Construction Classics

Beam extraction hardware

Extraction is the thinnest subject in the open cyclotron literature and the one amateur builders ask about most. These four reports span fifteen years at Berkeley and answer the same question four different ways: pulse the deflector hard and fast, perturb the magnetic field until the beam walks out on its own, drive the beam out with a second RF field, or give up on preventing sparks and engineer the power supply to survive them.

Why these documents can be hosted

All four are University of California Radiation Laboratory reports prepared for the US Atomic Energy Commission — UCRL-3476 and UCRL-10655 carry contract W-7405-eng-48 on their title pages, and UCRL-8578 carries the government-sponsorship disclaimer. AEC contractor reports are treated as US Government works and carry no copyright; pre-1978 distribution without a copyright notice is an independent public-domain basis. UCRL-10655 additionally states Printed in USA. Price 50 cents. Available from the Office of Technical Services, U.S. Department of Commerce — a priced US Government publication. None of the four carries a copyright notice, publisher imprint, or journal-reprint marking. Verified against the scans and the OSTI records; determinations made August 2026.

1948 — switch it fast enough and it leaves

UCRL-95 is the oldest document in this collection to describe a working piece of extraction hardware, and it is unusually circuit-level. The 184-inch synchrocyclotron pulled its beam out with a pulsed electrostatic deflector working against a magnetic channel, and this report is the pulse generator that drove it.

The engineering is set by one number. With about 0.1 inch of separation between successive turns and deflector bars an inch apart, the pulse has to reach full voltage inside the time the ion group needs to arrive — so the 10%-to-90% rise time must be of order 0.1 µs. Everything else follows from that: two water-cooled pulse transformers connected symmetrically about ground, oil-filled in a lucite tank, wound on Westinghouse 2-mil oriented Hipersil C cores, with a lead structure forming an effective 2-ohm transmission line to match into the primary. The switch is a battery of sixteen paralleled Kuthe 5C22 hydrogen thyratrons, eight per transformer, because one 5C22 will pass 5000 A for 0.1 µs and they needed more. The trigger is a 1000 V, 20-ohm pulse into the grids, each thyratron fed from its own artificial transmission line so that all sixteen fire together; a transformer so built survived a 300 kV output test. What the deflector actually needed was about 75,000 V/cm — roughly 200 kV across bars an inch apart — to shift the centre of rotation of 200 MeV deuterons.

Two pieces of honesty make the report worth more than its circuit. The thyratrons were rated 16 kV and arced plate-to-grid above 11 kV in this service, so the supply was held below that — derate a pulsed switch for what your circuit does to it, not for its data sheet. And the achieved 10–90% rise time was 0.15 µs against the 0.1 µs requirement, blamed on capacitor internal inductance, and reported as a shortfall rather than quietly restated as the specification.

The tube types are museum pieces and the voltages are far beyond an amateur bench. What transfers is the reasoning chain — turn separation sets rise time, rise time sets the switch, the switch count follows from its current rating — and the fact that a fast high-voltage pulse into a capacitive electrode is a transmission-line problem, not a power-supply problem. The scan’s OCR is poor throughout; read it from the page images.

1956 — let the field do the work

UCRL-3476 is theory rather than hardware, and it is the entry point to regenerative extraction. Its introduction is a compact history of the idea: Tuck and Teng proposed in 1950 removing synchrocyclotron beams with magnetic field perturbations near the outer radius — a peeler, a region of radially decreasing field, followed at some angle by a regenerator, a region of radially increasing field. LeCouteur and Judd analysed it exactly for a linear field region, LeCouteur extended it to rapidly falling fields, and Crewe and colleagues applied it at Liverpool and then at Chicago, the latter extracting beam with a regenerator alone.

Stubbins extends those analyses to a regenerator-only system in the nonlinear field near maximum energy, for the 184-inch. The useful part for a builder is the structure: the required perturbation is determined in a stated seven-step procedure (p. 6), which turns "shape a magnetic bump until the beam walks out" into an ordered calculation with the regenerator shape, its position, and the growth rate of the radial oscillation amplitude as outputs. Axial motion is treated too, briefly — the perturbation that walks the beam out radially can also lose it vertically.

Regenerative extraction has one property that recommends it at small scale: it adds no high voltage. The cost is that it demands field control at the outer radius, which is exactly where a small magnet is least well behaved.

