Commercial SiC MOSFETs lose more than half their voltage in space

The reliability contradiction

The presentation quantifies why designers want silicon carbide: versus silicon, SiC power devices offer roughly ten times higher breakdown voltage, about one-hundredth the on-state resistance, about three times higher temperature operation, and about ten times the thermal conductivity. A 2011 NASA concept for high-power solar electric propulsion at about 400 kW, cited in the deck, illustrates the pull: the bus voltage rises from 120 V to 300 V as power grows, and higher-voltage, lower-loss switches are what make that practical.¹

Against those numbers stand the accelerated-lifetime results the deck cites for Wolfspeed 1,200 V, 20 A generation-2 MOSFETs. High-temperature reverse-bias testing at drain voltages up to 1,620 V extrapolates to a mean failure time around 3 by 10^7 hours (about 3,400 years) at 800 V. Time-dependent dielectric breakdown testing extrapolates past 10^8 hours (about 11,000 years) at 20 V on the gate. By every terrestrial measure these are exceptionally durable parts.

Heavy-ion campaigns by NASA, ESA, JAXA, and others, covering devices rated 650 V to 3,300 V, nevertheless observed single-event burnout at typically one half of rated drain voltage, with the analyzed 1,200 V devices failing near 500 V when linear energy transfer exceeded 10 MeV square centimetres per milligram. Ion-induced degradation of gate and drain leakage currents appears before burnout. These findings do not say every particle destroys a switch; a lethal event must deposit enough energy, arrive at the right solid angle, strike within the sensitive area, and find the device biased in a critical state.

From cross-section to mission lifetime

The authors estimate a failure rate for a 1,200 V part operated above 500 V in orbit. They assume the burnout cross-section saturates above 10 MeV-cm2/mg and start from the chip area of about 2 mm by 3 mm (0.06 cm2), reduced by a 50 percent sensitive-area assumption and a 50 percent duty-cycle assumption to a cross-section of 1.5 by 10^-2 cm2, behind 100 mils of aluminum shielding. Integrating the heavy-ion flux above the LET threshold then gives the environment term: 1,000 ions per cm2 per day for the worst-day solar particle event from CREME96, 10 for the 99-percent-confidence cumulative event spectrum of Xapsos et al., 0.9 for geostationary orbit at solar minimum, and 10^-4 for low Earth orbit.

The resulting mean times to failure are stark: about 1.6 hours on the modeled worst day, 160 hours for the 99-percent-confidence solar-event spectrum, 1,786 hours (roughly 75 days) in geostationary orbit at solar minimum, and about 1.6 by 10^7 hours, more than 1,000 years, in low Earth orbit. Those are scenario calculations built from a saturated-cross-section assumption and standard environment models, not flight observations. Their value is the orders-of-magnitude separation they reveal between environments, and the demonstration that a part number and a voltage rating cannot carry radiation reliability without orbit, shielding, duty cycle, and solar-event assumptions attached.

The strongest case for flying SiC anyway

The steelman is that the deck never argues against silicon carbide; it argues for disciplined use. The cited advantages are mass, cost, and power savings, the terrestrial reliability data are genuinely strong, and the authors' own conclusion is that performance advantages may justify use if the de-rating and the leakage degradation are acceptable. The failure channel is also unusually well characterized: four explicit lethal-ion criteria, a measured LET threshold, and a published conservative model give a designer something to design against.

Where a skeptic should push

The worst-case calculation deliberately compresses complicated device physics into a saturated cross-section and area assumptions. Real susceptibility varies with manufacturer, die design, lot, temperature, bias, and accumulated degradation; the authors themselves recommend careful heavy-ion testing of any commercially available SiC MOSFET proposed for space, rather than transferring their example unchanged. The extrapolated terrestrial lifetimes deserve the same skepticism in the other direction: they are extrapolations from accelerated stress, just as the orbit numbers are extrapolations from beam tests. And de-rating below 500 V is not automatically sufficient. Ion-induced leakage degradation can precede catastrophic burnout and erode margins silently, and a converter architecture may expose switching nodes to overshoot beyond nominal bus voltage, so a device sitting at a comfortable average stress can still see lethal transients.

High-voltage lunar grids pay a radiation tax inside every converter

The non-obvious implication is that raising distribution voltage to save cable mass multiplies series devices, shielding, and qualification effort if switches cannot use their nameplate rating: the cable-mass saving and the converter-radiation tax live in the same trade. For lunar surface grids and large electric-propulsion vehicles, where high voltage is the obvious answer to long runs and high power, the usable bus voltage may be set by derated switch stacks rather than by the wiring. The opportunity is still real: SiC's low on-resistance and high-temperature capability can shrink power management and distribution hardware substantially, and the authors' recommended path, careful heavy-ion testing of any proposed commercial part, is a defined qualification route rather than an open research problem. The threat is system optimization performed on terrestrial ratings, producing a bus voltage that no derated switching stack can safely interrupt when a solar particle event arrives.

The bottom line

High confidence: ordinary high-temperature and dielectric lifetime tests do not bound heavy-ion burnout for commercial SiC MOSFETs, and the demonstrated susceptibility near half rated voltage is a primary-result finding, not a model output. Moderate confidence: substantial voltage de-rating, orbit-aware shielding, and lot-specific radiation data can make the performance advantages usable. Any claim of space readiness for a commercial SiC switch without biased heavy-ion testing leaves the dominant demonstrated failure channel unanswered.