Why Matrix Transformers Exist
In a 48 V to 1 V converter delivering 500 A to a server or AI accelerator rail, a single transformer cannot be efficient: the secondary is one turn at enormous current, the winding resistance becomes the dominant loss, and the window cannot physically hold enough copper. The matrix transformer solves this by splitting one transformer into several smaller ones, each handling a fraction of the current, with primaries in series and secondaries in parallel — or the reverse.
The effect on loss is significant: total copper loss falls roughly in proportion to the number of elements, while core loss stays manageable because each core carries a smaller flux swing. What changes is the mechanical and assembly complexity.
Topologies and the Two Scaling Rules
| Configuration | How it scales | Best for | Main penalty |
|---|---|---|---|
| Series primary, parallel secondary | Current divides by N, voltage per element divides by N | High-current low-voltage outputs (VRM, 48 V to 1 V) | Secondary termination resistance and current sharing |
| Parallel primary, series secondary | Current divides by N, output voltage multiplies | Higher output voltage at moderate current | Primary current sharing and drive complexity |
| Interleaved / integrated PCB winding | Windings formed in PCB layers rather than wire | Very low profile, high frequency | Copper thickness limits and PCB cost |
Current sharing is the central design problem: elements never match perfectly, so the one with the lowest impedance carries more current and runs hotter. A 5 % impedance mismatch in a four-element matrix puts roughly 60 % of the extra current in one element, which is enough to dominate the thermal design.
Controlling Current Sharing
Three levers, in order of effectiveness: identical geometry and turns count with tight tolerances; a small intentional resistance in each element’s path to swamp the mismatch; and thermal coupling so that a hot element transfers heat to its neighbours rather than running away. In practice, matrix designs specify tighter than usual tolerances on inductance and DCR, and often accept a small efficiency penalty for a shared thermal mass.
Measurements matter too: DCR measured at 20 °C on a bench will not predict sharing at 100 °C, because copper’s temperature coefficient divides the current differently as elements heat. Ask for DCR and inductance distribution per lot, and for a current-sharing test at rated load.
Winding Construction Options
- PCB integrated windings. Best for very low profile and high frequency; copper thickness of 2–4 oz limits current per layer, so several layers are paralleled.
- Foil secondaries. Cheap and effective at 100–300 kHz; termination is the weak point.
- Stamped or bus-bar secondaries. For very high current; requires tooling and careful joint design.
- Litz or parallel strands. For higher frequency where AC loss dominates; costs fill factor.
Where Matrix Designs Fail
Most failures are mechanical rather than magnetic: a solder joint on one secondary that adds 50 µΩ and skews the sharing; a core that is not clamped identically across elements; a potting process that leaves voids in one element only. The design review should therefore include a current-sharing measurement at temperature, not only a room-temperature inductance check.
Related reading
- Coupled Inductor Design for Multiphase Buck Converters
- GaN and SiC High-Frequency Inductor Design
- Data Center Server PSU Magnetics
- Planar Magnetics Design
We build matrix and planar transformer assemblies with matched elements, per-lot DCR and inductance distribution, and a current-sharing test at rated load and temperature. Send your output current, switching frequency, and height limit to sales@sirui-tech.com and the Magnetic Core team will propose an element count and winding construction.