Ion Sources for Small Cyclotrons
The ion source is the cyclotron subsystem where the open literature comes closest to handing a builder its numbers directly: filament sizes, arc voltages, gas flows, slit dimensions, and service intervals, measured on machines from 6-inch teaching devices to commercial PET cyclotrons. This page collects that record for the internal sources a small machine actually uses — the hot filament, the hooded arc, and the Penning (PIG) source — and names the four things that move beam current: slit geometry, gas flow, arc regulation, and position relative to the first gap. It is the published record a design can be checked against, not a fabrication recipe; the examples run from amateur and teaching machines through the AEC-era workhorses to commercial medical sources, each cited at its own scale. The design guide’s ion-source domain holds the underlying 114 sourced rules.
What the source must deliver
The requirements are location, timing, and a modest current: ions at the machine center, close to the median plane, inside a narrow slice of the RF cycle. A cyclotron accepts nothing else. With an internal source and sinusoidal dee voltage, the usable starting-phase window is roughly −90° to +20° relative to the voltage crest — about 30% of the cycle (Smirnov 2021, p. 56). Everything the source emits outside that window, or above or below the aperture, is lost in the first turns. The center is also the machine's blind spot for focusing: the field index is zero at r = 0 by symmetry, so nothing magnetic holds the first turns together vertically — "by a fortunate coincidence, electrostatic focusing by the accelerating fields is effective for low-energy ions" (Humphries 1986, p. 524). Source geometry and RF phase do the work the magnet cannot.
The current demanded of the source is small. A working internal arc makes microamperes-to-milliamperes of ions available at its exit slit, while a first amateur machine accelerates nanoamperes; published slit currents run three to five orders of magnitude above what a small machine captures (Forringer 2004, pp. 88–91). That surplus gives a clean debugging rule: "one should be suspicious of the ion source if the measured beam current is very low in the region close to the ion source, i.e. the regime of large turn spacing" (Koeth 2006, p. 1). Low current close in makes the source the first suspect; current that appears close in and dies farther out points first at RF, vacuum, or field.
An internal source sits at the machine center inside the main vacuum; an external source feeds ions in through the pole via axial injection. External sources buy species flexibility and a cleaner chamber at the cost of an injection line and an inflector; none of the amateur machines documented in the builds directory uses one. This page covers internal sources; how the finished beam leaves the machine is a separate subject (beam extraction), and plasma theory beyond what a builder needs lives in Wolf's Handbook of Ion Sources.
Filament sources: a hot wire in the gap field
The simplest working source is a bare tungsten filament at the chamber center, run hot enough for thermionic emission and biased a modest negative voltage so its electrons ionize hydrogen along the field lines; the dee field pulls ions straight from this column. Lawrence's first cyclotrons ran exactly this open arc, the magnetic field itself collimating the electron stream (Clark 1981, p. 232). UCRL-476, the one report written for someone about to build a small machine, gives the whole recipe: a 0.64 mm (0.025 in) tungsten filament drawing about 25 A DC — "an automobile headlight filament has been used, but the breakage has been high" — with the arc running 0.5 to 2 A at 100–500 V once chamber pressure reaches the 10−4 mm Hg range (UCRL-476, 1949, pp. 7–8).
Two failure modes dominate filament practice, and both have standard fixes. First, vibration: a filament carrying mains-frequency AC in a tesla-class field shakes itself apart, so heating is DC or radio-frequency AC. The Iowa State 1.5 MeV machine heated its spiral filament at high frequency "to minimize self-destructive magnetic effects" (McGuire 1961, p. 7); the Argonne 60-inch used a 220 kc transmitter, the Carnegie synchrocyclotron 20 kc, and UCRL-476 floated its DC supply across a storage battery purely to filter ripple (ANL-5907, p. 10; NYO-780, p. 22; UCRL-476, p. 8). Second, temperature sensitivity: emission follows the Richardson law, so a 10% change in filament temperature swings emission roughly tenfold, and evaporation life collapses just as fast in the other direction — a 1 mm tungsten wire lasts ~8,300 h at 2500 K but ~46 h at 2900 K (Wolf 1995, pp. 27–31). Regulate filament current tightly and run the fattest filament the supply affords at the lowest temperature that gives enough arc.