1958 — drive it out with RF

UCRL-8578 starts from an observation that reads like a complaint: particles are lost by vertical betatron blowup in the low electric field of a synchrocyclotron anyway, so why not use a resonant amplitude increase deliberately. The report analyses a radiofrequency perturbation with a gradient on the outward side of a reference orbit, arriving at an equation of Hill type — the unstable solutions are the ones that extract.

The worked case is a cyclotron with 50 MeV deuterons in a 17,000 gauss field, computed on an IBM 650, and it needs a gradient of 4.3 kV/cm over a 60° azimuth sector, against a stated design ceiling of under 5 kV/cm. Set that beside the 1948 pulser at the top of this page, which needed about 75,000 V/cm: the RF route asks for roughly a factor of fifteen less field, and buys it with control complexity instead. Two features turn out to matter: the turn separation is influenced by the perturbation itself, and the orbits must be phased so that the amplitude maxima occur near one azimuth, because that is where the extraction channel is. Do not expect a single kick — the report is explicit that orbits precess, take a few influential encounters, and drift off the perturbation in between.

The stated advantage is the one worth remembering. Frequency and gradient can both be changed to match conditions, which sidesteps the difficulty magnetic extraction has in variable-energy machines — where a fixed magnetic perturbation is only correct at one energy. Any machine intended to run at more than one energy inherits that problem.

1963 — assume it will spark

UCRL-10655 is the direct companion to UCRL-10654, already hosted here: that report gives the electrode materials, the achievable gradients and the spark-damage ranking, and this one gives the power supply that feeds them. Read together they are a complete electrostatic deflector design.

The premise is a concession. Sector-focused cyclotrons need higher deflector gradients than ordinary cyclotrons because the particles are more energetic, and the way to get a higher operating gradient is not to prevent sparks but to limit the energy each spark can deliver — exciting the deflector from a low-energy-storage supply so that electrode surface heating per spark stays controlled (abstract, p. 6; spark behaviour discussed from p. 19). Electrode damage, not breakdown voltage, is the real limit. The whole supply stores only about 2.5 J at 120 kV, and a thyratron crowbar grounds the oscillator’s screen grid — not the high voltage — to cut power to the deflector within a few microseconds of a spark.

What makes this report unexpectedly useful at amateur scale is the parts list. The supply is a pair of high-frequency Cockcroft-Walton rectifiers for the 88-Inch Cyclotron, each a six-stage circuit delivering 120 kV at 5 mA, driven by a 100 kc oscillator. The stack is built from inexpensive silicon diodes rated 600 V and 0.75 A with about 2 µs storage time, assembled on etched circuit boards, with 900 pF 30 kV ceramic capacitors similar to those in television sets between decks. Output is controlled by plate modulation of the oscillator and regulated to 0.01%.

A 1963 national laboratory reached 120 kV using television capacitors, cheap diodes and a driven multiplier rather than a transformer — which is exactly the topology an amateur would reach for today, and the report says why the values were chosen.

Reading them together

The four documents are a fifteen-year argument about where to put the difficulty. In 1948 it went into the switch: enormous instantaneous current, ferocious rise time, tube count driven by ratings. By 1956 it had moved into the magnetic field, where the hardware is passive but the field shaping is exacting. By 1958 it had moved into a second RF system, cheap in volts and expensive in control. By 1963 it had moved into the power supply, on the assumption that the electrodes will spark and the design’s job is to make sparks survivable.

For a small machine the honest reading is that none of these is easy, which is why extraction remains the gate most amateur builds never pass. The 1963 answer scales down best, because limiting stored energy is easier at small scale, not harder.

Design rules extracted from these documents

27 Design Guide rules come from these four reports — UCRL-95, UCRL-3476, UCRL-8578 and UCRL-10655 — the largest single addition the extraction domain has had, and the reason it is no longer this collection’s emptiest subject. Representative examples:

• dg-756 — design the deflector supply to limit the energy per spark — the 88-Inch stores only 2.5 J at 120 kV
• dg-764 — derive pulser rise time from turn separation, not from the switch
• dg-766 — derate pulsed switches for what operation does to them: 16 kV thyratrons failed above 11 kV
• dg-772 — the three requirements any regenerative extraction scheme must satisfy
• dg-780 — prefer electric extraction elements in a variable-energy machine
• dg-763 — magnetically shield glass tubes sitting in the stray field — mundane and mandatory

Used by the beam extraction deep dive.