FeNi Powder Cores for Server Power Supplies

A server power supply is not one converter. It is a chain: an AC front end with power factor correction, an isolated DC-DC stage that provides safety isolation and steps the voltage down, and a point-of-load layer that feeds processors, memory and networking at low voltage and high current. Each boundary asks something different of its magnetic components.

That matters for core selection, because the material that suits the isolated transformer is often a poor choice for a high-current output inductor. Treating the supply as a single design problem tends to produce a compromise that is wrong at both ends.

Where magnetic components sit in a server PSU

In a typical architecture the PFC stage corrects input current and produces a high-voltage DC bus. An isolated DC-DC converter then transforms that bus to the rack distribution rail, and downstream converters step from the rail to the voltages the load actually needs. Magnetics appear at every one of those boundaries: the PFC inductor, the isolated transformer, the resonant or output inductor, and filter inductors on the downstream rails.

What the high-current stages demand

Output and intermediate inductors carry continuous current with a comparatively small ripple superimposed on it. The magnetic design is therefore dominated by DC bias behavior rather than by AC loss alone. What decides the part is how much inductance remains at full load, how much heat the core produces under sustained current, and how much board area the answer occupies.

• Continuous DC current dominates the flux state of the core
• Inductance at full load, not nominal inductance, sets ripple and loop stability
• Sustained load means thermal rise accumulates rather than averaging out
• Rack airflow is shared with processors, so local ambient is above intake temperature
• Long service life puts weight on thermal margin rather than peak efficiency

Why FeNi powder cores are considered here

FeNi powder cores are pressed from insulated alloy particles, which creates a distributed air gap throughout the magnetic path rather than a single physical gap. For a high-current inductor the practical consequence is a gradual, predictable inductance roll-off as bias increases, instead of a sharp saturation knee.

Two effects follow. The converter keeps a predictable ripple characteristic at full load, and the core avoids the fringing flux that a discrete gap produces — useful when the inductor sits close to sensitive traces on a dense board.

Where ferrite still fits better

This is not an argument that powder cores replace ferrite in a server PSU. The isolated DC-DC stage usually switches at high frequency, where low-loss ferrite grades keep efficiency high and material cost low. Ferrite also suits the transformer function itself, which needs high permeability and tight coupling rather than high stored energy per unit volume.

The practical split is by function: ferrite for high-frequency transformation and low-power high-frequency filtering, powder cores for the higher-current energy-storage and filtering roles where bias stability sets the size.

Engineering requirements to define first

• Switching frequency of the specific converter stage
• Continuous current and worst-case peak current
• Required inductance at full bias, not at zero current
• Ripple current amplitude and waveform shape
• Local ambient temperature and the allowed temperature rise
• Available board area and component height
• Whether the inductor is a discrete part or integrated into a module

Design considerations in dense racks

A server enclosure is a harsher thermal environment than the datasheet ambient suggests. Air is preheated by upstream components, airflow per component is limited, and the same chassis must support several load profiles. Two consequences follow for magnetics.

First, the inductor should be evaluated at its hot-spot temperature rather than at intake temperature. Second, because thermal margin is shared across the board, a part that looks comfortable at nominal load may still be the wrong choice if it has no headroom at sustained maximum. Measuring a prototype at the real operating point is more informative than extrapolating from a single datasheet figure.

Application examples

• Rack power shelves feeding multiple servers from a shared distribution rail
• Intermediate bus converters between a distribution rail and downstream low-voltage rails
• Point-of-load converters close to processors, where board area is the binding constraint
• Redundant supply paths, where both converters run below maximum but never rest

Selection checklist

• Separate the transformer function from the energy-storage function before choosing material
• Specify inductance at the real bias current and real temperature
• Compare candidate cores at the same operating point, not by part number
• Check hot-spot temperature at sustained load, not only nominal load
• Confirm the core geometry fits the available footprint and height
• Validate with a prototype measurement before committing to tooling

Custom development for server power

Server power designs rarely settle on an off-the-shelf core. Current level, height limit and thermal budget usually combine into a geometry and permeability grade that has to be developed for the application. That work covers core geometry, permeability selection and the wound inductor around it, followed by prototype evaluation and production scale-up.

For the high-frequency end of the same problem, see Choosing Inductors for AI GPU VRM and the High Current Power Inductor Design guide. The High Flux Cores Guide covers FeNi grade and series data.

Frequently Asked Questions

What type of magnetic core is used in server power supplies?

Several types, because a server supply contains several converter stages. High-frequency isolated stages typically use low-loss ferrite grades, while high-current output and filtering inductors are candidates for powder cores such as FeNi, where inductance must hold under sustained DC bias.

Why are powder cores used in high-current inductors?

Powder cores contain a distributed air gap formed by insulated particles rather than a single physical gap. That produces a gradual inductance roll-off under DC bias instead of a sharp saturation knee, which helps the inductor hold its value at full load.

Can FeNi powder cores replace ferrite in a server PSU?

Not across the whole supply. The materials suit different functions. Ferrite is generally preferred for high-frequency transformation and low-loss high-frequency filtering, while FeNi powder cores are considered for higher-current energy-storage and filtering roles where bias stability sets the size.

What should be specified when requesting a server power inductor?

Switching frequency of the stage, continuous and peak current, required inductance at full bias, ripple current and waveform, local ambient and allowed temperature rise, and available board area and height. Without the bias point, quoted inductance values are not comparable between suppliers.

How does rack thermal design affect magnetic component selection?

Air inside a server enclosure is preheated by upstream components, so the local ambient at the inductor is usually higher than intake temperature. The inductor should be evaluated at its hot-spot temperature under sustained load, because thermal margin is shared with the rest of the board.