Feature articles |
By Nick Flaherty
Power is an increasingly important factor in system design, spilling over into everything from AI data centres to e-mobility, drones and robotics.
From the power needs of AI data centres to e-mobility and lightweight, efficient battery power and motors for humanoid robots, more efficient power systems with higher levels of integration are being shown at the PCIM conference and exhibition in Nuremberg, Germany, in June.
Silicon carbide (SiC) with trench structures is helping reduce the high cost of devices in the AI data centre, while new architectures for gallium nitride (GaN) are boosting the reliability of power systems while reducing the size and weight for robotics.
A key example is that Renesas is increasing its capacity by a factor of eight to meet growing demand over the next couple of years. “We believe GaN will break through with the transition next year and 2028 will be significant for us and our competitors,” Pietro Scalia, head of power system architecture at Renesas Electronics tells eeNews Europe.
Despite the different process technologies used for power and logic, integration is increasing in power devices.
Toshiba Electronics Europe has started shipping engineering samples of a device that integrates a microcontroller (MCU) with power MOSFETs for three-phase brushless DC (BLDC) motor drive, enabling direct control of small automotive motor systems operating below 40W.
This is aimed at automotive applications that need more compact three-phase BLDC motors to drive electric valves, HVAC dampers, small pumps, fans, and grille shutters. These applications require highly integrated solutions to reduce component count and shrink the size of the electronic control unit (ECU).
The AEC-Q100 TB9M040FTG integrates a 32-bit Arm Cortex-M23 core, a vector co-processor, flash memory, a three-phase BLDC motor driver with built-in MOSFETs, a 5 V power supply for optional components and sensors, and a LIN transceiver, all in a 6 mm × 6 mm VQFN36 package.
Toshiba’s Vector Engine (VE) co-processor accelerates field-oriented control (FOC) processing with very short FOC cycle times, reduces CPU load, and helps minimise software size. It also features back electromotive force (BEMF) detection, enabling sensorless square-wave control.
Integrated protection features include undervoltage, overvoltage, and overcurrent detection, as well as thermal shutdown, charge pump voltage monitoring, and drain-source voltage (Vds) detection for both high- and low-side MOSFETs.
The RAK-Gan evaluation platform accelerates development of efficient motor control and power conversion systems.
Rutronik has developed an integrated evaluation platform enables customers to efficiently test, validate, and optimize GaN-based motor control and power conversion concepts.
The RAK-GaN provides a practical development environment for applications where efficiency, power density, and system integration are key using GaN HEMTs.
Although individual GaN components are still more expensive than conventional silicon MOSFETs, they offer clear economic advantages at the system level: reduced cooling requirements, lower material usage, and simplified system architectures often lead to lower total system costs.
At the core of the platform is the Infineon PSOC Control microcontroller family based on an ARM Cortex-M33, which is optimized for digital power control and motor control.
Features such as high-resolution PWM (below 100 ps), fast analog-to-digital converters (12 MSPS) with synchronous sampling, and integrated high-speed comparators (less than 10 ns) enable precise implementation of highly dynamic control algorithms. This is complemented by hardware accelerators such as CORDIC and a low-latency trigger architecture, ensuring deterministic control even at very high switching frequencies.
In combination with the integrated GaN HEMTs, this opens up new possibilities for high-frequency switching concepts. The kit supports both 48 V three-phase BLDC/PMSM motor drives and DC-DC buck converters. At the same time, integrated protection features such as inrush current protection, hardware-based overcurrent detection, and precise current sensing using advanced TMR sensors from Infineon ensure safe and robust system design.
“GaN is a key enabler for the next generation of efficient power electronics. With the RAK-GaN, we not only provide our customers with access to this technology, but also with a well-thought-out system platform that allows innovative concepts to be quickly evaluated and translated into market-ready solutions,” says Stephan Menze, Head of Global Innovation Management at Rutronik.
Texas Instruments is showing advancements in battery health monitoring to unlock longer battery life, enhance safety, and prevent critical failures through predictive battery management, single-stage power conversion, and 48 V architectures. These devices help accelerate EV development and bring next-generation vehicles to market sooner.
GaN and SiC wide-bandgap technologies deliver the high efficiency and power density engineers need for bi-directional inverters, intelligent energy storage, and megawatt-scale EV charging. Paired with edge AI for real-time, low-overhead control, this helps engineers optimize the grid.
