EPC’s New 100 V GaN Inverter Boards: 35 ARMS, 100 kHz, 80 V Bus

In short: On 27 October 2026 EPC announced the EPC91124 and EPC91125, two 100 V three-phase BLDC inverter evaluation boards that run from an 18 V to 80 V bus, switch at up to 100 kHz, and deliver 35 ARMS and 30 ARMS of continuous phase current (45 ARMS and 40 ARMS peak). Both use the same 79 × 80 mm layout, so a design team can compare two current classes without changing fixture or firmware. They are development hardware for battery-powered motion — cobot joints, AMRs, servo axes, power tools and drones — not a replacement for a 400 V cabinet drive. The number worth remembering is not the current rating. It is that 100 kHz is now a supported design point on a board you can buy for $690.
What EPC announced, in one table
| Board | GaN FET (100 V) | Continuous | Peak | Bus / PWM |
|---|---|---|---|---|
| EPC91124 | EPC2361, 0.75 mΩ typ., 3 × 5 mm | 35 ARMS | 45 ARMS (65 Apk) | 18–80 V / 100 kHz (150 kHz max) |
| EPC91125 | EPC2367, 1.2 mΩ typ., 3.3 × 3.3 mm | 30 ARMS | 40 ARMS (56.4 Apk) | 18–80 V / 100 kHz (150 kHz max) |
Both boards measure 79 × 80 mm including connectors, share the same gate-drive and sensing architecture, and sit within $10 of each other: $690 for the EPC91124 and $680 for the EPC91125. The FETs themselves list at $9.11 and $7.37 each in single quantities, which tells you where the cost sits — in the board, the current sensors and the assembly, not in the silicon.
Why two boards at all? The EPC2361 is the larger die: a 133 A, 0.75 mΩ device in a 3 × 5 mm package. The EPC2367 is a 101 A, 1.2 mΩ device in 3.3 × 3.3 mm. At 35 A per phase, conduction loss in one device works out to 0.92 W for the EPC2361 and 1.47 W for the EPC2367, so the bigger die buys thermal headroom at the same current — which is exactly what you want to measure when the enclosure is the constraint.
The “45 ARMS” headline is a peak number
EPC’s product pages are titled “45 ARMS” and “40 ARMS”. Those are peak figures. The continuous ratings are 35 ARMS and 30 ARMS, and the announcement does not say how long a peak may be held or at what duty cycle. If you are sizing a cable, a connector or a motor winding, use the continuous number.
One line in the release deserves a second read: “Thermal characterization under realistic operating conditions demonstrates continuous operation up to 40 ARMS with passive cooling.” It does not say which board, at what ambient temperature, with what airflow, or for how long. Forty ARMS is above the 35 ARMS continuous rating of the EPC91124, so either the sentence describes a specific test setup or it describes the smaller board under favourable conditions. Ask for the thermal report before you size a heatsink from a press release sentence.
Why 100 kHz is the part that matters
EPC’s own baseline is that conventional silicon motor drives “typically operate at switching frequencies around 20 kHz”. In practice, most cabinet drives ship at 2–4 kHz and are rarely pushed past 8–16 kHz, because every step up adds switching loss and derates the output current. Moving the same power stage to 100 kHz changes three things at once: ripple, dead-time distortion and electromagnetic behaviour.
| Switching frequency | PWM period | Phase-current ripple (48 V bus, 100 µH, 50% duty) | Dead-time share (50 ns) |
|---|---|---|---|
| 4 kHz | 250 µs | 30 A peak-to-peak | 0.02% |
| 20 kHz | 50 µs | 6 A peak-to-peak | 0.1% |
| 100 kHz | 10 µs | 1.2 A peak-to-peak | 0.5% |
The ripple column uses the standard buck-cell relation ΔI = Vdc · D(1−D) / (L · f) with a low-inductance servo motor of 100 µH. At 100 kHz the same motor sees one fifth of the ripple it sees at 20 kHz, which is why high-frequency GaN drives can shrink filter magnetics, cut torque ripple at the switching frequency and reduce audible noise. It is also why the current loop can update every 10 µs instead of every 50 µs.
The dead-time column is the cost side. Fifty nanoseconds is fifty nanoseconds at any frequency, but as a share of the PWM period it grows from 0.02% at 4 kHz to 0.5% at 100 kHz. That voltage error has to be compensated in software, or it shows up as current distortion and torque ripple at low speed — the exact operating point where a cobot joint spends most of its life. High-frequency drives are as much a control problem as a power-stage problem.
EPC also made a deliberate EMC choice: the boards are set up with dv/dt “under about 10 V/ns” rather than at the fastest edge the FETs can produce, and the gate drive can be re-optimised for faster transitions. At an 80 V bus, 10 V/ns means an 8 ns edge. Push it faster and you trade switching loss for ringing, radiated emissions and motor-cable reflections. That knob is the real engineering decision on this platform.
