Electrical
The Marx Generator: Stacking Capacitors Into One Giant Pulse
The Marx Generator manufactures one enormous voltage pulse from a bank of ordinary capacitors by a beautifully simple switch of topology: charge them all in parallel from a modest supply, then reconnect them in series in a few billionths of a second so their voltages add. Ten capacitors charged to 100 kV become a single 1 MV impulse. Patented by Erwin Marx in 1924, it is still the standard way to make lightning-scale test voltages and the beating heart of the world's most powerful pulsed-power machines.
- InventorErwin Marx, 1924 (Braunschweig)
- Erected voltageV_out ≈ N · V
- Erection time~10–100 ns
- Voltage efficiency~85–95%
- Standard test pulse1.2/50 µs lightning impulse
- Air breakdown field≈ 30 kV/cm (3 kV/mm)
Interactive visualization
Press play, or step through manually. The visualization is yours to drive — try it before reading on.
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
Charge in parallel, fire in series
A Marx generator solves a hard problem: you want a pulse at, say, 1 MV, but capacitors and DC supplies rated for a full megavolt are enormous and costly. Marx's insight was to charge cheaply and discharge expensively. N identical capacitors are wired in parallel through high-value charging resistors and charged from a modest DC supply to a per-stage voltage V — a few tens up to roughly 200 kV. Between each stage sits a spark gap, an air or gas switch that stays open while the bank charges.
When the gaps close, those same capacitors are suddenly reconnected in series, and their voltages add:
V_out ≈ N · V
Ten stages at 100 kV become a 1 MV pulse; twenty stages, 2 MV. The charging resistors that set up the parallel path are deliberately large, so during the fast discharge they behave almost like open circuits and the current takes the low-impedance series route through the gaps. The generator has two personalities selected purely by the speed of the event: slow and parallel while charging, fast and series while firing.
The overvoltage cascade — why it fires itself
The dramatic moment is erection: the self-propagating collapse of every gap within nanoseconds. Only the first (bottom) gap needs a deliberate trigger, often a three-electrode trigatron. The causal chain is this — when gap 1 breaks down, the bottom plate of capacitor 2 is yanked from ground up toward +V. Its top plate is still held near its charged potential through a slow charging resistor, which cannot move charge on a nanosecond timescale, so the voltage across gap 2 is momentarily forced toward 2·V, well above its static breakdown level. Gap 2 flashes over; that hoists the next node toward 3V, overvolting gap 3, and so on up the column.
It is a positive-feedback avalanche — electrically a row of dominoes, where each firing raises the standing voltage on the gaps above it and guarantees they exceed breakdown. The whole stack erects in roughly 10–100 ns. Designers line the gaps up in sight of one another so ultraviolet light from each spark seeds electrons in the next, slashing the statistical time lag and jitter. The enabling condition is impedance separation, R_charge ≫ √(L/C), so almost no current leaks back through the charging resistors during the pulse.
The numbers: voltage, energy, current, power
Take a compact ten-stage generator: N = 10, stage capacitance C = 0.1 µF, charged to V = 100 kV. Erected output is ≈ 1 MV. Energy is where the "no free lunch" lives. Each stage stores ½·C·V² = ½ · 0.1 µF · (100 kV)² = 500 J, so the bank holds 10 × 500 J = 5 kJ. In series the stack looks like a single capacitor of C/N = 10 nF charged to 1 MV, storing ½ · 10 nF · (1 MV)² = 5 kJ — identical. Voltage is multiplied by N; the energy is merely rearranged.
Speed comes from the discharge loop. With a loop inductance of order L ≈ 1 µH, the surge impedance is √(L/C) = √(1 µH / 10 nF) ≈ 10 Ω, and the peak current into a low-impedance load is I ≈ V / √(L/C) ≈ 1 MV / 10 Ω = 100 kA. The instantaneous power V·I touches ~100 GW for a fraction of a microsecond — a torrent of power the little bank could never sustain, released in a single flash.
Shaping the impulse: the 1.2/50 µs standard
Most Marx generators in industry are not physics spectacles but referees: they apply the standardized 1.2/50 µs lightning impulse defined by IEC 60060-1 to prove that transformers, bushings, cables and switchgear survive a lightning-like overvoltage. The raw erected pulse would be far too abrupt, so two families of resistors shape it into a double exponential:
V(t) = V₀·(e^(−t/τ_tail) − e^(−t/τ_front))
A small series front resistor R_f, together with the load capacitance, sets the fast rising edge (front time ≈ 1.2 µs, the time to reach peak). A larger parallel tail resistor R_t discharges the bank slowly, fixing the time to fall to half value (≈ 50 µs). Switching-impulse tests use a slower 250/2500 µs shape. The real output never quite reaches N·V: stray capacitance to ground, loop inductance and the charging resistors cost a few to fifteen percent, so the voltage efficiency η = V_peak / (N·V) is typically 85–95%.
