Brookhaven National Laboratory Tandem Van de Graaff

The Tandem Van de Graaff is the Brookhaven accelerator most of the published planetary single event effect results were taken on. Its distinguishing property for a test planner is that ion energy is tunable at the machine rather than by putting a degrader in the beam, so there is no straggling in the energy distribution reaching the part, and a single event effects campaign can be turned around in a week [2].

Exposures are run in vacuum, and the energies are the lowest of the four accelerators on the radiation testing page: ranges in silicon of 30 to 200 micrometers, so parts are delidded and flip chips thinned, or at the high-LET tune not tested at all [1].

Two other Brookhaven beams are not this facility. The NASA Space Radiation Laboratory runs at energies whose ranges in silicon allow a whole circuit card assembly to be irradiated without delidding; its envelope, its spilled beam structure and the error that structure introduces are recorded on the radiation testing page. The Alternating Gradient Synchrotron was used to deliver 10 GeV per nucleon gold ions at 12 MeV-cm2/mg linear energy transfer at normal incidence, at a range far exceeding that of 200 MeV proton recoils, which was sufficient to latch up any device with a measurable cross section below that LET [6].

Laboratory data

Parameter	Value
Operator	Brookhaven National Laboratory [2]
Location	Upton, New York, United States
Commissioned	Not published
Type	Tandem electrostatic accelerator, heavy ion single event effects
Floor area	Not published
Capabilities	Heavy ion beam line
Simulant or terrain	Not applicable
Instrumentation	Not published
Ground truth	Not published. Test methods follow ASTM F1192 or JEDEC JESD57 [4][5]
Fidelity limits	Monoenergetic single-species beam; 30 to 200 micrometer range in silicon [1]
Access	User facility. One week campaign turnaround achievable [2]
Cited by	radiation testing, sensors and IMUs

No operator-written facility description is held. The envelope below comes from two users’ facility comparisons and from the ion, energy and range figures recorded in the campaigns themselves [1][2][3].

Capabilities

Heavy ion beam line

Parameter	Value
Working volume	Vacuum chamber; dimensions not published [4]
Test article limits	Ceramic cover removed; flip chip thinned, or not tested at the high LET tune [1]
Vacuum	Yes. Heavy ion exposures are run in vacuum [4]
Temperature	Set by the user. Campaigns here report room temperature to 90 C [3]
Illumination	Not applicable
Slope	Not applicable
Gravity offload	Not applicable
Instrumentation	Supplied by the user

Two tunes. The low-LET tune covers 1 to 6 MeV-cm2/mg at 85 to 200 micrometers of range in silicon, with the ceramic cover removed, flip chips thinned, and no heat spreader or heat sink [1]. The high-LET tune covers 6 to 85 MeV-cm2/mg at 30 to 85 micrometers, and flip-chip parts should not be tested on it at all.

The ranges the campaign records give are consistent with that: bromine-81 at 287 MeV reaches 36 micrometers in silicon at LET 37.5 MeV-cm2/mg at normal incidence, and silicon-28 at 187 MeV reaches 74 micrometers at LET 7.8 [3].

Because ion energy is tunable at the machine, no degrader is needed to reach an intermediate energy, and the beam therefore arrives without the energy spread a degrader introduces [2]. That is what makes a one week turnaround possible.

What it does not reproduce

The spectrum. Monoenergetic, single-species, as at every accelerator here. See the radiation testing page.

Penetration. The shortest ranges of the four facilities, 30 micrometers at the high-LET tune, so every part is opened and many are thinned [1]. A result taken here is a result on a modified part.

A stated envelope. Nothing held is written by the operator. The two tune rows above come from a JPL test-engineering guideline whose author states in its own foreword that the facility table was already out of date at publication [1].

Where it sits among heavy ion sources

A Van de Graaff was one of four ion source categories in the earliest written JPL heavy ion upset practice, alongside fission fragment sources, alpha emitters and cyclotrons [9]. That guideline carries no device data, and its facility recommendations name the 1984 machines then available to JPL [9]. What still applies here is the equipment it lists for the experimenter to bring: vacuum chamber, beam energy and flux measurement, uniformity measurement, a positioning test card and a device exerciser.

The alternatives differ mainly in penetration and in how fast the ion can be changed.

• Cyclotron cocktails. The Berkeley 88-Inch Cyclotron offers four cocktails at 4.5 to 30 MeV per nucleon, 1 to 100 MeV-cm2/mg and 40 to 1400 micrometers of range in silicon, and switches species in about two minutes by retuning frequency [8]. Those are operator-stated capabilities, not measurements [8]. The Texas A&M beam is run in air, which the compendia pair with this facility’s vacuum work [4][5].
• Very high energy beams. The Michigan State facility delivers 60 to 143 MeV per nucleon in air, penetrates commercial package overlayers without delidding, and agreed with Brookhaven results on a Matra HM65656 SRAM [10]. Switching ions takes a 24 hour retune and beam time cost 2300 to 2700 dollars per hour at commissioning. The same paper records a plastic packaged IDT71256 whose cross section was two orders of magnitude below the delidded part on a degraded krypton beam, because the ions ranged out [10]. The NASA Space Radiation Laboratory pursues the same goal with millimeter ranges and a Bragg peak method, which its author states still needs validation across device types and package configurations [7].