The power path from grid to processor is being reshaped with emerging 800 VDC architectures while maximising power density and conversion efficiency. At PCIM, Infineon will demonstrate its semiconductor technologies for efficient battery storage systems, uninterruptible power supplies, solid-state transformers (SSTs), and solid-state circuit breakers (SSCBs). A demonstration stack for SSTs, as well as SSCB components based on CoolSiC JFET technology, enables fault isolation within microseconds and delivers high robustness for future DC grids.
Rapidly rising AI computing workloads are driving a sharp increase in data centre energy demand and accelerating the shift toward new power architectures such as HVDC sidecars and DC microgrids down to processor cores.
Silicon carbide (SiC) is increasingly of interest for data centre power.
Toshiba has also started test-sample shipments of a 1200 V SiC MOSFET with a trench gate for power supply systems in AI data centres. Housed in a top-side-cooled QDPAK package, the surface-mount device delivers high current capability, improved heat dissipation, and higher power density in the power stage, which is essential for power conversion in AI data centres. The MOSFET is also suitable for renewable-energy equipment, including photovoltaic inverters, uninterruptible power supplies (UPS), EV charging stations, and energy storage systems.
The MOSFET uses a proprietary trench-gate structure, which achieves a low on-resistance per unit area (RDS(on)A). The device has a typical RDS(on) of 7.0 mΩ, a gate-drain charge (Qgd) of 33 nC, and a DC drain current (ID) of 172 A. Compared with Toshiba’s third-generation 1200 V SiC MOSFET, it halves the on-resistance and improves the figure of merit (RDS(on) × Qgd), which represents the trade-off between conduction loss and switching loss, by approximately 52%. The device also supports a low gate-drive voltage (Vgs-on) of 15 V to 18 V. These features enable highly efficient operation and reduced heat generation in data centre power supply systems, contributing to improved overall system performance.
With the rapid expansion of generative AI, increasing power consumption has become a pressing issue for data centres. In particular, the widespread adoption of high-power AI servers and the growing deployment of 800V HVDC (high voltage direct current) architectures are driving demand for power supply systems with higher power conversion efficiency and power density. The TW007D120E addresses these requirements, combining low conduction loss, low switching loss, and enhanced thermal performance to support more efficient and compact power system designs.
Toshiba is preparing for mass production later this year and is developing a version for automotive applications.
Two 3.3 kV SiC power module families from Wolfspeed provide high-power half-bridge and full bridge implementations in industry-standard footprints. These allow design engineers to reduce power stages and move to a 2-level topology for 2 kV and higher DC-link architectures with the choice of baseplate and baseplate-less SiC power modules.
“The release of this 3.3 kV MOSFET voltage node in two complementary footprints was a strategic decision,” said Guy Moxey, vice president of Wolfspeed’s Industrial & Energy business. “We understand the urgency our customers are facing to scale power infrastructure, and these two families enable both established grid-scale players and emerging players with modular architectures to move quickly.”
The high-power half-bridge baseplate SiC power module (LM platform) is designed for applications over 800 amps in converters for solar, grid-scale energy storage, and wind-power infrastructure. The module provides a 42% improvement in switching losses over other SiC modules and more than 90% over IGBTs, both measured at 125°C on a 1.8 kV bus in the same package.
The scalable full-bridge without a baseplate is part of the Wolfspeed WolfPACK family and is designed for modularity, offering flexibility to configure multi-level, series-stacked, or parallel converter architectures with consistent, matched performance — and is optimized for solid-state transformers (SSTs) and modular renewable energy infrastructure.
Both families are engineered for the relentless demands of always-on infrastructure. The WolfPACK module leverages cutting-edge sintered die attach and epoxy encapsulant material to deliver a significant improvement in power cycling performance over standard silicon gel encapsulated modules. Similarly, the baseplate module achieves improved system durability and power cycling by using sintered die attach and a copper die-top system.
This enables solid-state transformer systems to deliver over 50% footprint reduction compared to traditional equipment through improved switching performance and system architecture improvements.
Both families achieve improved switching over temperature, reducing magnetics and EMI filter sizes, ultimately leading to system power density and reduced system costs.
Navitas is also showing 3300 V, 2300 V, and 1200 V Trench Assisted Planar (TAP) SiC devices for AI data centres. These use a robust SiCPAK press-fit module, alongside recently announced 5th-generation GeneSiC TAP MOSFETs in QDPAK and TO247-LP packages.
These have been used for a 20 kW 800 V-to-6 V power delivery board aiming 97.5% peak efficiency, eliminating the traditional 48V intermediate bus converter (IBC) stage while enhancing overall system efficiency, reliability, cost-effectiveness, and power density.
