With GaN power FETs, SiC MOSFETs, and other wide-bandgap devices showing up in more space power systems, a question keeps coming up: when you report SEE cross-section data for a GaN device, should the x-axis be LET in silicon or LET in gallium nitride?
This matters because LET is material-dependent. A given ion at a given energy deposits a different amount of energy per unit path length in GaN than it does in Si, because the materials have different densities and electronic stopping powers. For example, at the same beam energy, the LET in GaN is roughly 1.6–2× higher than in Si for many heavy ions, owing primarily to GaN’s higher density (6.15 g/cm³ vs. 2.33 g/cm³ for Si). So reporting in one convention vs. the other shifts your entire cross-section curve along the x-axis.
The case for always using LET(Si):
For practical radiation hardness assurance, LET(Si) is the right choice and arguably the only defensible choice for mission applications. Here’s why:
-
The environment is defined in LET(Si). Space radiation environment models (CREME, SPENVIS, OMERE) output integral LET spectra in MeV-cm²/mg in silicon. When you compute an SEE rate using the RPP method or any Figure of Merit approach, you’re integrating your cross-section curve against a differential LET flux in silicon. If your cross-section data is in LET(GaN), you’d need to convert and that conversion introduces unnecessary complexity and opportunity for error.
-
Requirements are written in LET(Si). When a program spec says “the part shall be SEL-immune to LET = 75 MeV-cm²/mg,” that’s LET in silicon. It doesn’t matter what the device is made of. The requirement reflects the environment, not the device physics.
-
Comparability across technologies. If you have a design trade between a Si MOSFET and a GaN FET for the same application, you want their cross-section curves on the same x-axis so you can make a direct comparison. LET(Si) gives you that. LET(material) does not.
-
It’s what the beam facility gives you. When you go to TAMU or LBNL, the cocktail table lists ion species with LET values in silicon. You pick your test points in LET(Si). There’s no reason to convert after the fact.
When LET(material) is appropriate:
The material-specific LET becomes relevant when you’re doing research into the physics of why a device responds the way it does. If you’re studying charge collection mechanisms, modeling energy deposition in the active region with TCAD or Monte Carlo transport codes, or trying to understand why the SEE threshold correlates (or doesn’t) with the bandgap or critical electric field of the material, then you need to think in terms of actual energy deposited in the material. That’s a research question, not a mission assurance question.
Some published academic papers on GaN and SiC SEE report data in LET(material), which is fine for the physics discussion but creates confusion when someone tries to use that data for a rate prediction or a requirements comparison. At minimum, papers should state clearly which convention they’re using. Ideally, they’d provide data in both.
Practical implications for testing:
When you test a GaN power FET at a heavy-ion facility, you should:
- Plan and report your test points in LET(Si), because that’s what your beam facility provides and what your rate prediction tools expect.
- If your test report is going into a radiation database, label the x-axis unambiguously: “LET (MeV-cm²/mg) in Si” — not just “LET.”
- If you need LET(GaN) for research analysis, you can always convert after the fact using stopping power tables (SRIM/TRIM). Going from the standard convention to the material-specific one is straightforward; the reverse requires you to know exactly which ion and energy was used at each point.
Bottom line: Use LET(Si) for all practical RHA work, test planning, requirements verification, rate prediction, and data reporting. Reserve LET(material) for mechanism studies and physics papers, and even then, always provide the LET(Si) values alongside so the data is usable by the broader community.
What are other people’s experiences with this? Has anyone run into issues where the LET convention caused confusion in a review or in trying to use published GaN/SiC data? Are there cases where LET(material) is actually the better choice for practical work?