The cost of the tandem route is therefore the preparation of the part. Nothing held states a per-hour price or an annual hour budget for this facility.

Campaigns run there

MEMS accelerometer latchup against temperature, reported 2018. Two Colibrys parts were exposed under bromine-81 and silicon-28 [3]. The SF1600S showed a non-destructive latchup threshold between LET 7.88 and 11.44 MeV-cm2/mg above 70 C at maximum recommended bias, and the MS1002A between 11.44 and 19.71 above 70 C. The temperature dependence is the result worth carrying: on the MS1002A the latchup cross section at LET 37.5 and 1e7 ions/cm2 was 6.1e-6 cm2 at 75 C against 1.5e-6 cm2 at room temperature, a factor of four [3]. A latchup screen run at ambient understates the flight case by that factor for this part.

JPL compendia heavy ion exposures, 2007 to 2012. The vacuum heavy ion half of both JPL compendia was run here, with the in-air and longer-range work at Texas A&M [4][5]. Entries taken in this configuration include the Elpida EDS5104ABTA-75 512 Mbit SDRAM, upsetting above LET 2.7 MeV-cm2/mg, with a recoverable mode-register functional interrupt above 3.8 and a run of over a thousand consecutive errors above 5.4, the latter clearing on rewrite or power cycle [4]. The same part family in the later compendium upset and suffered functional interrupts below LET 2 at about 2e-9 cm2 per bit [5]. The sample size and method caveats that apply to every compendium entry are on the Texas A&M page.

Commercial parts for a Class D payload, reported 2018. Fourteen commercial parts were screened for latchup at this facility and at Texas A&M [3]. Four latch destructively: the LTC2440 ADC, the LTC6655 reference, an ECS 4 MHz crystal oscillator and the Altera EN5322QI regulator, which fails permanently at LET 42.8 MeV-cm2/mg. The method point is range: a 287 MeV bromine beam cannot pass the 12.76 micrometer copper redistribution layer of the DRV8412 motor driver, so a clean result there would have been a false negative, and the authors confirmed stackup by cross sectioning and SEM [3].

Europa Clipper candidate parts, reported 2019. The Applied Physics Laboratory heavy ion work used a 15 MeV per nucleon beam only, with LET coverage stopping at 75 to 83 MeV-cm2/mg, so saturation cross sections above that are extrapolated for most parts [11].

Test conditions that decide whether a result transfers

Three later works state conditions the screening data above depend on.

Preparation and exercise state. A 2016 JPL tutorial chapter states that heavy ion parts are de-lidded and tested in vacuum because the ion range is tens of micrometers, and that bias and exercise state dominate results for complex devices [12]. On some microprocessors the cross section at a given LET rose by an order of magnitude when the exercising program used the cache [12], so a processor cross section cited without its test program is not a usable number. The chapter is analytical and irradiated no hardware; its worked rate example uses hypothetical parts. It tabulates a typical ion menu from lithium at LET 0.45 to gold at 82.3 MeV-cm2/mg with no facility named [12].

Misdiagnosed mechanisms. The 2019 Europa Clipper characterization of the RH3845 step-down controller used the Berkeley cyclotron and a pulsed laser rather than this machine, but it bears on how any beam result is read [13]. A destructive event first attributed to single event burnout was traced to transient-induced cross conduction in the external synchronous MOSFETs, and raising the in-line gate resistance from 2 to 10 ohms removed it up to 100 V input at LET 79 MeV-cm2/mg. The authors state the root cause is still under investigation and that the effect can occur electrically without a beam, so the beam cross section may measure an instability the beam only triggers. The mitigation was a board-level choice, not a part derating.

Acceptance thresholds. The 2021 NASA radiation hardness assurance guidelines define the categorization framework, the accepted LET acceptance thresholds and what a single event effects test requirement must specify [14]. They contain no measured part results, and the SET signature tables apply only to radiation hardened parts, not commercial ones. The 42.8 MeV-cm2/mg ceiling of the 2018 screen has to be read against those thresholds rather than as a rating [3].

What is not established

No operator-written description of the machine is held, so commissioning date, floor area, flux range and beam uniformity are unpublished here. The tune ranges rest on a guideline whose author calls its facility table out of date [1]. The campaigns above are screening data on two to three parts per type [3][4][5], and the 2018 screen reports no upset, functional interrupt or dose data and is bounded by an LET of 42.8 MeV-cm2/mg, below the customary 75 to 85 qualification point [3]. Whether a result on a delidded or thinned part transfers to the packaged flight part is not answered by anything held. The facility has not been compared with the others on identical parts beyond the single SRAM agreement reported by Michigan State [10].