On the Eve of SST Mass Production: Three Barriers Facing High-Frequency Transformers

Industry Insights · August 2026

Between announcing a product and shipping it in volume, what usually stands in the way is neither the topology nor the control algorithm — it is a handful of unremarkable-looking magnetic components.

Solid-state transformer (SST) industrialisation has accelerated noticeably through 2026. According to a power equipment research report published by Guosheng Securities on 19 July, more than ten major companies released SST products in the first half of 2026 alone. Delta, Sinephro, TGOOD and Eaton have each brought forward their own solutions, some already in production with annual capacity exceeding 200 units. BloombergNEF projects the global SST market will reach RMB 120 billion by 2030.

The force driving this acceleration is compute. NVIDIA has confirmed deployment of an 800V data centre power architecture in 2027, moving SST from proof-of-concept into production readiness. Research from Huaan Securities indicates an 800V DC architecture can cut outside-rack electrical equipment cost by 75% and reduce distribution losses by 53% — a return large enough to justify redesigning the entire chain around it.

But an industry inflection point is not the same as deployment readiness. What actually constrains the pace is cost and reliability. And on the reliability side, the pressure concentrates heavily on magnetics.

1. High-frequency operation means replacing line-frequency magnetics

The fundamental difference between an SST and a conventional transformer is not that it is 'smart' — it is that power conversion moves from 50Hz line frequency up to 20kHz and beyond. The immediate benefit is a substantial reduction in magnetic component volume, which is the physical basis for the power density SST promises.

Put differently: high-frequency operation in an SST means swapping bulky line-frequency magnetics for a set of components operating at high frequency. System integrators compete on topology, control and integration — but efficiency and volume, the two metrics that ultimately decide the outcome, are settled at the magnetics level.

The complication is that once frequency rises by two to three orders of magnitude, secondary effects that were safely ignored at line frequency each become a first-order engineering problem. The industry currently converges on three: insulation reliability, high-frequency loss suppression, and parameter consistency.

What materials do high-frequency magnetics use: the parameter boundaries of four soft-magnetic materials

Replacing line-frequency magnetics with high-frequency ones starts with re-selecting the material. SST operating frequency typically sits in the 10kHz–100kHz range — a band that rules out conventional silicon steel and pushes ferrite, amorphous and nanocrystalline to the front. Their core differences show up in three figures: saturation flux density, usable frequency and loss.

The point of this table is not the numbers themselves, but the constraint they encode: saturation flux density decides how small the core can be, usable frequency decides whether the material is even viable, and loss decides the thermal margin. Ferrite's low HF loss comes at the cost of Bs of only 0.3–0.5T — meaning at the same power the core must be larger, or the operating flux density must be pushed well below saturation, effectively giving up part of the volume advantage. Nanocrystalline, with higher Bs and very low mid-HF loss, has become one of the mainstream choices for SST HF high-power transformers; the trade-off is higher demands on processing and cost.

2. Barrier one: insulation reliability

The medium-voltage side of an SST interfaces directly with a distribution network in the 10kV class, meaning the high-frequency transformer must provide galvanic isolation across a very large potential difference. On its own this is nothing new — line-frequency transformers have handled it for over a century.

The difficulty comes from three requirements holding simultaneously: high potential difference, high operating frequency, and small volume.

High-frequency operation materially changes the failure modes of an insulation system. A structure that runs stably for years at line frequency may exhibit sustained partial discharge above 20kHz, driven by the higher rate of change in the electric field. Partial discharge does not cause immediate breakdown; it erodes the insulation slowly, so a unit that passes every factory test can still fail after several thousand hours in service. That failure mode is difficult to expose in short validation cycles — and it is precisely the mode data centre and grid applications can least tolerate.

At the same time, the power density SST targets requires insulation distances to be compressed as far as possible. Every millimetre removed from creepage and clearance eats into the safety margin. Residual voids in the potting process, the thermal class of the insulating material, the handling of turn-to-turn and layer-to-layer insulation — any compromise at any of these points is magnified at high frequency.

3. Barrier two: loss at high frequency and high power

As frequency rises, core loss and AC winding loss both increase. This is physics; it cannot be avoided, only optimised through material and structural design.

On the core side

Core material selection sets the baseline. Ferrite performs well on loss at high frequency but has a relatively low saturation flux density. Nanocrystalline alloys offer higher saturation flux density and suit high-power applications, but demand more from processing and cost control. Worth noting: the high-frequency soft-magnetic nanocrystalline alloy developed by Academician Wang Weihua's team at the CAS Dongguan Institute of Materials entered volume delivery in 2026 and has been designed into the supply chains of Samsung, BYD, V-MAX, Delta and Lite-On — breaking a technology monopoly previously held by overseas suppliers. For domestic SST magnetics, that is a material improvement in availability.

Putting loss into numbers makes it concrete: for nanocrystalline, public datasheets quote typical specific loss of roughly ≤20 W/kg at 20kHz / 0.5T and ≤50 W/kg at 100kHz / 0.2T; by contrast, ferrite tends toward saturation failure at 0.5T flux density (some grades no longer function normally at 0.5T). That forces the ferrite in a HF transformer to operate well below saturation, partially offsetting its volume advantage. Material loss is therefore not a one-time selection — it is a continuous trade-off among Bs, frequency and loss.