For grid and energy infrastructure, Navitas will showcase two SST topologies enabled by Navitas GeneSiC UHV and HV technology:
A full SST cell developed with EPFL in Switzerland integrates the primary converter stage, transformer, and secondary conversion stage using a novel single-stage topology, leveraging Navitas 3300 V and 1200 V SiC technology. A 50 kVA bi-directional Active Front End uses the 3300 V SiCPAK MOSFET modules together with Texas Instruments’ C2000 real-time microcontrollers and UCC218915-Q1 gate drivers.
GaN is also being used to improve the efficiency of AI data centre power systems.
A 10 kW 800 V-to-50 V DC-DC platform features 2.1 kW/in³ power density and 98.5% peak efficiency, using the latest 650 V and 100 V GaNFast FETs from Navitas to deliver industry-leading efficiency, power density, and performance for 800 VDC and ±400 V power architectures.
At the same time ST Microelectronics is adding seven new GaN enhancement-mode transistors (HEMTs) to its 700V PowerGaN series aiming at AI servers. These have continuous currents that range from 6 A to 29 A, and typical RDS(on) from 53 mΩ to 270 mΩ as well as low internal capacitances and low gate charge, inherent in GaN wide-bandgap technology.
The devices are housed in DPAK, TO-LL, and PowerFLAT surface-mount packages with proicing from $0.63 to $2.25 for orders of 1000 pieces and are widely supported by major electronic design automation libraries and toolchains. The TO-LL and PowerFLAT devices provide a Kelvin source connection that separates the gate-control circuit from the main power path to maximize noise immunity, protect the gate driver, and preserve timing margin.
“Broadening our PowerGaN portfolio with new 700V devices extends the benefits of gallium-nitride technology into medium-power and high-power applications,” said Mario Aleo, Executive Vice President, Power & Discrete Sub-Group, STMicroelectronics. “We will continue to expand the portfolio with additional voltage ratings and features, reinforcing our commitment to GaN for tomorrow’s AI servers, humanoid robotics, industrial power, and advanced consumer power applications including home appliances.”
Power Integrations has also developed two ultra-slim, compact auxiliary power supply reference designs for 800 VDC AI data centres based on its 1700V GaN parts.
The single-output, 15 W design is only 30 mm by 30 mm with a 7 mm profile, while the isolated, six-rail, 35 W design is only 80 mm by 60 mm with an 8 mm profile. The designs are optimised specifically for the Nvidia Kyber liquid-cooled, blade-rack architecture, the designs free up approximately 30% space on densely packed main power distribution boards (PDBs) with an estimated 30% reduction in the BOM count—streamlining design and improving overall reliability. The designs are highly efficient with at least 88% efficiency across line and load.
The design examples describe 35 W and 15 W flyback auxiliary power supplies for internal components such as MCUs, gate drivers, and op-amps, which deliver critical control and housekeeping functions to ensure reliability, efficiency and system safety. These use the 1700 V-rated InnoMux-2 devices that easily support 1000 VDC nominal input voltage in a flyback configuration and can deliver flat efficiency of 90 percent in discontinuous conduction mode (DCM) while maximizing power delivery.
“As the only company offering single-HEMT 1700 V GaN devices, Power Integrations can design these best-in-class, highly efficient flyback converters with a low BOM count while maintaining wide safety margins on an 800 V bus,” said Jason Yan, Senior Training Manager at Power Integrations. “The only alternative solutions are discrete, costly silicon carbide (SiC) devices which require 30 percent more components and space to operate.”
Renesas Electronics has also introduced the industry’s first bidirectional switch using depletion-mode (d-mode) GaN technology, capable of blocking both positive and negative currents in a single device with DC blocking. This targets AI data centres and onboard electric vehicle chargers to simplify converter designs by replacing conventional back-to-back FET switches with a single low-loss, fast-switching, easy-to-drive device.
Current converter designs use unidirectional silicon or silicon carbide (SiC) switches, which block current in only one direction when in the off state. As a result, power conversion must be divided into stages with multiple switched-bridge circuits, creating a fourfold increase in switch count and reducing efficiency.
Bidirectional GaN changes this significantly. By integrating bidirectional blocking functionality on a single GaN product, power conversion can be achieved in a single stage using fewer switching devices. GaN devices also switch fast, with low stored charge, enabling higher switching frequencies and higher power density.
The 650 V SuperGaN devices are based on a proprietary normally-off d-mode technology developed by Transphorm that is simple to drive and highly robust. The TP65B110HRU combines a high-voltage bidirectional d-mode GaN chip co-packaged with two low-voltage silicon MOSFETs, with a high threshold voltage (3 V), high gate margin (±20 V), and built-in body diodes for efficient reverse conduction.
