Fusion Machines: Pinches, Focus Devices and Inertial Drivers

If you want to build something after reading this book, start here. Fusion is the most efficient energy release available inside ordinary three-dimensional physics, and unlike most of what follows it has a sixty-year experimental literature, hundreds of working machines, and a hiring market. Every family in this chapter is a different answer to one question — how do you hold matter hot enough and dense enough for long enough that nuclei fuse — and each answer produces a different machine, a different diagnostic and a different set of trades. Learn to tell them apart and you can read any fusion announcement in the world and say what it does and does not establish.

Before a word of theory, here they all are in one picture.

The chart above carries every machine this chapter names on the two quantities every fusion machine trades between — how densely the fuel is packed, and how long the machine holds it — so the difference between the families reads as a diagonal rather than a list. Choose a confinement approach, then a machine, and the read-out says what it squeezes, for how long, and exactly how far that machine's own standing goes.

Then look at the machines themselves. The gallery draws one cutaway per family, so each name on the chart above becomes a thing with electrodes, coils or beams in it.

The machine gallery

Six ways to hold fuel together long enough for nuclei to fuse, drawn as cutaways. Choose an architecture and the panel says how it confines the fuel and what that family has actually measured — each figure the one its own paper reports, and nothing added to it.

Architecture

The sheared-flow Z pinchThe dense plasma focusInertial electrostatic confinementThe tokamakThe stellaratorLaser inertial drive

Selected architecture: The sheared-flow Z pinch

How it holds the fuel. A large current runs straight down a column of gas and the column's own magnetic field squeezes it. Plasma moving faster on the axis than at the edge — a sheared axial flow — keeps the kink from growing.

What it has measured. On FuZE the plasma stays quiet for about sixteen microseconds and fusion neutrons come out steadily for five of them, at a pinch current near 200 kiloamps and a pinch radius of about three millimetres. The yield rises with the square of the deuterium fraction and vanishes with no deuterium, which is the signature of thermal fusion rather than a stray accelerated beam.

Quiet interval:

1.6E-5 s

Steady neutron production:

5E-6 s

Pinch current:

2E5 A

Pinch radius:

3E-3 m

Neutron yield per pulse:

1.25E5

A peak ion energy is not a temperature, and a neutron yield is not an energy balance. Each figure below is labelled with the quantity its own paper reports.

Read the sheet behind these figures

What every machine is trying to do

John Lawson's criterion is the sum every reactor has to pass. Get the density high enough, the temperature high enough, and hold it long enough that the energy released exceeds the energy invested. The combined quantity is usually written as the triple product.

The triple product — density times temperature times energy confinement time(18.1)

$$n\,T\,\tau_E \;\gtrsim\; 3\times 10^{21}\ \text{keV}\,\text{s}\,\text{m}^{-3}$$

The classic version was written for a power station on the ground, which can weigh what it likes and dump waste heat into a river. A generalised burn criterion for space keeps the magnets, radiators and power conversion inside the equations, adds a hot-ion mode, an uneven fuel mixture and a separate confinement time for the charged products, and then scores four fuels by what they cost to fly. That is the version this site cares about, because a reactor that has to launch is the first engineering gate between a power plant and a ship.

Two precision rules before we go any further, and they are not pedantry — each one protects a real claim from an easy rebuttal.

A peak ion energy is not a temperature. When a machine reports ions at a few hundred kiloelectronvolts, that is the energy of particles in a microscopic, nanosecond-lived structure, not the temperature of a bulk plasma. Write "peak ion energies in the range proton–boron-11 fusion requires," which is the real and remarkable statement.

Net gain is not over-unity. A reactor that returns more energy than it draws is burning fuel. That is an enormous achievement and it is an ordinary one. Keep gain and over-unity in separate sentences, always.

The instrument that follows puts the third quantity on the chart rather than leaving it in a slider, and it names the three gains the literature keeps apart — scientific, engineering, and wall-plug — each with its own denominator. What the chart itself reports is none of them: it is a distance from a criterion.