The Houghton College cyclotron shows the architecture at teaching scale: a standard AEI electron-microscope hairpin filament floating at about −90 V, heated by 2 A, with each filament lead bypassed to ground through a 0.001 µF capacitor against RF pickup (Yuly 2013, p. 3). The same program measured the bias knob in its own setup: "it appears that filament bias has no effect on the beam current" (Fuller 2013, pp. 48–49) — there, pressure and dee voltage moved the beam and bias did not, a useful caution against tuning the wrong knob first.
The chimney turns an open arc into an instrument
The open filament's weakness is that its plasma column is everywhere: it loads the RF system, sprays gas into the chamber, and emits ions from an undefined region. The fix, standard by the 1940s, is a hood — "later a hood was placed over this arc to confine the region of high gas pressure and reduce loading on the dee system. A reflector was added above the filament to use the arc electrons more than once" (Clark 1981, p. 232). The hooded arc's tube is the chimney: gas and arc burn inside it, and ions leave only through a small aperture facing the dee gap.
The Rutgers 12-inch documented the conversion directly: a chimney over the existing filament lets thermionic electrons run the full column to the median plane, and "a small aperture, 1/16 of an inch in diameter, opening towards the DEE permits ions to be drawn into the accelerating field" (Koeth 2006, pp. 2–3). The Cyclotron Kids' 2 MeV machine used a quarter-inch tube the same way, shielding the filament from the dee's RF field (Baumgartner 2013, p. 3). How much column to expose is itself a sourced number: the optimum ionization-column length is 0.35–0.4 of the internal dee aperture — 16 mm (5/8 in) for MIT's 1.6-inch aperture — because a longer exposed column loads the RF circuit with off-focus ions and drags down dee voltage (Livingston & Blewett 1962, p. 178).
Mature hooded-arc practice is best read from the machines that ran for decades. The Argonne 60-inch: graphite chimney of 7.9 mm (5/16 in) diameter with a 2.4 × 5.6 mm aperture, 3.2 mm tantalum hairpin, deuteron arc at 50–120 V and 6–8 A, gas at 0.1 cm³/min (ANL-5907, p. 10). The Oak Ridge 86-inch: a graphite arc chamber 14.3 mm OD × 272 mm long over a 4.3 mm tantalum filament drawing 300–450 A, arc 100–300 V at 0.5–1.5 A, protons escaping through a lateral slit "usually 0.062" by 2.5"" (ORNL-1196, p. 64). Graphite recurs for a reason: it "operates at high temperatures with a minimum of sputtering or evaporation" (Livingston & Blewett 1962, p. 178).
The PIG source: two cathodes and a hollow anode
A Penning or PIG source is a chimney source with cathodes closing both ends and the chimney itself serving as the anode: the magnetic field traps electrons oscillating between the cathode faces until they ionize something. Each trapped electron produces about eight ions on average, which is why the geometry is so efficient (Wolf 1995, p. 71). The discharge itself needs only a modest axial field — above roughly 0.1 T the arc parameters barely depend on B (Wolf 1995, p. 71) — and one source family has run in cyclotron center fields from 0.5 T on a test stand to 4.5 T in the Harper K100 (Forringer 2004, p. 27). Figure 1 shows the anatomy.
PIG sources run in two regimes, distinguished by how the cathodes emit. Cold-cathode sources sustain the arc by secondary emission under ion bombardment; hot-cathode sources let the cathodes (or a filament) run thermionic. Wolf's nominal envelopes — above 1 kV at 0.5–5 A for cold, under 1 kV at 1–50 A for hot (Wolf 1995, p. 71) — are typical rather than defining; measured cold-cathode sources straddle them (the Eclipse sustains 550 V). The AMIT bench measurements caught a single source crossing between them: below ~250 mA of arc the cathodes emit by secondary emission and the discharge impedance is high and positive; above, ion bombardment self-heats them into thermionic emission and the V–I slope turns negative (Obradors et al. 2017, p. 3). A "cold" cathode is not necessarily cool — thermal modeling of a 500 W source puts the cathode heads at 1992 K, where computed thermionic emission is still only 0.6% of the discharge current (Zakerhosseini et al. 2016, pp. 1–3).
