Safety

A small cyclotron concentrates most of the classic laboratory hazards into one apparatus: lethal voltages, kilowatt RF, unexpected X-rays, tonnes of magnetic force, an atmosphere of pressure on every lid, and flammable gas. None of these hazards is exotic, and all of them are managed daily in professional labs by procedures that cost little to copy. This page describes each hazard's mechanism — because a hazard you understand is one you can actually defend against — and its mitigation. The figure below maps where each hazard concentrates on a typical small installation; its numbered callouts link to the sections that follow.

Framing is deliberately conservative: where published sources disagree, the cautious bound is presented. This is information, not a substitute for training, local regulations, or professional review. Rules drawn from the literature carry their citations; the full sourced rule set is in the design guide.

Electrical: high voltage and stored energy

Mechanism. A cyclotron runs on multiple supplies that can kill: hundreds of volts for filaments and screens, kilovolts of dee bias and oscillator plate voltage, tens of kilovolts on ion-source and deflector electrodes. Currents of tens of milliamperes across the chest can stop a heart — the exact threshold moves with path, waveform and duration — and supplies throughout the machine can deliver far more. The subtler killer is stored energy, which remains after shutdown: filter capacitors hold their charge long after shutdown — hours or days if a bleeder is absent or failed, which is why discharge is proven with a meter, never assumed from elapsed time — and even coax cable is a capacitor — one documented build found that 6 m of HV cable at 30 kV stored ~0.4 J and sustained damaging arcs until the run was shortened (Rutgers 12-inch cyclotron, Cyclotrons 2010 presentation). Magnet coils store energy in their field instead: interrupting a coil circuit at operating current produces an inductive voltage spike that can arc through insulation or a person.

Mitigation. Treat every supply in the machine as lethal: interlock switches on all power-supply covers, and a grounding hook kept at the machine and applied to every capacitor and electrode before touching anything (Wouters, LASL recommendations for small cyclotrons). Discharge through a resistive stick first where stored energy is large, since a dead short across a charged capacitor is an arc in itself, and verify zero on a meter before hands go in. Fit bleeder resistors across HV capacitors, rated for the full voltage and sized to the capacitance rather than picked from a parts bin (the ~10 MΩ habit is fusor practice, at fusor scale). Still wait after shutdown, and let a meter reading prove the discharge: a failed bleeder discharges nothing (Farnsworth-fusor practice). Current-limit HV feeds with staged series resistance — one large resistor at the supply, a second at the chamber (Rutgers: 150 MΩ + 5 MΩ) — so a sustained fault delivers microamps, not amps; stored charge downstream of the resistors is not limited by them, which is why the hook and the bleeders remain the rule. Put meters in the grounded return leg, never the hot side. Cover or guard magnet coils whenever power exceeds 150 VA, current 30 A, voltage 130 V, or stored energy 5 J, and ground every core (Tanabe, magnet engineering lectures). Those are guarding criteria from magnet engineering rather than safe-touch limits; electrical-safety practice treats exposed conductors from about 50 V up as hazardous. Never open a coil circuit at current without a surge path across the coil (Wouters). Keep one hand in a pocket when probing live circuits, and never work on HV alone.

RF: exposure and burns

Mechanism. The dee system runs at tens of MHz and anywhere from tens of watts to kilowatts. RF at these frequencies does not trip the nerve reflex that DC and mains shock do — contact with an energized conductor cooks a deep, slow-healing burn before it hurts. RF also travels: without filtering, it rides out of the tank on every supply lead and control wire, making remote "safe" panels unexpectedly hot, and a high-power oscillator radiating into the room can exceed exposure limits (IEEE C95.1 / ICNIRP) near the tank. A home machine warrants the stricter general-public tier of those limits rather than the occupational tier: the exposed household is not a monitored workforce.