The pinch family: squeeze it with its own current

The Z-pinch is the simplest fusion machine anyone ever drew. Run a huge current straight down a column of matter and the column's own magnetic field crushes it inward. No external magnets, no auxiliary heating. It was largely abandoned in the 1960s because the column kinks and necks apart in about a nanosecond.

Two lines revived it. The first replaced the column with a cage of fine wires. At Imperial College, about a million amps in a quarter of a microsecond showed what actually happens inside: each wire stays put as a solid core wrapped in plasma, the plasma streams inward ahead of everything else, and because the streaming is uneven the imploding sheath leaves trailing mass behind that carries some of the current on its own path — which is exactly the compression the pinch loses at stagnation. Array geometry, not just stored energy, sets what these machines do. Even the choice of wire material changes the discharge.

The second repaired the instability directly. Give the column a sheared axial flow, plasma moving faster on the axis than at the edge, and the kink stays suppressed. On the FuZE device the plasma goes quiet for about sixteen microseconds and fusion neutrons come out steadily for five of them — roughly five thousand times as long as the column should have survived. Read the diagnostic the way the authors do: the yield rises with the square of the deuterium fraction and vanishes with no deuterium, which is the signature of thermal fusion rather than a stray accelerated beam. That distinction is the single most useful thing in this chapter. Zap Energy has since taken the concept to a repetitive, liquid-metal-cooled machine firing every ten seconds onto a curtain of molten bismuth, three runs of 1,080 shots each.

Alongside them sits magnetized liner inertial fusion, which trades implosion speed and perfect symmetry for a magnetic field. A beryllium tube of deuterium, a 10 tesla axial field, a laser preheat, then nineteen million amps in a hundred nanoseconds: the fuel stagnated near 3 keV and produced up to two trillion neutrons, while shots missing either the preheat or the field produced almost nothing. The field insulates the fuel and traps the charged products. Simulations of the same scheme with a deuterium-tritium ice liner reach gain above 100 at sixty million amps. Sandia's older light-ion driver programme is the pulsed-power branch of the same instinct: cheap stored electrical energy instead of lasers.

The kink, and the flow that holds it off

Two columns of the same plasma, side by side. The left one is a plain Z pinch: its own current squeezes it, and the kink grows until the column necks apart. The right one carries a sheared axial flow — plasma moving faster on the axis than at the edge — and stays quiet for the interval FuZE measured. Move the clock and watch the two separate.

Time since the pinch forms

Time since the pinch forms

1E-9 seconds

Advance the clockBack to the start

• Steady neutron production
• Measured quiet interval

At this instant: The column is still holding together. The sheared column is quiet, and the kink has not grown. Fusion neutrons are coming out steadily.

Measured quiet interval:

1.6E-5 s

Steady neutron production:

5E-6 s

Pinch current:

2E5 A

Pinch radius:

3E-3 m

Neutron yield per pulse:

125,000

Instability lifetimes survived:

5,000

Read the diagnostic the way the authors do: the yield rises with the square of the deuterium fraction and vanishes with no deuterium, which is the signature of thermal fusion rather than a stray accelerated beam.

Read the sheet behind these figures

The dense plasma focus, and the plasmoid it makes

This is the machine this site returns to most, because of what it builds. Two coaxial electrodes, a capacitor bank, a gas fill. The current sheet sweeps down the barrel and collapses on the axis, and along the way it does something no designer imposed: the current breaks into filaments, the filaments pinch, and then a kink instability coils the merged filament until a plasmoid knots itself into existence — the same way a telephone cord kinks. The structure is tens of micrometres across and holds a large share of the whole machine's energy.

The electrode geometry above is the apparatus. Filament and pinch morphology need time-resolved imaging, current, spectral and particle diagnostics; a generic torus is not a measurement of a focus shot.

The field's working manual is a simulation code fitted to a measured current waveform with four parameters, and it tracks real behaviour well enough to predict what a machine will do before it is built. The physics of the pinch and the beams it throws off is measured at nanosecond resolution across machines from a kilojoule to a megajoule, which is why more than thirty laboratories run them: the pulse a focus delivers to a target is the closest available stand-in for what a reactor wall must survive.

