Why Wide Bandgap Materials Are Replacing Silicon in Power Electronics
Silicon power devices hit a wall. A silicon MOSFET's theoretical limit for on-resistance times die area (the Baliga figure of merit) at 600V is about 50 mΩ·cm². Silicon carbide (SiC) devices achieve 5-10 mΩ·cm² at the same voltage. Gallium nitride (GaN) is even better in some configurations. That's not an incremental improvement — it's an order of magnitude, and it translates directly into smaller, lighter, more efficient power converters.
The physics behind this advantage is the bandgap itself. Silicon has a bandgap of 1.12 eV. SiC is 3.26 eV. GaN is 3.4 eV. A wider bandgap means the material can sustain a higher electric field before breakdown — about 10× higher for SiC compared to silicon. That higher critical field lets you build thinner, more heavily doped drift regions, which is where most of the on-resistance lives in a high-voltage power device.
SiC MOSFETs: The EV Power Train Workhorse
Silicon carbide has found its killer application in electric vehicle inverters. The main traction inverter converts DC from the battery pack (typically 400V or 800V) into three-phase AC for the drive motors. This is the single most power-hungry electronic system in an EV, typically handling 100-250 kW continuously.
Tesla was the first major automaker to adopt SiC MOSFETs at scale, using STMicroelectronics' devices in the Model 3 inverter starting in 2018. The efficiency improvement over silicon IGBTs was about 5-8% in real-world driving, which translated to roughly 5-10% more range for the same battery capacity. When your battery costs $10,000-15,000 and weighs 400 kg, that efficiency gain is worth a lot.
Since then, most EV makers have followed. BYD, Hyundai, Mercedes, BMW, and Lucid all use SiC inverters in at least some models. The supply chain has responded — Wolfspeed (formerly Cree), STMicroelectronics, Infineon, onsemi, and ROHM are all ramping SiC wafer and device production aggressively.
This connects to the ideas in Sensor Chip Design: CMOS Image Sensors, LiDAR Receivers, and.
SiC Manufacturing Challenges
Growing SiC crystals is hard. The physical vapor transport (PVT) method used to produce SiC boules operates at about 2,300°C — compared to silicon's 1,414°C Czochralski process. Growth rates are slow, about 0.2-0.5 mm/hour versus several mm/hour for silicon. Crystal defects like micropipes, basal plane dislocations, and stacking faults are more prevalent and harder to control.
The wafer sizes tell the story of maturity. Silicon moved to 300mm wafers in 2001. SiC is still transitioning from 150mm to 200mm in 2025-2026. Wolfspeed's new Mohawk Valley fab in New York is one of the first to run 200mm SiC at scale. The smaller wafer size means fewer dies per wafer and higher cost per device — currently about 3-5× the cost of equivalent-rated silicon IGBTs.
Substrate cost dominates the bill of materials. A 150mm SiC substrate costs roughly $800-1,200, compared to $50-100 for a 200mm silicon wafer. This is why alternative substrate approaches — like growing SiC epitaxy on polycrystalline SiC carriers or even on silicon — are getting serious R&D investment.
GaN HEMTs: Fast Switching for Consumer and Data Center Power
Gallium nitride power devices take a different approach than SiC. Most commercial GaN power devices are lateral high-electron-mobility transistors (HEMTs) grown on silicon substrates — a technology often called GaN-on-Si. The AlGaN/GaN heterojunction creates a two-dimensional electron gas (2DEG) at the interface with extremely high electron mobility, enabling very fast switching speeds.
For a related perspective, see Advanced Packaging Wars: TSMC CoWoS vs Intel Foveros vs Sams.
Where SiC excels at high voltage (1,200V+) and high current, GaN-on-Si devices currently top out at about 650V but switch much faster — often 10-100× faster than silicon with minimal switching losses. This makes GaN ideal for applications where switching frequency determines the size of passive components:
- Phone chargers and laptop adapters — GaN chargers from Anker, Apple, and others are 50-70% smaller than equivalent silicon chargers because the higher switching frequency allows smaller inductors and capacitors
- Data center power supplies — 48V-to-1V conversion for server CPUs benefits enormously from high-frequency switching
- Solar microinverters — Enphase's latest microinverters use GaN for higher efficiency and smaller form factor
- Class-D audio amplifiers — GaN's fast switching reduces dead time and improves audio quality at high power levels
The main GaN-on-Si players are GaN Systems (now part of Infineon), Efficient Power Conversion (EPC), Navitas Semiconductor, and Power Integrations. Texas Instruments has also entered the market with their own integrated GaN products.
The Enhancement vs. Depletion Mode Problem
Native GaN HEMTs are depletion-mode (normally-on) — the 2DEG channel conducts with zero gate voltage, and you need a negative gate voltage to turn it off. This is a safety problem for power electronics, where you want devices to be off by default so that a controller failure doesn't create a short circuit.
The industry has developed several solutions. GaN Systems and EPC use cascode configurations — a depletion-mode GaN HEMT in series with a low-voltage silicon MOSFET that handles the normally-off behavior. Power Integrations uses a proprietary p-GaN gate structure to create true enhancement-mode (normally-off) devices. Navitas integrates the GaN FET with the gate driver and control logic on the same chip, ensuring the device is managed correctly.
Related reading: Semiconductor IP Licensing: ARM, Synopsys, and the IP Block .
Comparing SiC and GaN: Where Each Wins
It's tempting to ask which technology is "better," but they're really optimized for different application spaces:
- SiC wins at high voltage (900V-3,300V) and high current (50A-800A). EV inverters, industrial motor drives, grid-tied solar inverters, rail traction.
- GaN wins at moderate voltage (100V-650V) and high frequency (100 kHz-10 MHz). Consumer chargers, data center power, point-of-load converters, wireless power.
- There's overlap in the 400V-650V range where either technology can work. Cost and switching frequency requirements usually determine the choice.
Vertically integrated GaN-on-GaN devices — where the GaN is grown on native GaN substrates instead of silicon — could eventually compete with SiC at higher voltages. But native GaN substrates are even more expensive than SiC substrates today, so commercial GaN-on-GaN power devices are still limited to niche military and aerospace applications.
What's Ahead
Diamond and gallium oxide (Ga₂O₃) are the next-generation wide bandgap materials that researchers are investigating. Diamond has a bandgap of 5.5 eV and thermal conductivity 5× higher than SiC — in theory, the perfect power semiconductor. In practice, growing device-quality diamond is extremely difficult and no one has demonstrated commercially viable diamond power devices yet.
Ga₂O₃ is more promising near-term. It has a bandgap of 4.8 eV and — crucially — can be grown from melt using techniques similar to sapphire production, which is potentially much cheaper than SiC crystal growth. The main weakness is poor thermal conductivity (about 1/10 of SiC), which limits power density. Researchers are exploring hybrid approaches where Ga₂O₃ devices are bonded to high-thermal-conductivity substrates for heat extraction.
Honestly, for the next decade, SiC and GaN will dominate. They're already in volume production, the supply chain is scaling, and costs are dropping with each year. Diamond and Ga₂O₃ are 2035+ technologies for power electronics at best.