Mitigation. Enclose the oscillator in a grounded screened box, and choke and bypass every circuit that connects to a tank element so RF cannot reach meters and supply lines (Wouters). Never adjust a live RF system by reaching into it — kill it, ground it, adjust, re-energize. Interlock the tank lid and screen-box panels with the RF drive. Keep people out of arm's reach of unshielded resonant elements at power; arm's reach guards against contact burns, while compliance with the IEEE/ICNIRP exposure limits is something to measure on the shielded, assembled system rather than estimate by distance. And remember that "the RF is only 50 W" is how burns happen.

X-rays: yes, even from a sub-MeV machine

Mechanism. This is the hazard newcomers most reliably miss. Long before the ion beam does anything nuclear, the machine is an X-ray tube by accident. Any vacuum gap holding tens of kV emits electrons by field emission from microscopic surface whiskers (the Fowler–Nordheim mechanism; Miley & Murali, Inertial Electrostatic Confinement Fusion) — the "dark current" that flows with no ion source running at all. Those electrons slam into the dee, chamber wall, or viewport at full gap voltage and produce bremsstrahlung X-rays. Amateur fusor experience puts the practically detectable onset around 18–20 kV on the electrodes (Kovalchick, IEC fusor thesis); conservatively, treat any vacuum gap above ~15 kV as an operating X-ray source. Dee-gap RF voltage counts, and what counts is the full instantaneous gap potential: a 30 kV peak swing on a single dee against a grounded chamber makes 30 keV X-rays, and two dees driven push-pull double the gap voltage — and the endpoint — for the same per-dee number. Glass viewports attenuate these photons far less than the metal around them — the weak points in the wall, and dark current — hence X-ray output — changes over time: it drops as electrodes condition and jumps after every air exposure (Miley & Murali).

Mitigation. Own a survey meter before first pump-down: a GM counter to find radiation and a calibrated ion-chamber or energy-compensated instrument to quantify dose rate. Check both instruments' low-energy response: thick-walled GM tubes and many energy-compensated meters badly under-read 10–30 keV photons, so the finder wants a thin-window (pancake) probe and the dose meter wants a calibration that covers the low-energy end. Survey the whole machine perimeter at first power-up, at every voltage increase, and after every vent-and-pump cycle, with special attention to viewports and thin walls. Shield with lead sheet where the survey says so — millimeters of lead stop sub-100 keV X-rays effectively — and adopt a zero-measurable-exposure goal at the operator position rather than working up to a dose limit. NCRP Report 144 (Radiation Protection for Particle Accelerator Facilities, which superseded NCRP 51) is the standard shielding reference and worth reading before designing any enclosure. For sizing and building that enclosure — which radiation a sub-MeV machine can physically produce, measuring the spectrum before choosing a material, and why penetrations rather than walls decide the result — see shielding a small cyclotron. The design guide’s safety rules collect the sourced accelerator-safety practice, X-ray precautions included. The figure below is the intuition version of the two levers that matter: distance, which is pure geometry for a point source, and shielding, which works in powers of two per half-value layer.

Induced radioactivity: where the line actually is

Mechanism. A proton below a reaction's threshold cannot drive that reaction. For the (p,n) reactions that dominate activation of common structural materials, thresholds sit comfortably above 1 MeV; the exceptions, exothermic capture channels with no threshold at all, are scoped below (values from the NNDC Q-value calculator, rounded):

• ⁵⁵Mn(p,n)⁵⁵Fe — 1.03 MeV (manganese is in practically every steel; ⁵⁵Fe is a 2.7-year EC emitter — IAEA NDS, retrieved September 2026)
• ⁵³Cr(p,n)⁵³Mn — 1.41 MeV (⁵³Mn at 3.7 million years is effectively stable: the channel opens with negligible activity)
• ⁵⁷Fe(p,n)⁵⁷Co — 1.65 MeV
• ⁷Li(p,n)⁷Be — 1.88 MeV
• ⁹Be(p,n)⁹B — 2.06 MeV (⁹B is particle-unbound, so this row marks neutron production rather than a lingering radionuclide)
• ⁶⁵Cu(p,n)⁶⁵Zn — 2.17 MeV
• ¹³C(p,n)¹³N — 3.24 MeV
• ⁶³Cu(p,n)⁶³Zn — 4.21 MeV
• ⁵⁶Fe(p,n)⁵⁶Co — 5.44 MeV
• ²⁷Al(p,n)²⁷Si — 5.80 MeV

