TECHNICAL ARTICLE
Under the official 80 PLUS certification criteria, Titanium requires 230V/277V redundant data center power supplies to reach 96% efficiency at 50% load. The newer Ruby tier raises the requirement to 96.5%, or 97% for the specified higher-voltage AC/DC input classes. As GaN and SiC devices continue to reduce switching losses, the losses that remain—core losses in transformers and inductors, AC copper losses, leakage inductance spikes, parasitic capacitance, insulation stress, and thermal rise—increasingly determine how far high-frequency design can go.
| Core thesis: High-frequency operation shrinks magnetic components, but it does not create power density for free. Once wide-bandgap semiconductors compress the loss budget on their side, magnetics stop being a commodity accessory and become the shared boundary of topology, control, and thermal design. The next generation of server power supplies is not just a GaN or SiC swap—it is a co-design of magnetic materials, windings, planar structures, and digital control. |
Table of Contents
Industry Overview: The Efficiency Race in AI Data Centers
Six Mechanisms That Make Magnetics the Constraint
Where the Value Lands: Materials, Structures, and Systems
Engineering Practice: What This Means for Magnetic Component Design
What to Watch Next
FAQ
1. Industry Overview: The Efficiency Race in AI Data Centers
Server power supplies must deliver more power in less space—within a 1U rack unit or compact power modules. AI accelerators keep pushing per-rack power upward: the Open Rack V3 specification now includes a 12kW PSU module, while next-generation systems are moving toward higher-power three-phase shelves. Power density and heat flux rise in lockstep.
1.1 The Efficiency Bar Keeps Rising
| Certification | Condition | Efficiency Requirement |
|---|---|---|
| 80 PLUS Titanium | 230V/277V internal redundant PSU, 50% load | 96% |
| 80 PLUS Ruby | 230V/277V internal redundant PSU, 50% load | 96.5% |
| 80 PLUS Ruby | 380V/800V DC or 400V/480V AC internal redundant PSU, 50% load | 97% |
The closer efficiency gets to its ceiling, the harder it becomes to compensate for every watt of magnetic loss elsewhere in the power train.
1.2 Higher Frequency Is the Path to Power Density
Raising switching frequency reduces the volt-seconds each cycle must handle, shrinking the required core cross-section. A transformer designed for 500kHz can be significantly smaller than its 100kHz equivalent. TI’s PMP23146 server auxiliary power reference design uses a planar transformer at 400kHz, illustrating how frequency and planar magnetics can work together to compress volume.
But higher frequency also pushes core, winding, insulation, and parasitic parameters into the dominant zone. The design target is no longer “the highest frequency”—it is the system optimum across magnetic volume, semiconductor losses, passive losses, EMI, thermal rise, and cost.
2. Six Mechanisms That Make Magnetics the Constraint
2.1 Core Loss Scales with Frequency and Flux Swing Together
Higher frequency allows a smaller core cross-section, but hysteresis and eddy-current losses climb. Ferrites, nanocrystalline alloys, and metal powder cores each behave differently across frequency, temperature, DC bias, and waveform shape. Datasheet typical values cannot substitute for testing under real converter waveforms.
2.2 AC Copper Loss Is Not DC Resistance
Skin effect concentrates current at the conductor surface; proximity effect distorts current distribution in adjacent windings. At high frequency and high current, the effective resistance of copper foil, flat wire, and PCB windings rises dramatically. Layer arrangement matters more than simply adding copper thickness.
2.3 Leakage Inductance and Parasitic Capacitance Pull in Opposite Directions
Reducing primary-to-secondary spacing lowers leakage inductance but increases interwinding parasitic capacitance and common-mode noise. Increasing insulation distance improves safety but raises leakage inductance and volume. Planar transformers must simultaneously optimize layer stacking, shielding, and winding interleaving.
2.4 LLC and DAB Topologies Demand Controllable Magnetic Parameters
In resonant LLC converters, magnetizing inductance and resonant inductance set the gain curve, soft-switching range, and transient response. In dual-active-bridge (DAB) converters, leakage inductance itself becomes an energy-transfer element. Parameter variation directly impacts efficiency and current sharing across paralleled units.
2.5 High Power Density Hides Hotspots Inside the Magnetic Component
Internal thermal resistance in cores and windings is high—a normal exterior temperature does not rule out internal hotspots. Potting, clamps, busbars, and cold plates reshape the thermal path. Thermal design must be verified together with electromagnetic simulation, material lifetime, and insulation class.
2.6 Production Consistency Determines Whether Lab Designs Scale
Air-gap dimensions, winding placement, copper thickness, lamination, magnetic material batches, and assembly pressure all shift inductance and loss. Automated winding, PCB-based planar magnetics, and in-circuit testing earn their keep by converting the laboratory optimum into a stable manufacturing process window.
