Engineers designing high-power AC-DC supplies know the word "paralleling" all too well: when a PFC boost stage or an LLC primary has to carry ten-plus kilowatts, a single 650V silicon superjunction MOSFET cannot deliver a low enough on-resistance, so designers add a second device, and then a third. HXY Electronics (HuaXuanYang) offers a different path with its 750V SiC MOSFET HXYS355N75MPI (750V/355A/5mΩ, TO-247H-4L Kelvin-source package): one high-current SiC device replaces the parallel group, moving the current-sharing, symmetric layout, and multi-channel drive problems inside the device and simplifying the power stage of high-power supplies.

1. The Layout Engineer‘s Obsession with Symmetry

Anyone who has laid out parallel power MOSFETs knows the obsession: three devices arranged in a perfectly symmetric pattern, gate traces matched in length and width, source loop areas calculated device by device, with any asymmetry raising the fear that one device will hog the current. This is a recognized precision task in high-power supply design — and a recognized place where designs fail. Every hour spent perfecting the symmetric footprint is an hour not spent on the rest of the platform. The cost of failure is real: a hot-running device and a few efficiency points in the mild case; lost dynamic current sharing, overstress on a single device, and a blown power stage in the severe case.

2. Three Difficult Ledgers of the Paralleled Approach

Putting multiple silicon superjunction MOSFETs in parallel means balancing three ledgers at once, and none of them is easy.

The first ledger is current sharing. Static sharing depends on how well the on-resistances match: the device with lower resistance takes more current, and the positive temperature coefficient of RDS(on) pushes the current back — but this self-balancing mechanism only works when initial parameters and thermal conditions are closely matched. Dynamic sharing is trickier still, determined by the matching of threshold voltage VGS(th) and gate charge Qg: the device with the lowest threshold turns on first and turns off last, inherently absorbing a larger share of the switching loss. Industry application notes state the conclusion bluntly: parameter spread between paralleled devices is unavoidable, and a current derating of typically around 20% is required — three 40A devices in parallel do not give you 120A of capability, but roughly 96A of usable current. As for switching loss, paralleling does not dilute it; it can actually increase it through dynamic imbalance.

The second ledger is gate drive. With N devices in parallel, total gate charge is N times that of a single device, and the driver‘s peak current capability and drive loss scale up accordingly. Every device needs its own series gate resistor to damp parasitic oscillation in the gate loops — identical values, symmetric routing. Some vendors go as far as recommending ferrite beads on the gates or separate totem-pole outputs when paralleling. Common source inductance is the hidden driver here: during switching, the source current transition induces a voltage across the shared source trace inductance that directly erodes the effective gate voltage, unequally from device to device. Industry application notes call it the most serious stray circuit element in paralleled layouts.

The third ledger is thermal management and reliability. Paralleled devices must be thermally coupled — mounted on the same thick copper baseplate or a common heatsink so junction temperatures stay close; separate heatsinks are widely documented as the worst possible arrangement. Superjunction devices carry an additional risk: in fault-current-limiting situations that force the device into the saturation region, the temperature coefficient is no longer guaranteed positive, and the self-balancing mechanism disappears. Add N devices meaning N times the solder joints, mounting stress, and potential failure points, plus redundancy design that typically selects devices for 1.2 to 1.5 times the total current — and the true cost of the paralleled approach is far more than unit price times N.

3. The Single-Device Idea: Moving the Problem Inside the Package

Flip the thinking: if one device offers a lower on-resistance than the theoretical parallel value of three, do these problems simply stop existing?

The core parameters of HXY Electronics‘ 750V SiC MOSFET HXYS355N75MPI anchor exactly this idea. On-resistance is 5mΩ — three paralleled 40mΩ superjunction devices give a theoretical 13.3mΩ, before the roughly 20% current-sharing derating is applied. The single SiC device cuts on-resistance to less than half of the parallel group‘s theoretical value, and every milliohm is fully available for the job, with no sharing losses. At the 40A RMS current typical of a 12kW-class PFC stage, 5mΩ corresponds to roughly 8W of conduction loss — a fraction of what the three-device group faces once high-temperature coefficients are included (detailed below).

The current rating is 355A, giving more than six times margin over the 40A-class continuous current, with peak current excursions comfortably inside the device‘s capability. The 750V blocking rating provides roughly 1.65 to 1.88 times margin over the 400V PFC bus — more headroom than the 1.2 to 1.4 times typical of 650V devices, easing voltage-stress anxiety during surges and abnormal operating conditions.

The TO-247H-4L package is a four-pin Kelvin-source package: the drive loop is separated from the power loop inside the package. The common-source-inductance problem described above — which a three-device, three-pin layout must manage through careful board layout — is structurally removed at the package level. For a single-device design this matters even more: half of the parasitic battle that used to be fought in layout is now won by the package.

On the drive side, HXY Electronics‘ SiC product line supports +15~18V turn-on and 0V turn-off (unipolar drive): no negative rail is required, a single positive supply suffices, and the simplified drive architecture carries over. The design work of N symmetric gate resistors collapses into one gate loop. As for current sharing inside the device — whatever the internal construction, what lands on the engineer‘s bench is a single transistor whose parameter matching and thermal consistency are guaranteed by the manufacturing process, no longer consuming the layout engineer‘s attention.

For higher-voltage scenarios with an 800V DC bus (three-phase high-power modules, energy-storage PCS), HXY Electronics offers 1200V devices (such as HXYS165N120MPI, 1200V/165A/13mΩ) following the same selection logic; this article does not expand on that class.

