Heavy Ion SEE Test Planning: LET, Fluence, Range, Bias and How Much Is Enough

Beam time is expensive, scarce and unforgiving: a campaign that tested the wrong effect, at the wrong bias, with ions that stopped short of the sensitive volume, produces data that looks fine and means nothing. This guide is the list of decisions to make before you arrive, in the order they matter, with the standard that governs each and the reason behind its numbers. The standards page holds the documents themselves, the facilities, and the longer reasoning.

1. Decide why you are testing

There are three different campaigns that go by the name "SEE test", and the plan is different for each.

• A destructive screen. Show that the part does not latch up, burn out or rupture up to a stated LET at the worst case bias and temperature. This is the campaign that rules a part in or out. It needs high LET, high fluence, worst case conditions, current monitoring and, for latchup, a decision in advance about whether protection will be credited.
• A characterisation. Measure the upset, transient or functional interrupt cross section as a function of LET well enough to fit a curve and compute a rate. It needs several LET points spanning the threshold and the plateau, enough events per point for the statistics, and the flight operating mode.
• A system or application test. Irradiate the part in its flight circuit, or a board, and watch what the system does. It answers the propagation question directly and it is the only way to verify a mitigation such as latchup protection or a watchdog.

Destructive screening comes first, because its outcome decides whether the other two are worth doing. Do not bury a latchup screen in the last hour of a characterisation run.

2. The LET ladder: why 37, why 75, why in between

Requirements for destructive effects are written as a threshold LET the part must exceed. Two numbers dominate. 75 MeV cm2/mg is close to the LET of iron, the heaviest abundant cosmic ray ion; the integral flux above it behind normal shielding is a few times 1e-6 per square centimetre per day, small enough that a part immune to 75 sees essentially no destructive events over any mission. 37 MeV cm2/mg is the threshold below which the integral flux rises steeply (it is about twenty times higher than above 75, and rising fast), and it is the LET requirement that is commonly written for missions and part types where 75 cannot be demanded: shorter missions with fewer susceptible devices, where the residual rate above 37 is tolerable, and vertical power MOSFETs, where the mechanism responds to ion energy and range in a way that makes an LET based requirement above about 37 impractical to demonstrate, so 37 with derating and range control has become the working standard. The standards page has the full argument and the SEL test LET tool computes what LET your mission actually needs for a given reliability rather than taking either number on faith.

For a characterisation the ladder is different: you want points below, around and well above the threshold. Four to six LETs from about 2 to 60 or more, with one placed to pin the threshold once you have seen where it is, is a common shape.

3. Fluence and event counts

The standard test method asks for each run to continue until a fluence of 1e7 ions per square centimetre or a set number of events, around 100, whichever comes first. The reason is statistics. A cross section from ten events has a 30 percent uncertainty and from a hundred a 10 percent one; and a null result to 1e7 bounds the cross section at about 2.3e-7 cm2 at 90 percent confidence, which for a device of a few square millimetres is a factor of a hundred below its area, enough to say a threshold has not been reached. For destructive screening, the fluence has to be large compared with one over the cross section you are trying to exclude: if the sensitive area for burnout in a power device might be 1e-4 cm2, 1e7 ions per square centimetre gives a thousand chances to see it, which is convincing, and 1e5 gives ten, which is not. Flux matters too: keep it low enough that events can be counted individually and that heating and dose do not distort the measurement, typically 1e3 to 1e5 per square centimetre per second.

Total dose accumulates during heavy ion testing. A run to 1e7 at LET 60 deposits close to ten kilorads (dose in rad is 1.6e-5 times LET times fluence); several such runs can degrade a commercial part enough to change its response. Track the dose per device and plan fresh samples for the high fluence runs.