LPPFusion's line is the aneutronic one. Their published overview reports the highest confined ion energy measured in any fusion experiment, a mean of 240 keV — in the range proton–boron-11 needs — plus the cleanest fusion plasma on record after fitting a hand-polished beryllium anode, and a physical mechanism for the fuel: at gigagauss field strengths electrons can only accept energy in discrete steps, so the plasma stops cooling itself with X-rays. The preparations for firing on hydrogen-boron read like a work order — decaborane vapour as the fuel, bought isotopically pure, with detectors that count the two rare side reactions to recover ion energy, density and confinement time from outside the chamber.

Dense-plasma-focus devices produce fusion reactions. Testing the aneutronic route requires a reported proton–boron-11 yield from FF-2B, with the beam-target and thermal fractions separated.

And the same device has been costed as an aircraft powerplant: a parametric study for a reusable single-stage-to-orbit vehicle sweeps thrust, specific impulse, capacitor energy density and gain, and reports how many gigawatts are left over for everything else on board once the engine runs better than break-even.

Inertial electrostatic confinement: fusion you can own

The cheapest working fusion machine in the world is a wire cage in a vacuum chamber. Philo Farnsworth — the inventor of electronic television — patented it in the 1960s: an outer cathode fires electrons inward through a nearly transparent inner anode, and where they crowd at the centre they build a virtual cathode, a well of negative potential standing in open space with no physical electrode at all. Ions fall into that well, oscillate through it, and fuse.

It works. A tabletop device at 40 kilovolts and 2 milliamps produced ten thousand neutrons a second, steadily — and, more valuably, the authors asked where the fusions were happening and answered it two ways, from the current scaling and from a probe. Most reactions were fast ions striking background gas rather than beam meeting beam. Modern particle simulations set against probe measurements show the same self-organisation: bright spokes through the cage openings, a virtual anode built out of nothing but trapped ions, and a nested virtual cathode of electrons at the highest voltages.

This is the family a school, a workshop or a determined amateur can actually build, and it is where a great many plasma careers start.

Inertial drivers, and the fuel that decides the machine

Squeeze a small quantity of fuel hard enough and fast enough and it ignites before it can fly apart. The National Ignition Facility is the famous version of that idea, and Daniel Jassby's audit of the fusion ledger is the right way to learn it. A fusion claim only becomes readable when energy out, energy in and fuel inventory are on one page — and when the page says which gain is being quoted and against which denominator. Here is that page, for every machine this chapter names.

The fusion ledger

A fusion headline is a fraction, and nearly every argument about fusion is an argument about its denominator. Put three things beside every machine — the energy it got out, the energy that figure is being divided by, and the fuel it had to hold to do it — and the claims sort themselves out. Choose a machine and read its three lines. The axis below places a machine only where its own sources publish a gain, and seven of these twelve do not.

Choose a machine

JET, the 2021 deuterium–tritium campaignThe National Ignition Facility, the shot of 8 August 2021HB11, laser-driven proton–boron-11Magnetized liner inertial fusion with a deuterium–tritium ice liner (simulation)A dense plasma focus as an aerospace powerplant (design study)Magnetized liner inertial fusion on the Z machine, 2014FuZE, the sheared-flow-stabilized Z pinch, 2019Century, Zap Energy's repetitive Z-pinch core, 2024FF-1, the LPPFusion dense plasma focusA tabletop inertial electrostatic confinement deviceProton–boron-11 in the Large Helical Device, 2023The laser block-ignition road map (proposal)

Back to the reference point

• Measured
• Simulated
• Proposed
• No published gain

JET, the 2021 deuterium–tritium campaign

Energy out:

59 megajoules of fusion energy in one shot, at a transient peak fusion power of 13 megawatts.

Energy in:

The denominator is the external heating deposited in the plasma. The campaign is reported as a gain rather than as a number of joules in.

Fuel inventory:

At least 100 milligrams of tritium per shot. The whole civilian tritium inventory is about 30 kilograms, held from CANDU reactors at 30,000 to 100,000 US dollars a gram.

