What Happens When the Ferrite Core Stops Being a Standard Shape?

The paper prompting this discussion is Potential of filament-based additive manufacturing for Ni–Zn ferrite: towards components with complex geometries, by Marwane Dherbécourt, Ghali Sqalli and colleagues, published in Progress in Additive Manufacturing on 8 September 2026. It demonstrates a manufacturing route and characterizes toroids. My interest is what its results allow us to design next, and where the RF evidence stops.

The printer shapes the precursor

The feedstock contains 43% ferrite by volume. After fused filament fabrication, solvent and thermal debinding remove the polymer; sintering produces a porous ceramic. The finished magnetic part is therefore fundamentally different from a ferrite-filled plastic print whose binder remains in service.

That distinction changes the engineering problem. Polymer spacing in an unsintered composite interrupts magnetic connectivity. In the ceramic route, particle bonding, grain growth and residual pores determine the magnetic response. A filament recipe cannot serve as the final material specification.

The study uses sintering temperatures of 1150–1300 °C. A desktop printer supplies one manufacturing stage; feedstock preparation, controlled binder removal, furnace processing, handling and inspection remain substantial work. Section 5.1 lists 1, 2 and 3 hours, but Table 3 and the magnetic plots identify 1, 2 and 4 hours. I use the table’s conditions when interpreting the samples.

There is a useful manufacturing opportunity here: a custom part can be changed without commissioning a new pressing die. That can matter greatly for development or short runs. It does not establish cheaper production per qualified component. Furnace occupancy, failed parts, dimensional finishing and inspection all belong in that calculation.

What the magnetic figures actually establish

Section 6.4 measures wound toroids with nine copper turns and a Keysight 4294 impedance analyzer over 0.1–100 MHz. Table 3 spans μ′ = 27–132 at 1 MHz; the abstract’s 80–130 describes the more successful processing region, not every specimen.

Figure 22 shows higher low-frequency permeability for the hotter four-hour treatments, followed by pronounced dispersion. The sample with the highest starting permeability does not preserve that advantage across the entire sweep. Choosing the largest left-hand value misses the frequency-dependent design problem.

In Figure 24, the printed validation core and 4F1 reference track reasonably closely at lower frequencies. Their peaks, roll-offs and loss curves visibly separate at higher frequencies. The reported approximately 52 MHz relaxation feature is not a low-loss operating-frequency rating, and the plot does not demonstrate interchangeable broadband behavior.

The manufacturer’s 4F1 specification helps put that comparison in context: initial permeability is approximately 80 at 25 °C, no more than 10 kHz and 0.25 mT. Its stated power-transformer optimization is 4–10 MHz. Those are different quantities and conditions from a relaxation frequency. Neither its large-signal loss data nor its magnetic-field limits transfer to the printed material merely because the small-signal plateaus resemble each other.

I would call this evidence of a useful magnetic ceramic with a promising reference comparison. I would not substitute it for 4F1 in a qualified design without further tests. It gives no basis for substituting for mixes 31 or 43, either: chemistry, microstructure and frequency response must each be established.

Permeability becomes impedance through geometry

My earlier toroid-selection discussion makes the same distinction: permeability describes part of the design. For a closed, approximately uniform magnetic path in the linear regime, define effective cross-section A_e, path length l_e, turn count N and angular frequency ω = 2πf. With the e^jωt convention:

Thus μ′ contributes inductive reactance and μ″ contributes equivalent series resistance. Here μ_r is the homogenized response of the processed ceramic, including its microstructure; the equation does not remove its pores. Winding resistance, capacitance, leakage flux and fixture effects still have to be added. Above a winding resonance, interpreting everything through this simple magnetic branch can be misleading.

Consider an independent illustration using hypothetical finished dimensions: A_e = 25 mm², l_e = 50 mm, N = 5, μ′ = 80 and μ″ = 8 at 1 MHz. The approximation gives L′ ≈ 1.26 μH and Z_mag ≈ 0.79 + j7.90 Ω. These are calculated teaching values, not measurements of the paper’s cores or an RF.Guru product.