A sub-MeV proton machine — which is what a first amateur cyclotron is — therefore produces essentially no induced activity in its structural materials — a statement about practical yield rather than an absolute, since thresholdless capture on structural metals is energetically open at any energy: the beam is below every threshold above, and the Coulomb barrier suppresses what little else is energetically allowed. Two scoping notes keep that statement honest. The thresholds above are properties of the listed nuclides, not of matter — mid-Z targets can be energetically open far lower (¹¹⁵In(p,n): 287 keV, Coulomb-suppressed but open), so the material inventory in the beam path defines the real boundary (see experiments by energy). And thresholdless channels — ¹⁰B(p,α)⁷Be, ¹²C(p,γ)¹³N, ¹⁶O(p,γ)¹⁷F — leave radionuclides in bombarded targets at any energy, sometimes well beyond trace level: ⁷Be from boron accumulates measurably under sustained beam. Survey targets before handling them. That comfort erodes in three stages. Above ~1 MeV the first structural channel opens: ⁵⁵Mn(p,n)⁵⁵Fe at 1.03 MeV — and manganese is an ingredient of practically every steel, at the few-tenths-percent level of the census's own magnet-steel analyses — with chromium behind it (⁵³Cr, 1.41 MeV, though its product ⁵³Mn is so long-lived its activity is negligible) and iron's minor isotope at 1.65 MeV (all NNDC Q-value calculator, retrieved September 2026); near 2 MeV lithium and beryllium contaminants begin producing neutrons and copper's first channel opens (⁶⁵Cu, 2.17 MeV), while ¹³C in graphite waits until 3.24 MeV; the major channels (⁶³Cu at 4.21, ⁵⁶Fe at 5.44 MeV) open by ~4–5.5 MeV and the machine activates in earnest; and deuterons are a different animal entirely: many (d,n) reactions are exothermic, and a deuteron beam implants deuterium into whatever it strikes, turning the target into a D–D neutron source at energies where protons are harmless. Fusor practice treats sustained D–D output above roughly 6×10⁵ neutrons/s as the point where shielding, not just time and distance, becomes necessary (Hull, IEC fusor documentation). That is a community rule of thumb rather than a dose criterion; measurable dose rates exist at meter-scale distances well below it, and the jurisdiction's limits are what decide. Measure with the right instrument: the GM and ion-chamber meters this page recommends for X-rays do not read neutron dose. Deuteron operation, or any running above the (p,n) thresholds, needs calibrated neutron dose instrumentation or dosimetry alongside the photon instruments.

Mitigation. Accelerate protons (or H₂⁺), not deuterons, on a first machine. Below ~1 MeV, proton activation of the ordinary copper, steel and aluminium the machine is built from is closed on the tabulated (p,n) channels — the first, manganese in steel, opens at 1.03 MeV; target-specific reactions and the capture channels above still need their own check, target by target. If the design grows past that, face beam-strike surfaces with graphite, which activates far less than copper (Oak Ridge 86-inch practice), survey targets and slits after runs before handling, and keep a run log. Any machine energetic enough to activate materials is energetic enough to need real shielding design — NCRP 144 again — and, in most jurisdictions, registration (see Legal).

Vacuum: implosion and stored atmospheric force

Mechanism. Atmosphere pushes on every evacuated surface with 101 kPa — about 10 N/cm² (~1 kgf/cm²). A modest 30 cm-diameter chamber lid carries roughly 7 kN, the weight of a small car, continuously. Metal that yields dents — though a thin flat lid can buckle suddenly and tear feedthroughs and windows with it, which is why the lid calculation below exists; glass fails by implosion, collapsing inward and then spraying shrapnel outward, and scratched or stressed glass can let go without warning.