Gain, as quoted:

about 0.4 (Measured) · no more than 0.2 (Measured)

Reading: Read the denominator before the headline. The peak gain is about 0.4 and the quasi-steady figure 0.33, below the 0.67 the same machine reached in 1997; and because pure deuterium beams were fired into an overwhelmingly tritium plasma, at least three quarters of the output came from beam ions striking thermal ions, which leaves a thermonuclear gain of no more than about 0.2.

The reference point. JET is a tokamak, and this chapter is about pinches, focus devices and inertial drivers; it is here because it is the worked example in the sentence above, and because it is what every other ledger on this page gets read against.

Read the sheet

Seven of the twelve machines here publish no gain at all. That empty column is not a gap in this table — it is what the literature says, and it is the most useful thing on the page. A neutron yield, a triple product or a delivered energy per shot is a real measurement, and it is not a ledger. None of these seven is placed on the axis, because placing one would take a number nobody published.

A gain above one is a reactor burning fuel, and nothing else. That is an enormous achievement and an ordinary one, and it is not over-unity — keep the two in separate sentences, always.

The open question this ledger draws is one empty region. No machine running aneutronic fuel has reported energy out above energy in with its fuel inventory published. That measurement, on any machine here, is what would fill it.

Now the fuel. Deuterium-tritium is the easiest reaction to light and the worst one to build around: it sends most of its energy out as 14 MeV neutrons, which will not turn in a magnetic field, cannot be aimed, and damage the machine on the way past. The aneutronic alternative is the reason this chapter exists.

The reaction that changes the machine(18.2)

$$p + {}^{11}\mathrm{B} \;\longrightarrow\; 3\,{}^{4}\mathrm{He} \;+\; 8.7\ \mathrm{MeV}$$

The reaction has now been made where a reactor would actually make it. A TAE and Japanese team dropped boron powder into the Large Helical Device, fired 2 MW hydrogen beams at 160 kilovolts, and counted the first proton-boron-11 fusion in a magnetically confined plasma — about 150,000 counts a second with boron present, under a thousand without. The laser route is set out as a full road map with numbers: a picosecond petawatt pulse pushing a solid-density plasma block by light pressure, a second pulse wrapping the fuel in kilotesla fields, fourteen milligrams of fuel and a gigajoule a shot. The status report from inside that programme is admirably honest: the best shots sit about four orders of magnitude below breakeven, and it names every route it knows for closing the gap. And the standard theoretical objection — that proton-boron radiates away more than it makes — is answered on its own ground by the degenerate-plasma calculation, which finds a window where the loss channels themselves change and the sum comes out positive.

Magnetic confinement, and the compact machines

The mainstream route is worth knowing accurately, because everything else is measured against it. The incumbent roadmap is written by two people inside it, and its central observation is that the triple product climbed steadily from the 1970s and then flattened. Its central number is the one nobody puts on a poster: the world's commercial tritium stockpile is about thirty kilograms and a single large machine wants two thirds of it, so every future plant must breed its own fuel from lithium in a blanket that has not yet been demonstrated.

The compact-torus family is the alternative this chapter is most interested in, because its plasmas hold their own shape. The physics is magnetic helicity: helicity is very nearly conserved while energy is not, so a tangled magnetically dominated plasma sheds energy, keeps its linkage, and relaxes into the lowest-energy state still open to it — the same argument covering a laboratory spheromak, a solar corona loop and a jet thousands of light years long. Built on it: the dynomak, a complete spheromak power plant with blanket, magnets, shielding and cost estimate; the Princeton field-reversed configuration, a metre-and-a-half-wide tube heated by an odd-parity rotating magnetic field, running deuterium and helium-3 for about a thousand times less neutron loading than a deuterium-tritium tokamak; a national-laboratory case for small field-reversed reactors made directly to the committee that sets the decade's priorities; and General Fusion's acoustically driven magnetized target fusion, where hundreds of pneumatic pistons hammer a sphere of spinning liquid lead-lithium and geometric focusing turns one gigapascal at the wall into ten at the plasma surface.

