How a stellarator works

A tokamak needs an electric current flowing through the plasma to close its magnetic field lines. A stellarator gets that same helical twist from the shape of the external coils: instead of driving a current, it twists the conductors into non-axisymmetric three-dimensional shapes, so the field comes out twisted to begin with. No transformer, no current to sustain.

A stellarator's plasma needs no induced current, and the bootstrap current that still appears is orders of magnitude smaller than a tokamak's, so there is no current to lose: the disruption disappears, the millisecond collapse that dumps the plasma's whole energy onto the wall. In principle the machine can hold the plasma in steady state for as long as external heating lasts. Wendelstein 7-X is designed for discharges of up to 30 minutes, per the IPP.

Wendelstein 7-X, at the Max Planck Institute for Plasma Physics (IPP) in Greifswald, carries 50 non-planar superconducting coils about 3.5 metres high and weighing six tonnes each, per the IPP. They are not fifty copies of one part: there are five different geometries, repeated across the machine's five field periods. All seventy coils of the set were tested on CEA's rig at Saclay, quench test included — the momentary loss of superconductivity.

What made that design possible was numerical optimization. Instead of drawing the coil and seeing what field comes out, the codes start from the magnetic field that is wanted — one that minimizes fast-particle losses and parasitic bootstrap currents — and work backward to the conductor shape that produces it. W7-X is the machine that tested that method at scale: its coils produce a quasi-isodynamic field, in which the field-strength contours close in a way that keeps particle orbits inside the plasma instead of drifting out, which lowers neoclassical transport and the bootstrap current together. The IPP measured that neoclassical losses fell to about 30% of the heating power.

On top of the 50 modular coils sits a second set of 20 planar coils, superconducting as well: 70 in all, per the IPP. The planar ones are ordinary rings, each in its own plane, and their job is not to confine but to tune the configuration: added to the field of the modular coils, they let the magnetic configuration be shifted without touching the geometry of the twisted pieces.

None of this tolerates approximations. The 6,200 wall tiles and the ten divertor modules were installed to a precision of one to two millimetres, per the IPP; before the first plasma the field was measured with flux-surface diagnostics, to check that the closed magnetic surfaces were where the calculation said they would be.

The edge of W7-X does not end in a smooth surface: in its standard configuration it carries a chain of magnetic islands that look like five lobes around the plasma in a poloidal cut. The island divertor uses that geometry instead of fighting it: it puts the plates exactly where the islands meet the wall, and that is where the particles escaping the edge are collected.

In W7-X the plates follow the twisted contour of the plasma edge as ten broad strips along the wall of the vessel, per the IPP. There the incoming particles, along with the impurities they carry, are neutralized and pumped away. The original graphite tiles were later replaced by water-cooled elements of carbon-fibre-reinforced carbon, so the heating energy could be raised without overloading the wall.

With 2.5 tesla on the plasma axis the electrons gyrate at about 70 GHz; W7-X's ECRH heats at the second harmonic of that frequency, 140 GHz, per the IPP. Each gyrotron delivers close to 1 MW for 30 minutes; ten were available in campaign OP1.2, with up to twelve planned and an ultimate target of 15 MW, and about 7.5 MW were actually coupled into the plasma.

ECRH is, per the IPP, the only heating system already able to run continuously, which is exactly what a machine built for steady state needs. The first hydrogen plasma, on 3 February 2016, was a 2 MW microwave pulse that reached 80 million degrees and lasted a quarter of a second.

In 2018 W7-X set the stellarator world record for the fusion product: 6 × 10²⁶ degrees × second per cubic metre, with an ion temperature of about 40 million degrees, a density of 0.8 × 10²⁰ particles per cubic metre and an energy confinement time of 200 milliseconds, per the IPP.

On 15 February 2023, after an upgrade that added water cooling to the wall components and doubled the heating power that could be coupled in, it sustained a plasma for eight minutes with an energy turnover of 1.3 gigajoules and 2.7 MW of average heating: 17 times the best value before the upgrade, 75 megajoules. On 22 May 2025, closing campaign OP 2.3, it set a triple-product record sustained for 43 seconds, with about 90 frozen hydrogen pellets injected by an Oak Ridge National Laboratory device, temperatures up to 30 million degrees and a plasma-to-magnetic pressure ratio of 3%; the energy turnover rose to 1.8 gigajoules over 360 seconds.

The other reference stellarator is LHD (Large Helical Device), at Japan's National Institute for Fusion Science (NIFS), in Toki. It is a heliotron: it uses a pair of continuous helical coils instead of modular ones, an older but proven architecture, with a 3.9 m major radius, a 0.6 m minor radius, a 3 T design field and 30 m³ of plasma, per NIFS. It began its deuterium experiments in March 2017 and in that first campaign reached an ion temperature of 10 keV associated with an internal transport barrier.

The advantage is clear: steady state with no current to sustain and no disruptions to fear. So is the cost: every coil is a twisted part unlike the others, to be manufactured and assembled to millimetre tolerances across a volume of several metres. For decades that price left the stellarator behind the tokamak, simpler to build and to model.

What changed the balance was numerical optimization, and the machines that validated it. Neither W7-X nor LHD is a power-plant candidate as built: both are plasma-physics experiments — W7-X does not even use deuterium-tritium fuel — aimed at demonstrating that an optimized stellarator confines as well as a tokamak and sustains it for longer. The next step the community discusses is a stellarator burning deuterium-tritium, with shielding and heat extraction, something that does not exist today at any scale.