At Pretzl Connect 2026, Nexperia argued that GaN versus SiC versus silicon is the wrong question – and that engineers designing for AI data centres, robotics, and electrification need to judge power devices by how they behave in the application, not on paper.
For much of the past decade, the case for wide-bandgap (WBG) semiconductors has been made on switching performance. Gallium nitride (GaN) and silicon carbide (SiC) switch faster and waste less energy than silicon, and that efficiency story drove early adoption. According to Nexperia, that chapter is now largely written.
“Historically, the perception was that you should focus on the technology that allows you to achieve the best efficiency,” said Filippo Scrimizzi, Regional System Application Director at Nexperia. “The world is changing. It is not just performance, it’s a mindset – understanding how the market is evolving.”
Rebuilding capacity
Before turning to WBG, Olaf Vogt, Director and Head of Product, Application Marketing and Engineering, addressed the company’s position. Nexperia lost control of its assembly and test sites in Guangdong and Wuxi in October 2025 and has spent the year since rebuilding supply.
“This is not an easy time for us, but we have a clear recovery plan, which is already in the implementation phase,” said Vogt. Capacity is being significantly expanded at its sites in Malaysia and the Philippines, alongside new foundry and OSAT partnerships.
Why the focus is shifting
Scrimizzi pointed to three converging markets: automotive electrification, the energy transition, and AI infrastructure. In data centres, rack power is climbing beyond 250kW, distribution voltages are rising and architectures are moving from single-phase to three-phase. Switching performance gains are well understood; the next gains, Nexperia believes, will come from thermal performance, power density, packaging, system optimisation, and reliability.
WBG devices also enable new topologies. In a next-generation bidirectional on-board charger, a bidirectional switch allows the AC/DC and DC/DC stages to merge, removing the DC link capacitor and reducing bill of materials, at the cost of greater control complexity. With cars parked for more than eight hours a day, Scrimizzi said, vehicle-to-grid and vehicle-to-load turn the car into an energy store.
The data sheet trap
The session’s central argument was that headline specifications mislead. Nexperia compared its 650V Cascode GaN (Gen 1) with a competitor’s e-mode GaN device. On paper, the competitor looked equal or better on every parameter: 25 versus 33mΩ RDS(on), 134nC versus 173nC Qoss, and lower thermal resistance.
Yet in a 400-to-230V non-isolated buck converter running at 200kHz, the Cascode device achieved higher efficiency at higher power levels. Scrimizzi was careful not to dismiss e-mode: where switching losses dominate, it performs better. But where conduction and soft-switching behaviour matter more, the result reverses.
“Before deciding on the final design, don’t just trust the number you read in the datasheet,” he said. “Test on the bench according to the final application requirements.”
SiC showed the same pattern. Nexperia’s SiC MOSFETs and a competitor’s were equivalent at room temperature, but between 25 and 175°C, Nexperia’s conduction losses grew by around 1.6 times at 40A, compared with around 2.3 times for the competitor.
Matching technology to the application
For AI data centres, Nexperia is lining up 750 and 1200V SiC MOSFETs from 17 to 80mΩ for power supplies, SiC JFETs for hot-swap and solid-state circuit breakers, and very-high-voltage devices above 3.3kV for solid-state transformers. Top-side cooled packages are key to fitting more power into a fixed rack form factor.
Humanoid robots are an emerging opportunity for GaN. Each robot uses 25 to 35 body actuators, drawing 0.5 to 2kW per joint, with potentially 200 to 300 GaN switches per robot, according to Yole Group. Most run on 48V batteries, which Nexperia is targeting with its Gen 2 GaN in the 80 to 100V range. Higher switching frequencies mean smaller passives, smoother motor control, and compact drives inside the joint, while silicon still serves low-power joints.
Optimisation, not adoption
Nexperia is positioning its integrated device manufacturer model as the way to industrialise WBG at scale. In 2026 its GaN portfolio is set to almost double, with around 60 SiC diodes and MOSFETs releasing, including a 750V SiC MOSFET family in industrial and automotive-qualified versions.
Scrimizzi also argued that the package is now as important as the die. “Even the device can be considered a system,” he said, pointing to top-side cooled and leadless packages as enablers of thermal performance.
His closing message was simple. “We do not believe there is a single technology that will be the winner in the market.” The next phase, in other words, is about application optimisation – and bench results, not datasheet tables.