4. Range and energy: the mistake that hides

An ion has to reach the sensitive volume with its nominal LET. Beam range in silicon depends on the ion and its energy; a low energy cocktail at high LET may have a range of only 30 to 40 micrometres, which is fine for a memory cell a few micrometres down and useless for a latchup path through a deep well, the drift region of a power MOSFET, or a flip chip die irradiated through its backside. The rules of thumb: know the depth of the structure you care about, demand a range that exceeds it with margin (a range at least twice the depth is a common minimum, and for power devices the test method specifies range and energy explicitly), and remember that tilting to raise effective LET shortens the projected range. Higher energy per nucleon facilities exist for exactly this problem, and the facility list shows energy and range alongside LET for each. The beam calculator computes the LET and range that survive air gaps, windows, degraders and overlayers.

Most packaged parts have to be opened. Plastic is removed chemically; ceramic lids mechanically; flip chip and stacked die parts may need thinning and irradiation from the back, which changes the range requirement again. Decide this before the samples are bought, because it decides how many you need.

5. Bias, temperature and operating mode

Latchup is worst at the maximum supply voltage and the maximum temperature the part will see; a screen at nominal voltage and room temperature can pass a part that latches in flight. Test at the maximum rated or the maximum application voltage, and heat the die to the flight maximum, commonly 85 or 125 degrees Celsius. Burnout and gate rupture in power devices depend on drain and gate voltage directly; those tests sweep voltage to find the safe operating area, and the derating advisor says what the answer usually looks like by technology. Upset cross sections depend on operating mode, clock frequency, data pattern and sometimes on supply voltage in the other direction (lower voltage, more upsets). Test in the flight mode, or in a mode you can argue is worse, and write the argument down.

6. The test setup

• Supply current monitoring on every rail, sampled fast enough to catch a latchup and logged, with a programmable limit and a power cycle. If latchup protection will be credited in flight, the test protection should mimic it.
• Event detection that matches the requirement. A transient test needs an oscilloscope trigger set at the amplitude the application cares about, and a record of every waveform. An upset test needs a memory readback fast enough that multiple upsets in a word are counted as one event, not missed. A functional interrupt test needs a definition of "functional" and a recovery procedure ranked by severity: software reset, hardware reset, power cycle.
• Dosimetry you trust. Fluence and flux from the facility, with the beam uniformity over the die confirmed.
• Remote control. Everything that changes a condition must be changeable from outside the vault.
• A run log with ion, energy, angle, effective LET, fluence, flux, bias, temperature, mode, events by type, and the dose to date on each sample. Every later analysis depends on it.

7. How many parts

Single event effects are a property of the design and process more than of the individual die, so test methods ask for a small number of samples, typically three or more, rather than the statistical samples a total dose test needs. The reasons to bring more are dose accumulation, destructive tests that consume samples by definition, and the need to keep an untested control. For power devices expect to destroy parts and plan accordingly. If the flight lot differs from the tested lot, the argument that the result transfers rests on the process being the same; record the lot and date codes.

8. How much is enough

Enough is when the evidence closes the requirement with margin. For a destructive screen that means a null result at or above the required LET, at the worst case bias and temperature, with a fluence that is many times one over the assumed area, with ions whose range covered the structure. For a characterisation it means a fitted curve whose confidence region, propagated to a rate, sits under the mission's tolerable rate; the rate tool and the SEECA workbench do that comparison, and the workbench also tells you, for a requirement you have not yet met, the effective LET and fluence a clean run would need, or that no null test can close it. Deciding that before the campaign is what turns a test into a plan.

9. When protons are the right beam, and when they are not

A 200 MeV proton test is cheaper, needs no delidding, reaches the whole die and reproduces the trapped proton environment of low Earth orbit directly. It also bounds heavy ion response only up to the recoil LET of about 15 MeV cm2/mg, so it cannot prove latchup immunity and it says nothing about the heavy ion rate of a part with a threshold above 15. The proton proxy tool quantifies what a clean proton result leaves uncovered for your part class, and the paired case studies show real parts where the proton test was reassuring and the heavy ion test was not.

Frequently asked questions

Related tools and pages

• Heavy ion beam calculator: effective LET, range and fluence at the die.
• SEL test LET tool: the LET your mission actually needs.
• Standards, facilities and data sources: the test methods and where to run them.
• Proton proxy tool: what a proton test does and does not prove.
• SEECA workbench: which parts need testing and how much.

References