From test halls to the Z machine
Beyond the test hall, Marx generators are the front end of the world's biggest pulsed-power machines. Sandia National Laboratories' Z machine begins with dozens of Marx generators (originally 36) storing on the order of 20 MJ; after further pulse compression it drives roughly 20 MA through a wire array in about 100 ns, producing the intense X-ray bursts used for fusion and high-energy-density physics. Marx banks also power EMP simulators, high-power gas and excimer lasers, flash X-ray radiography, plasma sources, and the injectors of some particle accelerators.
The largest impulse-test generators reach several megavolts — the biggest exceed 6 MV and stand several stories tall. At the other extreme, modern solid-state Marx modules replace spark gaps with stacked IGBTs or MOSFETs, trading peak voltage for the ability to fire thousands of times per second with precise, low-jitter timing — the workhorse of medical linacs, water treatment, and repetitive pulsed-power research.
Conditions, limits, and staying alive
The design walks a tightrope. Charging resistors must be large enough to isolate the stages during the pulse, yet small enough that the RC charging time (seconds) is tolerable. Gap spacings are tuned so every gap self-fires reliably at the overvoltage but never prematurely from stray light or field enhancement. As N grows, stray capacitance to ground increasingly steals charge from the erection cascade, capping practical efficiency and stage count. Loop inductance limits how fast the front can rise, and every shot erodes the gap electrodes and fatigues the capacitor dielectric, so lifetime is finite.
Safety is not a footnote. A modest bench Marx already stores kilojoules at a million volts; the charged capacitors stay lethal after the supply is switched off, and a partial erection or a gap that fails to fire can leave unexpected charge trapped. Such machines carry grounding sticks and interlocks and are never approached without shorting each stage to ground. One misconception is worth killing: the Marx does not create or amplify energy, and it is not a Cockcroft–Walton multiplier. It only rearranges a fixed store of charge from parallel to series — trading current for voltage in one spectacular instant.
| Feature | Marx generator | Cockcroft–Walton multiplier |
|---|---|---|
| Output form | Single fast high-voltage pulse (µs-scale) | Continuous / DC high voltage |
| Voltage relation | V_out ≈ N·V (series stacking of charged caps) | V_out ≈ 2N·V_peak (diode charge pump) |
| Switching element | Spark gaps (or stacked solid-state switches) | Rectifier diodes |
| Peak power | Very high — GW-class, but only for a brief flash | Low — limited by per-stage current |
| Typical use | Impulse testing, pulsed power, EMP simulators | X-ray tubes, accelerators, old CRT supplies |
| Inventor / era | Erwin Marx, 1924 | Greinacher 1919 / Cockcroft–Walton 1932 |
Frequently asked questions
Does a Marx generator create or multiply energy?
No. It multiplies voltage by reconnecting the same charged capacitors from parallel to series; the total stored energy ½·N·C·V² is conserved (minus resistive and corona losses). Voltage goes up by N, but the series stack's capacitance drops to C/N, so the energy books balance.
Are all the spark gaps triggered individually?
Usually only the first gap is triggered, commonly with a trigatron. The rest self-fire in a chain: each gap that closes overvolts the next toward 2V, 3V, and so on, so the whole column erects by positive feedback in tens of nanoseconds.
How is a Marx generator different from a Cockcroft–Walton multiplier?
A Cockcroft–Walton stack uses diodes and capacitors to pump up a steady DC voltage at low current, good for X-ray tubes and accelerators. A Marx uses spark gaps to produce a single high-voltage pulse with enormous peak power (GW-class) but only for microseconds.
Why doesn't the output reach exactly N·V?
Stray capacitance to ground, discharge-loop inductance, the load, and current lost through the charging resistors all bleed off some voltage. The resulting voltage efficiency η = V_peak/(N·V) is typically 85–95%, and it falls as the number of stages rises.
Why use spark gaps instead of ordinary switches?
A gas spark gap is historically the cheapest, fastest closing switch that will hold off hundreds of kilovolts and then conduct hundreds of kiloamps. Modern solid-state Marx generators swap gaps for stacked IGBTs or MOSFETs when repetition rate and low jitter matter more than peak voltage.
What sets the shape of the output pulse?
Wave-shaping resistors and capacitances. A small series front resistor plus load capacitance sets the rise; a larger parallel tail resistor sets the slow decay. The standard lightning impulse for insulation testing is 1.2/50 µs per IEC 60060-1.