One more lever belongs here because it is the purest example of engineering a plasma rather than out-building it. Alpha channelling launches a wave that takes energy from fusion-born helium nuclei before the electrons radiate it away and hands it to the fuel ions instead; a minority-ion catalysis scheme uses a trace of heavy ions as a middleman so that one wave can talk to both populations. Keeping the fuel ions hotter than the electrons is exactly the condition the harder fuels need.

Where the vacuum side plugs in

This is Part III, so the join matters. Two machines in the literature compress with fields generated by accelerated charged matter rather than with a bigger magnet: the peer-reviewed plasma compression fusion device, whose spinning charged cones heat the fuel with their own radiation and squeeze the plasma with the same fields. And the hypothesis this site tracks, in the coiner's own narrowed words: a driven, out-of-equilibrium electromagnetic vacuum environment that increases nuclear tunnelling rates by modifying the effective interaction potential — the screening, the polarisation, the correlated encounter rate — while leaving the nuclear energy release and the conservation laws intact. Lowering the Coulomb barrier itself is stated by its author as a hope rather than a claim.

What to watch: that hypothesis names its own experiment. Hold temperature and density fixed, drive the vacuum environment, and a measured reaction rate either moves or it does not. Every machine in this chapter is an instrument capable of running that test.

Measure every channel: the lesson of Castle Bravo

The most instructive fusion measurement in history was not a reactor. On 1 March 1954 the United States fired the Bravo shot of Operation Castle at Bikini Atoll. It yielded 15 megatons, nearly three times the six megatons its planners had estimated. The fuel included lithium deuteride, and part of that lithium was lithium-7, which the designers had treated as essentially inert. It was not. Struck by neutrons it made tritium, itself a fusion fuel, and that reaction — present in the fuel, absent from the model — carried the yield far past the prediction. The Los Alamos designer Carson Mark put the explanation in one line in July 1954: lithium-7 had been assumed to be a much less good fuel than lithium-6.

The principle is the one every energy result on this site has to meet, whichever direction the surprise runs. Every ingredient is a channel until a measurement shows it is not. A component assumed to sit out the reaction is an assumption, and the output meter is where the assumption is tested. The diagnostics engineer's job, named below, exists for exactly this: account for every input, every product and every path, and let the measured output, not the design sheet, have the last word.

The jobs this chapter feeds

This is the most directly employable subject on the site, and the list is not hypothetical — every one of these roles is being hired for now, by national laboratories and by private fusion companies. Plasma physicist. Pulsed-power engineer. High-voltage and capacitor-bank engineer. Vacuum and cryogenic systems engineer. Plasma-diagnostics engineer — the person who builds the instrument that tells you whether the neutrons were thermal or beam-target, which is the difference between a result and a press release. Neutronics engineer. Magnet engineer. Target-fabrication technician. Power-conversion engineer. Fusion systems engineer. If you want to build a warp drive one day, this is the trade to learn first, because a metric-engineering programme needs a small source of very large power before it needs anything else.

What the field added — July to September 2026

The season's clearest gain was architectural rather than experimental: the direct-conversion picture was taught properly. In an aneutronic machine the products emerge as an ion beam one way and an electron beam the other, and current is induced directly in a coil — no steam, no turbine, no Carnot ceiling. That is the reason the fuel choice decides the machine, and it is the reason a fusion-powered craft is a different object from a fission-powered one.

The season also produced the narrowed form of the vacuum-catalysed hypothesis quoted above, which matters here because it converts a large claim into a bench measurement any of these machines could make. And it produced a caution worth repeating: a fusion reactor that returns more than it draws is burning fuel. Say so, in its own sentence, every time.

---

Research measurements

• Independent replication at higher current, and the same deuterium-fraction test
• A reported proton-boron yield with the beam-target and thermal fractions separated
• A shot closing a substantial part of that gap on a named facility
• A machine reporting energy out above energy in, on aneutronic fuel, with the fuel inventory published
• A measured reaction-rate change at fixed temperature and density, with the drive as the only variable

Sources

The machines, in the order an engineer meets them — with the diagnostic that separates a result from an announcement kept in front.

