Three Teams Attack Fusion's Most Dangerous Glitch at Once

Jiggling a Plasma 20 Times a Second Tames Its Worst Energy Bursts

What if the trick to preventing a fusion reactor's most dangerous eruptions was simply to wobble the plasma — like tapping a fizzy drink bottle to release the gas gently?

ELMs — edge-localized modes — are the plasma equivalent of a pressure cooker's safety valve releasing in one violent burst. The plasma at the edge of a tokamak builds up pressure until it can't hold itself together, then fires a chunk of energy at the reactor wall. Left unchecked in a machine the size of ITER, a single big ELM could erode the wall in ways that end the experiment early. A team at DIII-D, General Atomics' tokamak in San Diego, tried something refreshingly mechanical: they used magnetic coils to wobble the entire plasma column up and down 20 times per second — a displacement of a few centimetres. Think of tapping a bottle of sparkling water repeatedly before opening it. Instead of one big release, you get many small ones. It worked. ELM frequency jumped from roughly 5 natural bursts per second to 20 per second, while the energy released per event dropped from about 10% of the plasma's stored energy to below 1%. The heat slamming into the divertor — the component that takes the most punishment — was cut in half. What's attractive here is the simplicity: tokamaks already have the coils needed to do this. No exotic new hardware required. The catch: the team studied just two experimental discharges — one with jogs, one without — so the statistics are thin. Overall energy confinement stayed roughly flat rather than improving. And nobody has yet shown this scales to the much hotter, higher-current plasmas that a power plant would run. Promising, but one data point is one data point.

A Puff of Nitrogen Gas Stops Plasma Explosions and Boosts Efficiency

Researchers at China's EAST reactor shot a puff of nitrogen gas into superheated plasma — and the violent eruptions stopped, while energy confinement improved by a third.

EAST, the large tokamak run by the Institute of Plasma Physics in Hefei, China, has walls made entirely of metal — tungsten and molybdenum — which makes it one of the best proxies for what ITER will look like. In this experiment, the team injected a 50/50 mix of nitrogen and deuterium gas into the plasma during high-performance operation. The effect was striking. Large ELMs, which had been firing about 300 times per second, disappeared completely within roughly a second of injection — and stayed gone even after the gas pulse ended. Think of adding a pinch of yeast nutrient to an overactive fermentation: the fizzing doesn't stop, but it shifts into a steadier, gentler pattern. The confinement improvement was the real surprise. The H98 factor — a score for how well the reactor traps heat — climbed from about 0.9 to 1.2, and total stored plasma energy rose from roughly 160 kJ to 200 kJ, a 25% gain. Most ELM-suppression techniques cost you confinement; this one improved it. The team identified a new kind of wave at the plasma's edge, called an edge coherent mode, which seems to regulate energy transport more gently than ELMs do. Gyrokinetic simulations using the CGYRO code — a kind of very detailed plasma weather model — identified it as a dissipative trapped electron mode. Here is the honest catch: this is a single experimental discharge. Nobody yet knows whether nitrogen causes long-term contamination on a metal wall over many minutes of operation, or whether the effect holds in different plasma conditions. One beautiful experiment is not a validated technique.

A Custom Chip Predicts Plasma Explosions in 4 Microseconds

If you could see an ELM coming 4 microseconds before it hits, you might have just enough time to stop it — and a team at DIII-D has now built a chip that does exactly that.

The window for preventing an ELM is absurdly small — we're talking microseconds, millionths of a second. Standard computers, which handle many tasks at once, are simply too slow for this kind of real-time plasma control. A team from SLAC National Accelerator Laboratory and DIII-D has addressed this by installing a different kind of chip — an FPGA — directly inside DIII-D's plasma control system. Think of an FPGA as a piece of reprogrammable silicon. Where a general-purpose chip is like a Swiss Army knife that can do many things adequately, an FPGA is like a specialist tool custom-forged for one job and nothing else. This one runs a small neural network — a pattern-recognition algorithm trained on plasma data — that reads 160 diagnostic signals simultaneously at one million samples per second. From raw signal to ELM prediction: 4.4 microseconds. One practical detail worth noting: engineers can update the neural network's internal settings — its learned weights — without rebuilding the chip from scratch. That matters because plasma behaviour changes between experiments, and you want to retrain your predictor without months of hardware work. The catch is important. The paper describes a hardware deployment, not a closed control loop. The chip predicts ELMs; it has not yet been shown to trigger a suppression action fast enough to prevent one. That next step — from forecast to intervention — is a significant engineering gap. The paper is also light on statistical accuracy metrics, so we don't yet know how often the chip gets it right across different plasma conditions. This is a promising tool, not a finished solution.

Three papers, one problem. That's not a coincidence — it's a sign the fusion field knows ELM control is one of its most urgent practical challenges before ITER turns on. What's interesting today is the diversity of the toolkit: mechanical wobbling, chemical injection, and AI hardware, all tested in parallel on different machines across three countries. None of these is the answer. But together they suggest something important: ELM control will probably not come from a single elegant fix. It will come from a menu of techniques that operators mix and match depending on the machine, the operating mode, and what they can afford to trade off in confinement. The EAST nitrogen result is the most surprising of the three — it's rare for a mitigation technique to improve confinement rather than degrade it. If that result survives longer discharges and replication on other machines, it will jump to the top of that menu fast.

The nitrogen seeding result from EAST is the one to track in coming months — watch for follow-up discharges testing longer injection durations and checking for wall contamination, which is the obvious concern in a metal-wall machine. For the DIII-D chip, the key open question is whether the prediction can be wired into an actual suppression action in time — that closed-loop demonstration is the necessary next step. Several of these results are likely to appear with more statistics at summer plasma physics conferences, including the European Physical Society's Conference on Plasma Physics expected in late June 2026.

• Wikipedia: Edge-localized mode — what ELMs are and why they matter
• ITER Organization: Science — the basics of plasma control