3. Where the Value Lands: Materials, Structures, and Systems
| Segment | Core Value | Key Requirements |
|---|---|---|
| High-frequency, low-loss magnetic materials | Low loss and stable permeability at target frequency, temperature, and DC bias | Consistency beats single-point peak specs |
| Planar transformers and inductors | Low profile and automation-friendly manufacturing | Layer stacking, parasitics, insulation, and thermal design complexity |
| High-frequency winding materials | Controlling current distribution to reduce AC losses | Litz wire, flat wire, copper foil, multilayer PCB—cost and process depend on power and frequency |
| Simulation and test equipment | Closing the data loop from material to structure to system | Impedance analysis, core-loss measurement, partial discharge, thermal imaging, EMI, power analysis |
| Digital control | Reducing device and magnetic losses through control | Soft switching, synchronous rectification, frequency and phase-shift modulation, parameter compensation |
| Power supply OEMs | Co-optimizing magnetics with GaN/SiC, topology, and thermal management | Multi-load-point efficiency, power density, and reliability as the final deliverables |
4. Engineering Practice: What This Means for Magnetic Component Design
For magnetics suppliers and power supply engineers, the shift is concrete:
Material selection must be waveform-verified. Loss curves measured with sinusoidal excitation understate losses under the square-wave, quasi-square, or resonant waveforms found in LLC and DAB stages. Ask for loss data under your actual operating waveform, flux swing, and temperature.
Winding design is an AC problem. Dowell-based layer analysis, interleaving, and litz-wire strand-diameter selection (e.g., 0.05–0.1mm strands sized to the skin depth at your switching frequency) should be part of the quotation package, not an afterthought.
Parasitics are design parameters, not defects. In DAB designs, leakage inductance can be deliberately engineered as the energy-transfer element; in LLC, the ratio of magnetizing to resonant inductance defines the operating region. Specify tolerances and verify them across production batches.
Thermal validation must reach the hotspot. Rely on thermocouple-embedded prototypes or thermal imaging of cross-sectioned samples—not just surface temperature—to validate winding hotspots and insulation margins.
Consistency is a deliverable. Documented process windows for air-gap assembly, winding tension, and core batch traceability are what allow a validated 96.5% design to ship at volume.
4.1 How TrafoPSU Approaches This
As a custom magnetic components manufacturer, TrafoPSU designs and produces high-frequency transformers, planar transformers, resonant inductors, and common-mode chokes for power conversion applications from 50W to multi-kilowatt. Our engineering workflow reflects the constraints above:
Core material portfolio spanning high-frequency ferrites through nanocrystalline alloys, selected against measured loss behavior at the customer’s operating frequency and temperature
Winding technologies including litz wire, flat wire, copper foil, and multilayer PCB planar structures, with AC resistance analyzed and verified at the design frequency
Topology-aware parameter control for LLC resonant tanks and DAB leakage-integrated designs, with defined production tolerances. See our integrated LLC resonant transformer design guide for a worked design example
EMI-side components—common-mode chokes engineered for target impedance bands and self-resonant frequency placement in high-frequency systems
Explore our planar transformers, high-frequency transformers, and inductors and common-mode chokes for data center and server power applications.
5. What to Watch Next
The count of 80 PLUS Ruby certifications across 230V, 380V, and 800V data center power supplies
Switching frequency, topology, and magnetics volume trends in multi-kilowatt and three-phase server power systems
Production adoption of planar transformers in main power stages (not just auxiliary supplies)
Loss curves and batch-to-batch consistency of high-frequency ferrites, nanocrystalline, and metal powder cores
How LLC magnetizing/resonant inductance and leakage tolerances move whole-unit efficiency
Long-term data on internal hotspots, partial discharge, and insulation lifetime in compact magnetics
Magnetic component makers’ data center revenue, customer certifications, and automated capacity expansion
The share of magnetics in server PSU BOM cost and R&D cycles as GaN/SiC losses fall
| Bottom line: The closer a server power supply gets to the efficiency limit, the less magnetic components are a deferred commodity purchase—and the more they become the core design variable that determines high-frequency capability, power density, and reliability. |
| Designing a High-Frequency Server Power Stage?TrafoPSU works with power supply OEMs to co-design transformers, resonant inductors, and EMI chokes for LLC, DAB, and PFC stages—from material selection and AC-loss-optimized windings to production-tolerance control. Send us your specifications:sales@trafopsu.com · Request a Quotation |
FAQ
Why do magnetic components limit server power supply efficiency?
After GaN and SiC devices reduced switching losses, core losses (hysteresis and eddy currents), AC copper losses from skin and proximity effects, leakage inductance spikes, and parasitic capacitance now dominate the remaining loss budget. At 80 PLUS Titanium (96% at 50% load) and Ruby (96.5–97%) levels, every watt of magnetic loss is difficult to compensate elsewhere.
What is the 80 PLUS Ruby efficiency requirement?
For redundant data center PSUs, 80 PLUS Titanium requires 96% efficiency at 50% load for 230V/277V internal units. Ruby raises this to 96.5%, while the specified 380V/800V DC and 400V/480V AC input classes require 97% at 50% load.
Why are planar transformers preferred for high-frequency server power supplies?
Planar transformers use PCB or copper-foil windings that enable low-profile designs and automated manufacturing, with better thermal spreading and predictable parasitics than wire-wound equivalents—though they demand careful optimization of stacking, shielding, insulation, and thermal paths.
How does switching frequency affect transformer size?
Higher switching frequency reduces volt-seconds per cycle, allowing smaller core cross-sections—a 500kHz transformer can be significantly smaller than a 100kHz design. However, core losses, AC winding resistance, EMI, and thermal rise all increase with frequency, so the goal is a system optimum, not the highest frequency.
Topics: Server Power Supply | Magnetic Components | High-Frequency Transformer | Planar Transformer | LLC Resonant | DAB Converter | GaN / SiC | Core Loss | 80 PLUS Titanium | Data Center Power