4. A Typical Application: The High-Frequency Leg of a 12kW Totem-Pole PFC

Consider the front end of a 12kW-class telecom rectifier or charging power module: single-phase input, bridgeless totem-pole PFC, 400V bus voltage, switching frequency set around 100kHz. The controller or DSP generates the PWM in CCM mode with cycle-by-cycle current control; each switch receives one isolated gate-drive channel with +15~18V turn-on and 0V turn-off, propagation delay in the hundred-nanosecond class to match the dead-time budget; in the power stage, the HXYS355N75MPI serves as the high-frequency-leg switch, its Kelvin-source pin separating the gate-loop stray inductance from the power loop so the gate waveform stays clean under high dv/dt switching edges; the load side is the boost inductor and bus capacitor, with a peak inductor current in the 60A class — well within the device‘s current margin.

The two approaches side by side:

Translated into losses: at the same 40A RMS current and 175C junction temperature, the three-device parallel group presents roughly 33mΩ of equivalent on-resistance and about 53W of conduction loss; the HXYS355N75MPI presents about 7.5mΩ and roughly 12W — a 41W difference that lands entirely on heatsink size, fan speed, and overall efficiency. Layer on the switching side: the SiC body diode‘s reverse-recovery charge is in the nanocoulomb class, so hard-switched commutation produces none of the recovery current spikes seen in silicon devices, and frequencies above 100kHz sit comfortably within SiC territory — while the dynamic current-sharing problem of the paralleled superjunction group only worsens as frequency rises. Industry field data already exists: on an existing PFC+LLC platform, a single SiC device directly replaced two paralleled superjunction MOSFETs with no topology or magnetics changes, and full-load temperature dropped by 10 to 25 degrees Celsius.

The boundaries of this substitution deserve equal clarity: in cost-sensitive scenarios below 65-100kHz where a 40mΩ-class device suffices, a single superjunction MOSFET remains the economical choice. The value window for a high-current single SiC device opens at the power level where silicon needs paralleling to survive.

The LLC primary side tells a similar story with its own nuances. Paralleling is equally common there, and although zero-voltage switching removes most turn-on loss, three silicon-specific burdens remain: the body diode conducts during every dead-time interval and its microcoulomb-class Qrr surfaces as recovery spikes when the opposite device turns on; the highly nonlinear output capacitance of superjunction devices narrows the ZVS window and makes dead-time tuning sensitive; and the dynamic sharing problem persists because each paralleled device still switches at its own pace. A single SiC device addresses all three at once — nanocoulomb-class Qrr, a more linear output capacitance that behaves predictably across the voltage swing, and one switching edge instead of N. The design effort that went into matching dead time across a parallel group converts into simply shortening it, which directly trims the primary-side circulating loss.

5. Design Pitfall Guide: Clear These Five Before Going Single-Device

First, a single device does not mean unlimited derating — thermal resistance is the real ceiling. 355A is the device‘s stated current capability; the practical limit is set by the temperature rise of "loss power times junction-to-case thermal resistance": back-calculate the allowable loss from the target junction temperature, then derive the usable current. Heatsink flatness and thermal-interface material application directly determine whether this high-current device delivers its parameters. Refer to the official datasheet for specific thermal-resistance values.

Second, route the Kelvin source independently. The fourth pin of the TO-247H-4L is the drive-dedicated source return; it must be routed separately from the power loop and returned to the driver ground nearby. Shorting the fourth pin into the power-source trace for convenience forfeits the Kelvin package‘s benefit.

Third, size the drive‘s peak current for a high-current device‘s gate charge. A 5mΩ-class device has a large active die area, and its gate charge is correspondingly substantial (refer to the datasheet for exact values). Provide ample peak-current capability in the gate driver, start with a small gate resistor, and converge on final values using measured switching waveforms rather than copying experience from lower-current SiC devices.

Fourth, 0V turn-off works, but high-dv/dt bridge legs warrant a Miller clamp as a safety net. HXY Electronics‘ SiC supports unipolar drive with 0V turn-off, eliminating the negative rail; in high-dv/dt positions such as the totem-pole high-frequency leg, Miller coupling during turn-off can still lift the gate voltage. Choosing a driver with an active Miller clamp — or keeping a small negative drive as an engineering option — makes the design more robust.

Fifth, budget for EMI up front. The single-device solution has steeper switching edges and higher di/dt, and switching energy previously distributed across multiple devices now concentrates in one location. Minimum power-loop area, bus capacitors placed close to the switches, and snubbers where necessary — these fundamentals cannot be skipped in a SiC single-device design.

6. Manufacturer and Summary

Paralleling was the silicon era‘s transitional answer for high-power supplies: sharing derating consumes 20% of current capability, symmetric layout and multi-channel drive consume engineering hours, and N devices leave behind a reliability bill of solder joints and thermal coupling. HXY Electronics‘ 750V/355A/5mΩ SiC MOSFET HXYS355N75MPI clears these costs in one stroke with a single device: conduction loss at 40A runs roughly 40W below the three-device parallel group, voltage margin rises above 1.65 times, and the drive circuit collapses from N channels to one. For teams planning a 5-15kW supply platform — or reworking a board because of current-sharing problems — this single-device option deserves a slot in the next design review. HXY Electronics focuses on power semiconductor product lines and provides selection evaluation and technical support for specific platforms.

This article is for reference only and does not constitute any purchasing or design commitment. Design selection should be based on HuaXuanYang‘s latest official datasheets; for application questions, contact HXY Electronics technical support (sales@hxymos.com).

声明 / Disclaimer
本文部分参数引自华轩阳产品手册或第三方规格书，数据有更新可能，仅供选型参考，最终设计请以最新版规格书与实测为准。
联系 Contact：sales@hxymos.com