Pinches and liners

• S. V. Lebedev et al. (2004), "Implosion dynamics of wire array Z-pinches: experiments at Imperial College." /library/stm-33d8266a67.
• Y. Zhang, U. Shumlak et al. (2019), "Sustained neutron production from a sheared-flow stabilized Z pinch," Phys. Rev. Lett. 122, 135001. /library/stm-751d0ee864.
• M. C. Thompson et al. (2025), "Century: Zap Energy's 100-kW-scale repetitive sheared-flow-stabilized Z-pinch system." /library/stm-49324933a9.
• M. R. Gomez, S. A. Slutz et al. (2014), "Experimental demonstration of fusion-relevant conditions in magnetized liner inertial fusion." /library/stm-d11937cf2a.
• S. A. Slutz & R. A. Vesey (2012), "High-gain magnetized inertial fusion." /library/stm-b7ffaf8ee4.

The dense plasma focus

• S. Lee & S. H. Saw (2017), "The plasma focus — numerical experiments, insights and applications." /library/stm-12981c85c9.
• V. A. Gribkov (2015), "Physical processes taking place in dense plasma focus devices." /library/stm-1c5e68db7e.
• E. J. Lerner et al. (2023), "Focus fusion: overview of progress towards p-B11 fusion with the dense plasma focus." /library/stm-05100e66da.
• E. J. Lerner & S. M. Hassan (2024), "Preparations for pB11 tests in the FF-2B dense plasma focus." /library/stm-8f6027c3eb.

Inertial electrostatic confinement

• P. T. Farnsworth (1966), "Electric discharge device for producing interactions between nuclei," US Patent 3,258,402. /library/stm-3a226d4c01.
• K. Yamauchi et al. (2001), "Neutron production rate and plasma characteristics of spherically convergent beam fusion." /library/stm-e386fdbf31.

Aneutronic fuel and direct conversion

• R. M. Magee et al. (2023), "First measurements of p11B fusion in a magnetically confined plasma," Nature Communications. /library/stm-9e2a22de0f.
• H. Hora et al. (2017), "Road map to clean energy using laser beam ignition of boron-hydrogen fusion." /library/stm-b5a7103035.
• W. McKenzie, D. Batani, H. Hora et al. (2023), "HB11 — understanding hydrogen-boron fusion as a new clean energy source." /library/stm-8dfcd2d56d.
• S. Son & N. J. Fisch (2004), "Aneutronic fusion in a degenerate plasma." /library/stm-9aa4e6f741.

Magnetic confinement, compact tori and systems

• S. Takeda & R. Pearson (2019), "Nuclear fusion power plants." /library/stm-4fed2cae63.
• D. Jassby (2022), "The quest for fusion energy" — the ledger, in its own numbers. /library/stm-2d18170fa6.
• P. M. Bellan (2017), Magnetic Helicity, Spheromaks, Solar Corona Loops, and Astrophysical Jets. /library/stm-17bc3baf7b.
• D. A. Sutherland, T. R. Jarboe et al. (2014), "The dynomak." /library/stm-45bcd277ce.
• C. Galea, S. Thomas, M. Paluszek & S. Cohen (2023), "The Princeton field-reversed configuration for compact nuclear fusion power plants." /library/stm-1aa14f2192.
• M. Laberge et al. (2013), "Acoustically driven magnetized target fusion." /library/stm-499bfa4562.
• D. Petkow et al. (2012), "Generalized Lawson criterion for magnetic fusion applications in space." /library/stm-626cb2cbf2.
• S. Knecht, R. Thomas, F. Mead, G. Miley & D. Froning (2006), "Propulsion and power generation capabilities of a dense plasma focus fusion system." /library/stm-727be17c15.
• D. Miller, ZPE All Stars interview, Hard Truths Podcast (8 September 2026) — the narrowed statement of the vacuum-catalysed hypothesis, and the bench test implied by it.
• National Security Archive (2024), Castle BRAVO at 70: The Worst Nuclear Test in U.S. History — the yield, the estimate and the lithium-7 explanation. /library/stm-edfd668a59.