Mitigation. Calculate every flat lid before pumping down — deflection, stress, and safety factor; the lid-deflection calculator exists for exactly this. Prefer metal chambers. Guard every glass element (bell jars, large viewports) with a polycarbonate shield or steel mesh, never rely on the glass itself, and retire any glass component with a visible scratch or chip. The sourced vacuum and pressure-safety practice is collected in the design guide’s safety rules.

Pump exhaust: oil mist in the room

Mechanism. An oil-sealed rotary-vane pump's exhaust leaves through its oil case and carries oil with it as a fine aerosol. The hobbyist write-up behind a documented amateur exhaust filter (library entry) puts it plainly: run such a pump for any length of time and “you will get a fine (submicron) oil mist out of the exhaust port”, a mist that “will look like smoke and is very hazardous to breathe”. The stock exhaust cap, the same write-up adds, “does nothing to capture the oil mist”. Mineral-oil mist carries published occupational exposure limits: NIOSH recommends no more than 5 mg/m³ over a working day (10 mg/m³ short-term), OSHA’s permissible limit is the same 5 mg/m³, and the listed effects are irritation of the eyes, skin, and respiratory system (NIOSH Pocket Guide, retrieved September 2026).

Two things set this hazard apart from every other on this page. It scales with hours, not beam: the pump runs longest exactly when the builder is beside it — roughing cycles, leak hunting, bakeouts — usually in the enclosed basement or garage the machine lives in. And it arrives first: a build’s vacuum system runs for months or years before the RF is finished, which is before any other hazard on this page exists at all. The census is full of machines that pumped down long before — or instead of — ever making beam. The stored-force hazard above arrives just as early; what is different here is accumulation: exposure grows with every pump-hour, whether or not the machine is ever finished.

Mitigation. Fit an exhaust mist filter: coalescing filters are a standard accessory for every common pump, and the documented homemade one — a wound water-filter cartridge and furnace-filter fabric in PVC fittings, about $45 — reports complete visible mist capture over 18 months of service, with the captured oil draining back into the pump. That is one builder's observation rather than a measured filtration efficiency, so ventilate anyway, and watch any filter's backpressure: a clogging exhaust filter loads the pump (library entry, with the original write-up linked there). Better still, duct the exhaust outdoors, terminating away from windows and air intakes and arranged so condensed oil drains back rather than pooling. That removes the mist and, with it, whatever the pump is pumping, including the ion-source hydrogen. Ventilate the pump space either way, and treat visible “smoke” at an exhaust port as the warning it is.

Magnet: projectiles and crush points

Mechanism. The pole-face attraction of a cyclotron magnet is F ≈ B²A/2μ₀ — in workshop units, (kilogauss)² × (area in in²) / 1.735 pounds (Wouters). Eight-inch poles at 10 kG (1 T) attract each other with ~13 kN — about 1.3 tonnes-force. The same field turns loose steel into projectiles: a wrench snatched from a hand accelerates through the gap with injuring force, and fingers between a tool and a pole, or between poles during assembly, are crush casualties. Permanent-magnet machines add a trap: they are never off — one built PM cyclotron magnet measured up to ~560 gauss of residual field at its zero setting — low enough that its builders called disassembly easy, and two orders of magnitude above the conventional 0.5 mT (5 gauss) pacemaker exclusion line of magnet-safety practice (Antokhin et al., RuPAC 2006, for the measurement) — and even electromagnets keep remanent field after the supply is killed.

Mitigation. Design the assembly sequence around the forces: jack screws or fixtures that control pole approach, never hands in the gap, and rigging rated for the attraction load, not just the weight. Keep ferromagnetic tools and stock away from an energized magnet; use non-magnetic tools for gap work. Post the field hazard and keep anyone with a pacemaker or ferromagnetic implant away. Cover coils per the electrical thresholds above, and treat a PM machine as permanently energized — because it is (the permanent-magnet section of the magnet page carries the documented assembly fixtures and force-reduction methods).