Doubling cross-section at the same path length and permeability doubles both magnetic terms. Doubling turns multiplies them by four only while the approximation holds; it also changes copper length and capacitance. A custom shape might achieve the required impedance with fewer turns and better conductor spacing. Whether that extends the useful band depends on the complete winding.

There is another measurement issue here. The paper describes the instrument and winding but does not provide a complete fixture compensation, parasitic extraction and uncertainty account. Its curves are valuable evidence; I cannot independently separate all winding effects from material dispersion near their upper-frequency features. Repeating the extraction with different turn counts would be a useful check.

A closed toroid is a favorable magnetic test shape

A toroid provides a nearly closed flux path. Change it into a short rod, a thin open structure or a core with a gap and the winding no longer sees the same effective response, even if the ceramic itself is unchanged.

For a uniformly magnetized ellipsoid, a useful magnetostatic illustration is the apparent susceptibility relative to the externally applied field:

D is the demagnetizing factor along the applied field. An approximately closed path has little demagnetizing penalty; an open shape can have a substantial one. With a real μ_r = 80 and an illustrative D = 0.1, 1 + χ_app is only about 9.88. That number is a shape illustration, not a universal rod-coil permeability formula. Nonuniform fields, partial winding coverage and complex permeability require a more complete model.

For a simple uniform core with a small deliberate gap g, neglecting fringing:

A gap lowers inductance and can improve bias tolerance by increasing reluctance. With the preceding illustrative 50 mm path and μ_r = 80, a 0.5 mm gap reduces the effective permeability to about 44.4. Fringing may increase nearby winding loss. A gap that helps an energy-storage inductor can be an unwanted reduction in a choke’s suppression impedance.

The opportunity is to design cross-section, winding space and flux path together. Narrow necks concentrate flux; sharp transitions encourage nonuniform fields; larger windows can improve insulation clearance while lengthening the magnetic path. The paper’s printed and sintered Benchy demonstrates shape-making capability. Purpose-designed RF flux paths, controlled functional gaps and improved antennas remain opportunities to investigate.

Porosity is part of the electrical design

The best reported printed densification is 88.6%, compared with 97.0% for the pressed control. The validation specimen is 83.9%. Figure 16 identifies roughly 30 μm inter-layer voids in C₈.

A pore is a region of high magnetic reluctance, but its effect depends on position and orientation relative to flux. Concentric deposition can look favorable under circumferential toroid excitation without proving the same behavior across printed layers. I would treat orientation-dependent permeability and loss as open questions before making three-dimensional flux paths. Similar grain shapes and broadly similar dimensional shrinkage do not establish magnetic isotropy.

Open porosity and total missing density are also different measurements. A small open-pore fraction does not mean that all remaining pores have disappeared. Closed pores and the distribution of voids still influence flux and mechanical behavior.

The replication comparison is encouraging, but small: C₈ and its repeat C_v give μ′ = 78 and 84 at 1 MHz. Their difference is about 7.7% relative to C₈. One repeat supports feasibility; it cannot establish lot tolerance, orientation independence or a guaranteed minimum. This is exactly the distinction in my earlier ferrite-tolerance article: geometry, low-level magnetic response, RF impedance and production spread need separate controls.

Shrinkage needs a dimensional model

Equation 3 labels V_s/V_g as shrinkage, although that ratio is the retained volume fraction. Conventional volume reduction is 1 − V_s/V_g. The table and discussion appear to use volume reduction, but not every entry reconciles with the tabulated axial and radial contractions.

For example, assuming the inner and outer diameters contract by the same radial fraction, C₈ gives 1 − (1 − 0.219)²(1 − 0.232) = 53.2%, close to its tabulated 52.8%. C₁₁ gives 54.8% from the same calculation, versus a tabulated 61.0%. The source does not resolve that difference. Different inner/outer contraction could invalidate the simplifying assumption; it would then require separate final dimensions.

I therefore would not derive a precise CAD enlargement factor from the headline volume number. Measure final inner diameter, outer diameter, height, gap width and mating faces for the selected process. A hole that closes too far affects the winding; unequal contraction changes A_e/l_e; a warped mating face creates an unintended gap. The required geometry is the fired geometry.