On the winding side

Skin effect concentrates current at the conductor surface, shrinking the effective conducting cross-section as frequency climbs. Proximity effect further distorts current distribution between adjacent conductors. Together they can push AC resistance far above DC resistance, driving loss up. A physical order of magnitude makes it tangible: copper skin depth δ ≈ 66/√f (mm), about 0.21mm at 100kHz and 0.066mm at 1MHz. In other words, once frequency enters the hundreds of kHz, the effective conducting thickness of a single solid conductor is already down to a fraction of a millimetre — which is why multi-strand Litz wire or thin copper foil must be used to spread the AC loss. Mitigation comes chiefly from conductor form — Litz stranding, copper foil, PCB copper layers — and from interleaved winding arrangements.

On the thermal side

Worse, rising power density simultaneously compresses the physical space available for heat removal. Loss increases while cooling gets harder — two trends pulling in opposite directions, compounding the thermal design difficulty. This surfaces the hard constraint of temperature rise: SST HF transformers typically must meet an insulation class — per IEC 60085, common classes are B 130°C, F 155°C and H 180°C (maximum hotspot temperature for the winding). At high power density, the magnetics' own temperature rise often approaches or even reaches these limits, so thermal design stops being a soft 'leave some margin' item and becomes a hard gate that decides whether a product passes certification.

It is worth adding that accurately measuring loss is itself a barrier. Per BigBit research, the global power magnetics market reached RMB 155 billion in 2025 and is expected to exceed RMB 170 billion in 2026. Yet testing capability lags on three fronts: 1MHz is now a mainstream requirement with 7MHz evaluation emerging; power test levels have climbed from an early 600W to 5kW as standard, with 20kW entering real use; and combined AC-plus-DC-bias testing has long been a gap — conventional instruments measure either AC loss or DC characteristics, and cannot reproduce the composite condition magnetics actually experience in service, which distorts loss evaluation.

Put simply: if you cannot measure it accurately, you cannot optimise it. This is the technical root of loss problems that stay hidden through prototyping and surface all at once in volume production.

4. Barrier three: parameter consistency

This is the real dividing line between a working prototype and a manufacturable product.

SST designs commonly use LLC and CLLC resonant topologies, which are acutely sensitive to magnetic parameters. Leakage inductance and magnetising inductance both enter directly into the resonant point calculation. Once batch-to-batch variation appears, the resonant point drifts, system efficiency fluctuates, and in severe cases the conditions for soft switching are compromised.

Building one prototype allows tuning toward the ideal value. Building a thousand — or ten thousand — depends not on tuning but on process capability. Variation enters at multiple points along the chain:

• Incoming cores — permeability varies naturally between batches of the same grade, directly affecting inductance
• Winding process — repeatability of winding position, turn spacing and layer spacing determines leakage inductance consistency
• Air-gap machining — small dimensional tolerances are amplified several times over in the resulting inductance
• Assembly — core pairing, clamping force and adhesive layer thickness all introduce variables

There is a clear industry consensus on this point: under SST high-frequency, high-power conditions, the three risks of core loss, leakage inductance deviation and core saturation are amplified exponentially. Batch variation that was tolerable in the line-frequency era translates here into measurable system efficiency loss.

5. The response: planar transformers and magnetic integration

Two main technical routes are being pursued against these barriers.

First, planar transformers. PCB copper traces replace conventional Litz windings, so winding geometry is set by PCB manufacturing tolerance — improving consistency by an order of magnitude. A flattened core also lowers overall height, suiting space-constrained applications such as server power supplies. Replacing 'improvement' with numbers is more concrete: versus a conventional wire-wound transformer, a planar design typically reduces volume and weight to about 30% (Standex data), raises power density to 3×, reaches typical efficiency of 98%, and holds leakage inductance to around 0.2% of primary inductance (all from vendor public technical data). Behind those figures are the geometric consistency from PCB tolerances, the larger copper surface for heat spreading, and the low leakage from interleaved layers.

Second, magnetic integration. Multiple discrete magnetic components are combined onto a shared core, reducing component count, shortening the power loop and lowering parasitics. A typical implementation integrates the resonant inductor with the main transformer, saving space and cutting interconnection loss.

That said, planar is not universally superior. PCB copper weight and layer count limit current-carrying capability, so Litz retains an advantage in scalability for high-current, high-power designs. Fixed layer spacing in planar windings delivers good consistency but narrows the tunable range of leakage inductance. Multilayer PCB cost is also unattractive at low volume. The same caveat applies to magnetic integration — it demands tighter parameter matching and process consistency than a discrete approach, and the higher the integration, the less room remains for adjustment.

6. Closing thoughts

Looking back across these three barriers, they are not three independent problems to be solved separately. They are a mutually constraining set: strengthening insulation encroaches on winding space and raises loss; compressing volume worsens thermal pressure and erodes consistency margin; pursuing maximum consistency drives up process cost and lead time.

Designing SST magnetics, therefore, is not about maximising any single metric. It is about locating the balance point that matches a specific topology and duty profile. That capability sits with neither the system integrator nor the semiconductor supplier — it requires dedicated magnetic component design and manufacturing expertise, engaged collaboratively from the early design stage.