This is important for the sidecar racks that convert incoming 480V three phase AC power to 800V DC power. It is also used for the conversion from 800V direct to 6V at the processor.
AI rack evolution Source: Renesas Electronics
“When we go to bidirectional GaN in the AC-DC stage Vienna rectifier and this is where we see a lot of positive traction. Bidirectional GaN fits really well for this applications and you can save a lot of space and a lot of money as it is a quarter the size of SiC,” said Scalia.
Unlike enhancement-mode (e-mode) bidirectional GaN devices, the d-mode bidirectional GaN switch offers compatibility with standard gate drivers that require no negative gate bias. This results in a simpler, lower-cost gate-loop design and fast, stable switching in both soft- and hard-switching operation without a performance penalty. Power-conversion topologies that require hard switching, such as the Vienna-style rectifier, can benefit from its high dv/dt capability of >100 V/ns, with minimal ringing and short delays during on/off transitions. The Renesas GaN device enables true bidirectional switching with high robustness, high performance, and ease of use.
VisIC in Israel is showing its third generation d-mode GaN technology, D³GaN with lower switching losses than SiC and up to 99.6% peak efficiency. The 750V Gen 3 devices have an on resistance of 4 mΩ in an automotive-grade. While Gen2 had an on resistance as low as 4mΩ, Gen3 achieves the same low resistance with a smaller area, with a a 58% Rdson x A improvement from Gen 2 @150 C.
Robots are rapidly evolving toward physical AI systems that can sense, think and act. At PCIM, Infineon will showcase semiconductor solutions supporting this evolution across industrial and domestic robots, humanoids and drones. Demos include high‑efficiency motor control and power management solutions based on CoolGaN power semiconductors, PSOC Control C3 microcontrollers and XENSIV sensors, enabling compact designs, precise control and robust operation in future robotic applications.
At the same time motion platforms powered by EPC’s seventh-generation GaN power devices include the EPC33110 integrated three-phase ePower Stage IC for compact humanoid-joint and drone-propulsion control. In the power-conversion stage, EPC is showing a 40 V eGaN FET with an ultra-low 0.84 mΩ RDS(on), specifically designed for synchronous-rectifier applications on the secondary side of a 48 V-to-12 V LLC converter, and a 100 V device with a typical Rds(on) of 0.75 mΩ. It is also highlighting the EPC2304 (200 V, 3.5 mΩ) and EPC2305 (150 V, 2.2 mΩ) for high-power isolated DC-DC conversion, and the EPC2057 (50 V, 6 mΩ) for compact intermediate-bus regulation.
The latest Gen7 devices up to 40 V are packaged in 3.3 × 3.3 mm dual-cool QFN packages, while the 100 V devices come in 3 × 5 mm dual-cool QFN packages. These devices deliver benchmark RDS(on) and low gate charge for high-efficiency, high-power-density modules and DC-DC power supplies used in AI data centres, robotics, and motor-drive applications.
These devices are being demonstrated on different platforms at PCIM Europe for the development of humanoid robotics and drone propulsion. The EPC91122 and EPC91132 are motor-drive reference designs based on the EPC33110. The EPC91122 will be demonstrated in both a large drone-propulsion system and a robotic-arm platform to show its suitability for high-torque, high-power-density motion architectures.
A small drone-propulsion system designed for lightweight, space-constrained applications with rapid dynamic response will also be showcased, using devices that contain all the critical functional circuits needed to support a complete motor-drive inverter, including gate drivers, regulated auxiliary power rails for housekeeping supplies, voltage and temperature sensing, accurate current sensing, and protection functions.
Beyond robotics and drones, GaN technology enables smaller and more efficient electrified platforms such as power tools and e-bikes where higher switching frequency and smaller system size directly translate into better performance, lighter weight and longer run time. The same GaN technology platform enables scalable, high-efficiency power conversion for emerging AI infrastructure, supporting the move to distributed power architectures and higher-density computing environments.
“PCIM Europe is the ideal platform to demonstrate how GaN is driving a new generation of intelligent motion and AI platforms,” said Alex Lidow, CEO and co-founder of EPC. “Engineers are embedding power electronics into moving and thinking structures, whether those are humanoid robots, drones or AI servers. Our Gen 7 GaN technology enables designers to reduce size and weight, while improving efficiency and reliability – critical requirements for scaling these systems from prototype to production deployments.”
PCIM Europe 2026 is in Nuremberg, Germany, from 9th to 11th June.
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。