For the cold-cathode PIG, the electrical architecture is one floating supply. The plasma sits within a few volts of anode potential and nearly the full arc voltage drops across the thin cathode sheath (Wolf 1995, p. 71), so the chimney can be grounded to the chamber with both cathodes strapped together on one negative feed — "for internal sources, the anode is usually grounded and 30–100 kV of rf voltage is used for extraction with a feeler or puller extending from the dee" (Clark 1981, p. 233). Hot-cathode and indirectly heated variants add heater or bombardment supplies on top. Ion energy at the cathode equals the arc voltage, which is why cold-cathode sources sputter faster than hot ones.
Section along the field axis
Plan at the median plane
Materials follow the temperatures. Cathodes: titanium is the accepted cold-cathode compromise, tantalum where the cathodes run hot (Wolf 1995, pp. 73–74) — medical-cyclotron sources use Ta discs. Chimney: copper, molybdenum, or graphite; grooved molybdenum anodes measurably outperformed the Eclipse baseline (Potkins et al. 2017, pp. 3–6), and the classical machines ran graphite arc chambers for decades (ORNL-1196, p. 64). Insulators: alumina to 1400 °C is the workhorse; boron nitride is excellent to 1200 °C but absorbs water and "outgasses badly," so bake it and bring the arc up slowly after every air exposure (Wolf 1995, pp. 344–345). Watch alloying traps — stainless steel forms low-melting alloys with tantalum and molybdenum above 900 °C (Wolf 1995, p. 344). And cool deliberately: an NSCL-lineage source melted copper parts when run without water (Forringer 2004, pp. 19–20), and the 86-inch put two dedicated water circuits on its source (ORNL-1196, p. 66).
Slit geometry is the strongest knob on output
Extracted current is proportional to arc current — roughly 10–100 (mA/cm²) per ampere of arc through the anode slit (Wolf 1995, pp. 71–72) — but the slit decides how much of that the machine can use, and in what shape. The measured record is unusually complete because Forringer characterized one source family on a DC test stand before installation:
- Slit beats hole. A tall slit gives a flat plasma boundary and a converging beam; a round hole of equal class gives a concave boundary, a diverging beam, ~50% larger normalized radial emittance, and half the luminosity at equal arc current (Forringer 2004, pp. 73–76). The Eclipse cross-check: converting anode and puller slits to equal-area round holes raised machine transmission from 19% to 30% but cut target current from 120 to 40 µA (Potkins et al. 2017, pp. 3–4). The trade runs on axial acceptance: a machine starved of it may prefer the hole's transmission, while the tall-beam weak-focusing geometry favors the slit's current.
- Width trades current against emittance, favorably. Doubling the slit from 0.25 to 0.51 mm (both 5.0 mm tall) raised beam current about 4.4× — 52 to 230 µA at 50 mA of arc — for ~1.7× the radial emittance (Forringer 2004, pp. 66–69).
- The machining details are published. Slits chamfered 10°, relieved to a 0.25 mm (0.010 in) land in a 0.5 mm wall, beam emerging about 70% of slit height (Forringer 2004, pp. 77–97); Russian design practice makes the slit's axial size several times its radial size with chamfer angles of 40–60° (Smirnov 2021, pp. 52–53).
- Bore has an optimum. For a hydrogen PIG, 7 mm anode bore maximized simulated electron density and 8 mm gave the highest measured beam in the KIRAMS-13; above ~9 mm secondary-electron production falls (Mu et al. 2015, pp. 3–5).