Gas handling: hydrogen

Mechanism. The ion source feeds on hydrogen, flammable in air from 4% up to roughly 75% by volume (supplier SDS: lower limit 4%, upper 76% — Airgas hydrogen SDS, retrieved September 2026) — an exceptionally wide range — with an ignition energy low enough that the SDS's handling precautions run to anti-static clothing. The quantities a cyclotron uses are tiny, but the failure mode is a leaking cylinder or regulator slowly filling an enclosed room. A second, sneakier failure: needle/metering valves are not shut-off valves — the manufacturer states plainly that they do not seal positively (Parker metering-valve datasheet) — so a "closed" metering valve left as the only barrier will bleed the cylinder into the room or the chamber.

Mitigation. Use the smallest cylinder that does the job (a lecture bottle lasts a long time at ion-source flow rates). Put a bubble-tight shut-off valve in series with the metering valve and close it whenever the machine is idle. Store and use the cylinder upright, secured, in a ventilated space, away from the HV and RF sparks the rest of the machine produces; leak-test every joint at pressure. Inexpensive hydrogen sensors exist and are worth mounting above the gas panel. And remember where the gas goes: the hydrogen the source uses leaves through the vacuum pump’s exhaust, so the exhaust line is part of the gas path. Duct it outdoors or ventilate where it discharges.

The three rules

Everything above compresses to three habits that professional accelerator culture treats as identity, not preference:

• Get a calibrated survey meter before you need one. Radiation is the one hazard here with no smell, sound, or sensation. If the project cannot afford a calibrated instrument, it cannot afford to run.
• Know your jurisdiction's registration rules before the machine can make X-rays, which is before first beam; some jurisdictions attach the obligation to installation or possession rather than operation. Many US states require registration of particle accelerators — sometimes at surprisingly low energies. The Legal section surveys all 51 US jurisdictions.
• Never defeat an interlock. Not once, not briefly, not because the run is almost done. Every interlock bypassed "temporarily" is a bet that the one person near the machine at the wrong moment is not you.

Sources

• L. F. Wouters, General Recommendations for the Design of Small Cyclotrons (LASL) — interlocks, grounding hooks, RF choking/bypassing, coil surge protection, pole-force formula.
• J. Tanabe, magnet engineering lectures (SLAC/USPAS) — coil covering thresholds (150 VA / 30 A / 130 V / 5 J), core grounding, cooling interlocks.
• Rutgers 12-inch cyclotron, Cyclotrons 2010 conference presentation — staged HV series resistance, coax stored-energy arc case, HV edge-field rules.
• G. H. Miley & S. K. Murali, Inertial Electrostatic Confinement (IEC) Fusion, Springer — field emission / Fowler–Nordheim prebreakdown, conditioning behavior.
• C. Kovalchick, IEC fusor thesis — observed X-ray onset at ~18 kV electrode voltage; zero-exposure monitoring practice. R. Hull, fusor documentation — D–D neutron shielding threshold ~6×10⁵ n/s.
• ORNL 86-inch cyclotron report — graphite facing of beam-strike surfaces; coil thermal alarms.
• NCRP Report No. 144, Radiation Protection for Particle Accelerator Facilities (2003; supersedes NCRP 51, 1977).
• NCRP Report No. 49, Structural Shielding Design and Evaluation for Medical Use of X Rays and Gamma Rays of Energies Up to 10 MeV (1976) — broad-beam half-value-layer ballparks used in the shielding-intuition figure.
• IEEE C95.1 / ICNIRP guidelines — RF exposure limits.
• NNDC (Brookhaven) Q-value calculator — (p,n) reaction thresholds, as retrieved 2026.
• Parker metering-valve datasheet — metering valves are not positive shut-off devices.
• NIOSH Pocket Guide to Chemical Hazards, “Oil mist (mineral)” — REL TWA 5 mg/m³ / ST 10 mg/m³, OSHA PEL TWA 5 mg/m³, health-effect listing; retrieved September 2026.
• Exhaust Filter for Harbor Freight Oil Vacuum Pump (hobbyist write-up, library) — the submicron-mist observation, the stock cap's inadequacy, and the $45 filter with an 18-month service report.

Verified against the cited sources as of 2026. Err on the side of the more conservative figure wherever this page and another source disagree — and if that other source is this site's own design guide, read the cited original.