How a tokamak works

At a hundred million degrees matter is ionized: nuclei and electrons run free and respond to a magnetic field, which makes them spiral around the field lines instead of crossing them. No material survives contact with that, so the plasma is held without touching the wall. A straight tube loses particles out of its ends; a ring, a torus, closes the path on itself and has no ends to escape through.

A torus alone is not enough: the toroidal field is stronger on the inboard side of the ring than on the outboard side, and that asymmetry makes particles drift outward. The tokamak's answer is to superimpose a second, poloidal field that wraps around the plasma in cross-section; the sum of the two gives helical field lines that average the drift out and bring the particle back in. And the cross-section is almost never a circle: most modern tokamaks elongate it into a D shape, because an elongated plasma with positive triangularity tolerates more pressure and more current before becoming unstable.

The toroidal field, the one running along the ring, is produced by D-shaped coils mounted around the chamber, each wrapping the torus like the ribs of a barrel. ITER has 18 of them, and reaching 5.3 tesla on the plasma axis means making them superconducting: their conductors are niobium-tin (Nb₃Sn) cooled with supercritical helium at 4 kelvin, about 269 degrees below zero.

The D shape is not a styling choice. A circular coil would have to resist an enormous bending moment, because the field is stronger on the inboard side; the D profile arranges the curvature so the magnetic forces act mostly as tension along the conductor. The modern alternative is to raise the field instead of the size: SPARC, from Commonwealth Fusion Systems and MIT, uses high-temperature superconducting tape to reach higher fields in a smaller machine.

On the axis of the torus, inside the hole of the ring, stands a column of coils: the central solenoid. It works as the primary of a transformer and the plasma as the secondary. As the solenoid's current changes, the magnetic flux threading the ring changes with it, and that variation induces a current of several million amperes in the plasma: 15 MA in ITER. That current does two things at once: it produces the poloidal field that closes the helical lines, and it heats the plasma ohmically.

From that comes the design's most awkward limit. A transformer only induces while the flux is changing, and the solenoid has a finite swing: once it runs out, the plasma current decays and the pulse ends. That is why a tokamak is intrinsically pulsed — ITER is aiming at pulses of about 400 seconds — and why a power plant would have to sustain much of the current non-inductively, with beams and radiofrequency waves, or accept operating in cycles. Losing control of that current has a name: a disruption, a collapse within milliseconds that dumps the plasma's entire energy onto the wall.

Around the assembly, in horizontal rings of different radius and height, sit the poloidal field coils. They do not confine on their own: they position and shape. They push the plasma inboard or outboard, stretch it vertically, give it the triangularity of the D and set where the X-point falls — the saddle point of the field where the closed magnetic surfaces stop being closed and the lines run off to the divertor.

An elongated plasma is more productive and also more fragile: elongation itself makes it vertically unstable, prone to running up or down within milliseconds. It is held by active control, fast coils that react to the measured position and correct it many times a second. The shape of the plasma is not something fixed when the machine is built: it is rebuilt on every shot, and held for as long as the shot lasts.

At the bottom of the D, below the X-point, sits the divertor: the part of the machine deliberately built to take what the plasma throws out. The open field lines running outside the last closed surface end on its targets, and with them arrive the exhaust heat and the helium ash, the product of the reaction, which has to be pumped away before it dilutes the fuel.

It is the most punished surface in a tokamak: the heat load, 10 to 20 MW/m², is ten times what a spacecraft takes on re-entry, so the targets are made of tungsten, the metal with the highest melting point, and are water-cooled. France's WEST, in Cadarache, has spent years testing the tungsten components ITER will use, and in February 2025 sustained a plasma for 1,337 seconds, just over 22 minutes, according to the CEA. The UK's MAST-U, in Culham, is testing another route: a Super-X divertor that lengthens the path of the field lines to spread that heat over more surface.

Between the plasma and the vessel goes the blanket, and its inner face is the first wall. The deuterium-tritium reaction releases 80% of its energy as 14.1 MeV neutrons, which do not see the magnetic field and fly straight out: the blanket stops them, heats up, and that is where a fusion power plant would collect the heat to raise steam. It also shields the superconducting coils, which work at 4 kelvin a short distance from a plasma at a hundred million degrees.

The blanket has a second job, and it is the one that decides whether fusion scales: tritium does not exist in any quantity in nature, it has to be made. A neutron from the reaction itself absorbed in lithium produces helium and tritium, so a blanket containing lithium can breed the fuel the machine consumes. No device has done it in a working regime yet: ITER's blanket is shielding, and breeding will be tested in test modules installed in two equatorial ports.

The induced current heats the plasma at first, but it loses its grip exactly when it is needed: the hotter the plasma, the less resistance it offers. The rest comes from two families of systems. Neutral beam injectors fire accelerated atoms that enter undeflected by the field, ionize inside and give up their energy through collisions. And radiofrequency antennas deposit power at the plasma's own frequencies: ion cyclotron heating (ICRH) on the nuclei, electron cyclotron heating (ECRH) on the electrons, the latter so focusable that it also serves to damp instabilities at a specific spot.

How well all that works is measured with two numbers. The triple product of the Lawson criterion combines density, temperature and energy confinement time (n·T·τE) and has to clear a threshold before heating from the alpha particles produced in fusion can take over from the external heating. And Q is the ratio between the fusion power released and the external power injected into the plasma: Q = 1 is scientific breakeven. The fusion power record in a tokamak still belongs to JET in 1997, 16 MW of fusion from 24 MW of heating, Q ≈ 0.67.

The whole assembly lives inside the cryostat, the vacuum envelope that thermally isolates the cold magnets from the room-temperature world. And to this day none of these machines was built to sell electricity: they are plasma physics experiments. In its final shots, on 3 October 2023, JET released 69 megajoules of fusion energy in a five-second pulse from just 0.2 milligrams of fuel, the record for fusion energy per pulse in a tokamak, as EUROfusion announced in February 2024.

ITER, under construction in Cadarache, is the first jump in scale: 500 MW of fusion power from just 50 MW of external heating, Q ≥ 10, sustained for about 400 seconds. Under the baseline the ITER Council endorsed in 2024, the start of research operation is planned for 2034 — with the machine already equipped — and the start of the deuterium-tritium phase for 2039. The next step the European community is discussing is DEMO, a machine that would have to breed its own tritium and put electricity on the grid as well as burn; it does not exist yet at any scale.