Alignment tolerances split into two classes. Concentricity is machined: an off-center cathode sharply cuts secondary-electron production and confinement lifetime (Mu et al. 2015, p. 4). Aim is adjusted: in the same group's extraction modeling, better than 50% extraction held only over a slit-to-puller rotation window of −0.2° to −1.5° — a window that narrow is specific to that geometry, but the degree scale of the tolerance is the message — and the 86-inch crew gave their source stem a Selsyn drive with three degrees of geared rotation because "the alignment of the ion source with the magnetic field and with the accelerating slits is critical" (ORNL-1196, pp. 64, 83–84). Build the chimney concentric in one setup; make the stalk's rotation finely adjustable and lockable from outside the vacuum.
Gas feed couples straight into the vacuum budget
In a tight, well-pumped system the running source's hydrogen is the chamber's dominant gas load; in a new or leaky machine, leaks and outgassing can still bury it, which is itself diagnostic. MIT measured the canonical ratio: base pressure below 1×10−6 mm Hg, operating pressure about 2×10−5 with the source gas flowing (Livingston & Blewett 1962, p. 198). Modern numbers agree: 2.5 sccm of H₂ held a well-pumped test chamber near 4×10−5 torr (Forringer 2004, pp. 17–29). An order-of-magnitude pressure rise when gas flows is normal operation, and the resulting scattering loss is calculable — the vacuum calculator turns chamber pressure and path length into a surviving beam fraction.
The chimney is what makes this bargain affordable: it holds the discharge interior at 1–10 Pa — a thousandfold pressure step above the chamber, sustained through a millimeter slit — and a discharge with good confinement ionizes up to half the gas fed to it, where poorly confined sources manage 10–20% (Wolf 1995, pp. 46, 58, 99; Table 5.5 on p. 90). Published consumption spans a wide band: 0.1 cm³/min on the Argonne hooded arc and 2–3 on the 86-inch (ANL-5907, p. 10; ORNL-1196, p. 62), 0.2–0.6 sccm across Wolf's Penning-source table (p. 90), and 3–5.5 sccm on the measured PET-class PIGs (Forringer 2004; Potkins et al. 2017). Feed the gas in at the cathode end of the chimney, away from the slit (Wolf 1995, p. 70; the AMIT source pipes it through the cathode cavity itself).
Admission hardware has been stable since 1949: "a needle valve having a long tapering needle," a deliberate thread leak, or a heated palladium diffuser (UCRL-476, p. 7). A modern fine-taper metering valve with a vernier handle makes flow settings repeatable — log turns-open as the process setpoint, valid while upstream pressure and gas stay fixed — but it is not a shutoff; the manufacturer states these valves are "not recommended for positive shut-off," so put a separate isolation valve in series (Parker Series 20/30 datasheet). Everything upstream of that valve is a hydrogen handling problem, covered in the safety page's hydrogen section.
There is also a pressure window on the accelerator side. The Houghton machine mapped it: beam appears between roughly 10−6 and 10−4 torr of hydrogen — too little gas starves ionization, too much scatters the beam and broadens the resonance (Yuly 2013, p. 5; Fuller 2013, pp. 51–52). Their largest single beam-current gain, an order of magnitude, came from running lower hydrogen pressure and lower filament bias, at zero added RF power (Yuly et al. 2010, pp. 17–18).
Striking, regulating, and quenching the arc
A gas discharge is not a resistor. Its V–I characteristic ladders through dark current, breakdown, normal glow (voltage nearly constant over decades of current), abnormal glow, and finally glow-to-arc transition when the cathode overheats — with hysteresis, so a lit discharge persists below its striking condition (Miley & Murali 2014, pp. 89–91). Two consequences are non-negotiable: the supply must be current-limited (ballast resistor or electronic current regulation), or the negative-slope regions run away; and the operating map must record voltage at the electrodes, since the ballast drops a large share of the supply setpoint during operation (Rovey et al. 2007, p. 2).