For a choke, useful loss still produces heat

A common-mode choke can benefit from magnetic loss because the resulting resistance damps unwanted current. A low-loss RF inductor has the opposite priority. In the simple magnetic branch, Q ≈ μ′/μ″; the imaginary part that helps a suppressor lowers an inductor’s Q. The application decides whether that loss is useful.

For the choke’s lossy branch, P ≈ I_CM,rms²R_CM at a stated frequency and operating point. That relation uses the current actually flowing through that branch. Increasing choke resistance usually changes the installed current, so a fixed-current comparison is not an installed thermal prediction. Parasitic current paths also matter.

My two-port measurement article separates three questions: the finished component’s complex impedance, the installed current distribution, and powered thermal performance. A printed core needs the same separation. For an ideal series impedance between equal real port impedances Z₀:

The fixture’s source and load determine that transfer. An antenna installation does not inherit its suppression figure. Under an appropriate calibrated π-network model, extracting Z_series = −1/Y₂₁ is another route to complex component data; neither operation turns a permeability curve into a universal rejection number.

A custom core could provide useful winding separation, rounded cable passages or a better thermal mounting surface. Those are design proposals. The paper does not test such a choke or establish an RF.Guru release, replacement core or QRO rating.

Small-signal similarity leaves the power question open

For sinusoidal voltage excitation of a reasonably uniform transformer core, peak flux is approximately B_pk = V_rms/(4.44fNA_e). For a current-driven closed path, H is approximately NI/l_e. These relations explain why adding cross-section can relieve voltage-driven flux stress, while cross-section alone does not reduce the applied field at fixed ampere-turns and path length.

Initial permeability supplies neither a saturation curve nor loss versus flux swing. DC bias changes the operating point and incremental permeability; temperature changes both magnetic response and heat removal. The paper provides no measured saturation, DC-bias or powered loss/temperature map sufficient to qualify an RF power component.

There is a mechanical consequence too. A porous ceramic can develop local thermal gradients around loss concentrations. Differential expansion between ceramic, winding, coating and mount can introduce stress or cracks. Larger thermal contact area may help cooling, while a rigid attachment may worsen stress. Those competing effects require thermal cycling and electrical checks on the assembled component.

Compact antennas still have an efficiency budget

A tailored core might make an RF loading inductor fit a restricted enclosure, reduce an awkward winding or concentrate magnetic material where it is useful. For an open magnetic antenna structure, demagnetization and field orientation become central. The toroid measurements alone cannot predict either antenna’s performance.

Loading can bring a short antenna to resonance. Resonance only cancels net input reactance at a frequency; it does not establish radiation efficiency. Refer all equivalent resistances to the same antenna current:

When radiation resistance is small, even modest added core loss can consume much of the accepted power. Extra dissipation can broaden a matched response while reducing useful radiation. A wider SWR curve is therefore insufficient evidence for a better compact antenna.

The classical Chu size-and-bandwidth analysis concerns the fields outside an enclosing region. Changing the core’s internal shape does not remove those external fields or the associated stored-energy constraint. Magnetic loading can improve a particular realizable design; it does not provide unlimited bandwidth and efficiency at fixed electrical size. Any comparison must include the full radiating structure, including feed conductors if they participate, and measure efficiency as well as match.

The design freedom I would pursue

I would first use this process where a conventional shape imposes an identifiable penalty: an excessively long winding, a poor cable bend, inadequate clearance or an inconvenient flux path. Then I would compare the custom ceramic against a conventional core in the same functional envelope, with the same electrical target and declared operating conditions.

The important comparison is whether the complete component achieves lower loss, a more useful impedance band, better thermal behavior or a mechanical arrangement that was previously impractical. It should include multiple builds, final dimensions, orientation, complex response, bias and powered temperature behavior. Shape freedom earns its value by solving one of those problems.

Dherbécourt and colleagues have made a credible step toward that freedom. My conclusion is that custom ferrite geometry deserves serious RF development work. The next advance will come from controlling the ceramic and the winding together, then demonstrating a useful component advantage.

Questions, antenna-factor records or height trials to share? Contact RF.Guru.