coordinator and Staff Writer for IFLScience, with a degree in Medical Genetics specializing in Immunology.

Social Media Coordinator and Staff Writer


Inside the TCV tokamak. Image Credit: Curdin Wüthrich /SPC/EPFL

Inside a tokamak, gaseous hydrogen fuel is subjected to intense heat and pressure until it becomes plasma hotter than the Sun’s core, creating the perfect environment for nuclear fusion. Powerful magnetic fields confine this plasma, preventing it from touching the walls and maintaining the reaction, but this is no easy feat – each coil must react to each variation in plasma, producing the perfect field to contain it and keep the reactor running.

Now, DeepMind and EPFL's Swiss Plasma Center (SPC) have come up with a new approach harnessing artificial intelligence (AI) to create one unified neural network to control all the magnetic coils, making a remarkable step forward in the pursuit of limitless energy generation. 


Their work was published in the journal Nature

Tokamaks are donut-shaped machines used in nuclear fusion research, and are likely to be the design future potential fusion reactors will use to generate energy if it ever becomes economically viable.

As the plasma swirls around the torus shape with a current equivalent to a bolt of lightning, energy is carried away from the plasma field by neutrons, which have no electrical charge and are unaffected by magnetic fields. These neutrons collide with an outer "blanket" where they are absorbed, heating the material and generating steam to power turbines.  

To control the magnetic field, earlier iterations of EPFL’s Variable Configuration Tokamak (TCV) used 19 individual algorithms for each coil, probing the plasma thousands of times a second and reacting accordingly to confine it.  


This new approach uses an AI designed by DeepMind that controls the variable voltage of each coil to maintain specific plasma configurations. Before they could use it on the real deal, the researchers used it on a simulation specifically designed to replicate the systems on the tokamak. 

Once trained, the algorithm could produce specific plasma configurations by using the exact settings required, and could even create various plasma shapes and maintain two plasmas in one simulation. 

Next was testing it on the real tokamak, and the results were extremely promising, mimicking the success it previously had in the simulation. 

Some of the shapes the AI could create using plasma. Image Credit: DeepMind and EPFL

Of course, this is only one key to the ultimate puzzle of fusion energy, but it is a huge step in the right direction.  


“Once this is done, this is not the end of the story. Then you have to make it a power plant,” said Gianluca Sarri, Professor at Queen’s University Belfast, UK, reports New Scientist

“And this AI is, in my opinion, the only way forward. There are so many variables, and a small change in one of them can cause a big change in the final output. If you try to do it manually, it’s a very lengthy process.” 


  • tag
  • nuclear fusion,

  • artificial intelligence,

  • physics,

  • nuclear power,

  • AI,

  • Deepmind