The cold-start recipe is the same across four decades of sources: "an arc is struck by raising the arc voltage to about 3 kV and increasing the gas pressure. Background ionization starts the discharge... [it] is stabilized by the arc supply current regulator or ballast series resistor" (Clark 1981, p. 233). The AMIT bench procedure matches: up to −3 kV on the cathodes and up to 10 sccm of gas, strike within seconds, sustaining voltage under 1 kV, then throttle the gas back down and regulate on current (Obradors et al. 2017). A bench PIG with a ballasted supply ignites at 1 kV or less (Rovey et al. 2007, pp. 2–3). Spec the arc supply's compliance for the ignition transient — 3–5 kV in the published recipes (Wolf 1995, p. 90) — with insulation and stored-energy margins to match, even though the running point is a few hundred volts. An arc supply at these voltages is lethal; the electrical practice belongs to the safety page's high-voltage section.
Running, the arc telegraphs its health through voltage. At constant current, arc voltage rises as gas flow falls — the discharge compensating for missing particles — "until the discharge becomes unstable" and quenches (Wolf 1995, pp. 71–72). Creeping arc voltage at fixed current therefore means gas starvation or a worn cathode, long before dropout. Arc power versus gas flow passes through a minimum (near 4 sccm on the AMIT source), a usable tuning landmark (Obradors et al. 2017, p. 3). The 86-inch ran the mature control scheme: arc voltage and arc current regulated as two independent loops, voltage by the arc supply, current by trimming filament temperature (ORNL-1196, p. 66). Below the thermionic transition (~250 mA in the AMIT geometry) the cold-cathode discharge showed high positive impedance, and a stiff current source plus modest ballast held it stable in those tests (Obradors et al. 2017, p. 3) — a result to verify on the actual circuit, since discharges also exhibit ignition transients, oscillations, and gas-flow feedback (Miley & Murali 2014, pp. 89–91).
Position: the source lives or dies in the first half-turn
Internal-source "extraction" is the dee's job. The RF field at the puller pulls ions from the slit; big machines do it with 30–100 kV of RF (Clark 1981, p. 233), and amateur machines with a few kV must recover field by closing the slit-to-puller gap. The scaling is Child–Langmuir: available current density goes as U3/2/d2,
— for protons at 10 kV across 5 mm (0.5 cm), about 220 mA/cm², thousands of times a small machine's needs, which is why weak extraction is a geometry problem rather than a space-charge one. The gap cannot shrink without limit: vacuum breakdown sets the floor, and tested hardware holds about 10 kV/mm — a 12.7 mm-radius puller held 50 kV across a 5.0 mm gap on the NSCL test stand, the K100 runs a 2.9 mm minimum gap (Forringer 2004, pp. 68–85), and the Eclipse puts its puller slot 2.3 mm from the anode (Potkins et al. 2017). Wolf's empirical breakdown limit,
is a bound for clean flat DC electrodes, and a source gap violates every one of its assumptions in service — RF voltage, a neutral gas jet from the slit, sputtered films — so it explains why the published gaps look generous; it is never a license to close them to its edge. Good extraction optics keep the aspect ratio near S = r/d ≈ 0.5 (a round-aperture rule, r the aperture radius; slit systems use the analogous half-width) and then match the plasma density to the field by tuning the arc (Wolf 1995, pp. 368–369).
Two placement rules close the loop. The first half-turn must clear the chimney body: the Rutgers study computed that with their 12.7 mm dee gap, first ions clear the chimney at roughly 200 W of RF drive, 50 W being far too low (Koeth 2006, p. 5) — a chimney is viable only when the dee voltage makes the first-turn diameter larger than the chimney. And the optimum source position is empirical: transmission-line voltage droop along the dee faces (measured up to 5%) shifts orbit centers, and large machines have needed source displacements of more than two inches to compensate (Livingston & Blewett 1962, p. 164). The 86-inch tuned beam intensity partly "by tilting the ion source" and, after commissioning, moved the whole source 38 mm to sit 25 mm below magnetic center (ORNL-1196, pp. 21, 108). Build the mount adjustable in position and rotation, and tune for beam, not for geometric center.
