GaN Traction Inverter Explained: AVL × VisIC Put D³GaN on the Road — 99.67% vs SiC

 

Bottom line: GaN in a traction inverter has mostly lived on slides and dynamometers. On Oct 13, 2026, VisIC Technologies and AVL announced that VisIC’s D³GaN GaN-on-silicon power modules are integrated into an AVL vehicle demonstrator: a high-speed e-axle driven by a GaN traction inverter, now running in a complete vehicle. Below I unpack four things: what 99.67% means in watt-hours, why partial load is where GaN wins, why 400 V comes first, and the GaN-on-Si manufacturing angle. All figures are from the company press release, not our own measurements.

What happened

• Who: VisIC (D³GaN power modules), AVL (inverter development and vehicle integration), and NXP Semiconductors (gate driver).
• Platform: AVL PI850e Inverter Development Platform plus AVL control software. The partnership started in 2024.
• Bench result (AVL Germany, WLTC): 99.67% at 10 kHz and >99.8% at 5 kHz. Under those conditions, inverter energy losses were more than 60% lower than a comparable SiC inverter.
• Positioning: today’s 400 V packs make PHEVs and range-extended hybrids the most attractive fit. The roadmap points to 800 V GaN modules.
• Spillover: the same platform is pitched for AI data-center (AIDC) power and “physical AI” such as robotics.

Scope check: this is not a GaN-for-AI-server-PSU story. It is GaN in the main EV drive inverter, where current levels, thermal cycling and lifetime requirements are an order of magnitude tougher. It also complements the 400 V to 800 V SiC story: the question here is whether GaN can push SiC aside at the same 400 V.

Engineering point 1: turn efficiency into loss before you judge “60%”

Flip the number. 99.67% efficiency means 0.33% loss, and >99.8% means under 0.2%. If 10 kWh flows through the inverter over a WLTC run, 0.33% is about 33 Wh. If you apply the “>60% lower” claim to the 10 kHz point, the implied SiC baseline lands around 0.8% loss (about 99.2% efficiency). That is my back-of-envelope estimate. The release does not give the SiC inverter’s absolute efficiency, device generation or switching frequency. The bigger takeaway is that inverter loss is now a second-decimal fight, so the next payoff is likely a smaller cold plate and a simpler cooling loop, not only a few extra kilometers of range.

Engineering point 2: partial-load switching loss is the battlefield

Inverter loss splits into conduction loss, which scales roughly with current squared and dominates at high torque, and switching loss, which you pay every PWM cycle. A drive cycle like WLTC spends most of its time at low-to-mid torque, so switching loss takes a bigger share of the total. That is GaN’s home turf: a lateral HEMT has no body-diode reverse-recovery charge like a Si or SiC MOSFET, so its edges are faster and cleaner. Note that 10 kHz efficiency is slightly lower than 5 kHz, but still above 99.6%. Higher switching frequency cuts current ripple and motor harmonic loss. The price is steeper dv/dt, which puts more stress on motor insulation and bearings and makes EMI harder. Only in-vehicle data will close that trade.

Engineering point 3: why 400 V PHEV/EREV first

Mainstream GaN-on-Si power devices are typically 650 V-class, which fits a 400 V bus with surge margin. An 800 V platform usually needs 1200 V-class switches, a much harder target for lateral GaN on a silicon substrate. That is why the release places the sweet spot at 400 V plug-in and range-extended hybrids. Their packs are small, the engine covers peak loads, and the e-drive spends long stretches at partial load, which amplifies GaN’s advantage. The 800 V GaN module is an announced plan. The route (higher-voltage devices, series stacking or multilevel topologies) and the timing are not disclosed.

Engineering point 4: the GaN-on-Si fab math and AI power

SiC cost is gated by the substrate: expensive wafers, with mainstream production still at 150/200 mm. GaN-on-Si can use existing 200 mm and 300 mm silicon fabs, which in principle raises the volume ceiling. VisIC also treats automotive validation as a ticket into AI data-center and robotics power, since traction inverters face harsher thermal cycling, vibration and lifetime demands than server PSUs. Still, fab compatibility does not mean GaN has already won on cost. Yield, package thermals (small GaN dies mean high heat-flux density) and the AEC-Q qualification timeline all still have to land.

Open questions

• 99.67%, >99.8% and −60% are dyno WLTC figures. Vehicle-level range and cooling reductions are not yet published.
• The SiC baseline (part, generation, switching frequency, cooling) is undisclosed, so the −60% reference is opaque.
• The demonstrator’s model, power rating and e-axle specs are not public.
• GaN short-circuit withstand time, plus the effect of high dv/dt on motor insulation and EMI, still need in-vehicle and long-term data.
• The 800 V GaN module path, automotive qualification and production timing are still blank.

Related videos

• From 400V to 800V: The Silicon Carbide Revolution
• How to Choose the Right Electric Vehicle Motor: Permanent Magnet vs. Induction vs. Axial Flux

Quick poll: will GaN take 400 V plug-in and range-extender drives from SiC first, or is the real fight only once 800 V GaN ships? And what worries you most: reliability, EMI or cost? Tell me in the comments.

Source: PR Newswire / VisIC Technologies: AVL and VisIC Bring High-Power GaN Traction Inverter Technology to the Road (Oct 13, 2026). Efficiency and loss figures are company-reported WLTC dyno results, not our measurements. The SiC baseline estimate is illustrative arithmetic.
Published on The Automation Core | Video: The Engineering Core YouTube channel