Species: H⁺, H₂⁺, and H⁻
For a small machine the species decision is short: accelerate protons, manage the H₂⁺ fraction, and leave H⁻ to machines with extraction programs. Commercial PET cyclotrons accelerate H⁻ because a negative-ion beam can leave the machine through a stripping foil instead of a deflector — a beam-extraction subject, treated there. The source side of the choice is what matters here: H⁻ is a minority species, made by "source plasma conditions which promote the attachment of electron to an atom, or upon charge exchange at surfaces or in vapor" (Clark 1981, p. 232), and it is fragile — "an electron detachment of negative ions due to a collision with plasma electrons frequently takes place because the absolute value of electron affinity is as low as about 1 eV," which is why high-current negative-ion extraction from an ordinary plasma is considered "extremely difficult" (Wolf 1995, pp. 293–294). The same PIG geometry produces it — the Eclipse extracts 800 µA of H⁻ from a 150 W arc (Potkins et al. 2017) — but the H⁻-specific optimizations (a cool ring between plasma and wall, cesium in the cathode) buy nothing for positive ions. Internal-source H⁻ at scale is real but hard-won: Milan's internal source delivered 45 µA of H⁻ at 45 MeV in 1981, against milliamperes of protons from the same source family (Clark 1981, pp. 234–235).
An amateur machine with a probe target and no extraction system gains nothing from H⁻ and inherits all its fragility; the practical species question is instead H⁺ versus H₂⁺. A hydrogen discharge makes both, and they resonate at different frequencies — an H₂⁺ ion behaves like two protons sharing the energy (Oliphant & Rutherford 1933, p. 269). Oliphant and Rutherford saw it in 1933: a fresh discharge "consists very largely of molecular ions, but after running for some time it changes over and becomes nearly all protons" (Oliphant & Rutherford 1933, pp. 261–262). The modern control is gas flow: a cold-cathode PIG run current-regulated at ≥2 sccm showed no detectable H₂⁺, while starving it to 0.5 sccm flipped the arc mode and molecular ions appeared (Forringer 2004, pp. 79–80). Healthy flow keeps the molecular fraction down in that source; the check that settles the species question on any machine is which frequency the beam actually resonates at: f = qB/2πm is a magnetic analysis built into the machine itself.
Consumables wear out on a published schedule
Filaments and cathodes are service items, and the intervals are documented well enough to plan around. Open and hooded filaments: 10–100 h is the handbook range for working arc sources (Wolf 1995, p. 62); the 86-inch averaged 85 h and budgeted two filament changes per week of continuous operation, two hours each (ORNL-1196, p. 21); the Argonne 60-inch got 25–30 h from filament plus chimney, 10–15 h on helium, and engineered a 15-minute changeout (ANL-5907, p. 10). Oxygen contamination and excess temperature are the killers; reversing DC heater polarity at half-life evens the erosion (Wolf 1995, pp. 30–31, 62).
Cold-cathode PIGs replace filament economics with sputter economics. The cathode is worn out when its erosion crater reaches about the anode bore radius, after which the discharge destabilizes (Wolf 1995, pp. 73–74) — a concrete inspection criterion — and arc power is capped near 1 kW per cathode by runaway thermionic emission. At sub-ampere hydrogen arcs the intervals are long: the Rutgers internal PIG runs more than 40 h between servicings (Koeth 2015, p. 2), and the Eclipse rebuild interval is 120 h at 150 W (Potkins et al. 2017). Heavy-ion machines at 1–15 A pay "a few hours to a day" (Clark 1981, p. 233). The upgrade path when cathode life matters is the indirectly heated block cathode, whose temperature is set by rear electron bombardment independent of the arc (Wolf 1995, p. 75). Machine spare cathode buttons before first light.
Published operating points, 1949–2017
Every row below except the Wolf handbook entry — the reference table's canonical ranges — is a specific source that ran, with its numbers as published. The outputs are measured at different points (exit slit, DC test-stand target, internal beam) and cover both H⁺ and H⁻, so they are context markers rather than a ranking; the spread itself is the point, microamperes to milliamperes from the same few architectures, scaled by arc power and slit area.