What is on the board — and what is not
Each board is a complete power stage, not a bare inverter:
- Three-phase bridge with gate drivers and regulated housekeeping rails
- Phase-current sensing through integrated Hall-effect sensors: 250 kHz bandwidth, 13.2 mV/A sensitivity, roughly ±125 A range
- Phase-voltage and DC-bus voltage monitoring, temperature sensing, input undervoltage lockout
- Encoder or Hall interface with two selectable voltage levels
- Schematics, bill of materials, Gerber files and reference firmware published on GitHub
What is not on the board is the microcontroller. EPC sells controller adapters separately — EPC9147A for Microchip dsPIC33, EPC9147B for Texas Instruments C2000, EPC9147C for ST STM32 and EPC9147D for Renesas RAxT. Budget for the adapter and for the time to port your control loop; the $690 buys the power stage, not a working drive.
The current-sense specification is worth doing arithmetic on. At 13.2 mV/A across ±125 A, full scale is ±1.65 V, so a 35 A phase current produces 462 mV. On a 12-bit converter with a 3.3 V reference, one count is about 61 mA. That is comfortable for torque control at rated current and marginal for a 2 A holding current in a stationary joint, where one count is 3% of the signal. If your application spends its time at low current, plan a second sense path with a lower range.
Also absent: any claim of reinforced isolation, a safety-rated torque-off, or CE and UL compliance. This is a bench evaluation platform, and the release does not pretend otherwise.
How much motor can an 80 V bus actually turn?
With linear space-vector modulation, the maximum line-to-line RMS voltage is about 0.707 × Vdc. Applying that to the rated continuous currents gives a realistic power envelope, assuming a 0.85 power factor and ignoring inverter losses:
| DC bus | Max line-line RMS | EPC91124 at 35 ARMS | EPC91125 at 30 ARMS |
|---|---|---|---|
| 48 V | 33.9 V | ≈1.75 kW | ≈1.50 kW |
| 60 V | 42.4 V | ≈2.19 kW | ≈1.87 kW |
| 80 V | 56.6 V | ≈2.91 kW | ≈2.50 kW |
These are estimates, not EPC figures, and they move with the motor’s power factor and the modulation limit you run to. But the order of magnitude is the point: this is a 1.5–3 kW class of hardware aimed at battery buses. A 400 V, 1.5 kW servo axis draws under 5 A per phase — the new boards are not a smaller version of that drive, they are a different machine built around a different bus.
Where 100 kHz bites on a real machine
- Cable reflections. An 8 ns edge on a 48–80 V bus behaves like a transmission line once the motor cable is longer than a metre or two. The motor terminals can see close to twice the bus voltage, and the inverter sees the ringing. With 100 V devices on an 80 V bus, that is a 20 V margin, not a comfortable one. Keep motor leads short, use shielded cable, and add an output reactor if the cable run is long.
- Bearing and shaft currents. Every switching event pushes common-mode charge through the motor. At 100 kHz there are 25 times more events per second than at 4 kHz, so shaft grounding, a common-mode choke or a shielded cable stops being optional on a direct-drive joint.
- Dead-time distortion at low speed. As the table above shows, 50 ns is 0.5% of a 10 µs period. Without compensation, a joint that creeps at 2% speed will show current distortion and torque ripple that a 4 kHz drive never had.
- Sense noise at low current. 13.2 mV/A is a wide-range compromise. A 35 A drive that also needs clean control at 1 A will want a second, lower-range current path or careful oversampling.
- The 100 V ceiling. An 80 V bus leaves 20% headroom. GaN FETs do not carry the repetitive avalanche rating that silicon MOSFETs do, so regenerative energy and cable ringing have to be clamped rather than absorbed. On an AMR, that means a brake chopper or a regen limit in the battery management system, sized before the inverter, not after.
What this changes for the drives already in your cabinet
Nothing this year, and probably not next. A GaN evaluation board is not a retrofit kit: it has no enclosure, no mains rectifier, no fieldbus, no safety functions and no certification. A 400 V Altivar or SINAMICS drive in a cabinet switches at a few kilohertz and will keep doing so. The spare-part list for that drive does not change because EPC shipped a 100 kHz board.
What does change is the comparison you make when a low-voltage axis is designed from scratch. If you are specifying a traction inverter for an AGV, a battery-powered test stand or a direct-drive joint, these boards are a cheap way to measure the trade-off between switching frequency, filter size and thermal headroom before committing to silicon. EPC has now shipped several motor-drive families in 2026 alone — the integrated 100 V module board EPC91122 in February, the 40 V EPC91121 in March, the EPC91128 to EPC91131 series in June, and now these two — which is a reasonable indicator that the reference designs are being used, not just announced.
For maintenance teams, the practical move is to know the switching frequency and derating curve of the drives you already run, because that number — not the semiconductor material — sets the motor temperature rise and the audible noise on the line today. Altivar drives, for example, let you raise the switching frequency into the 2–16 kHz range; most sites leave it at the factory setting because higher settings derate the output current. That is the same trade-off EPC is making at 100 kHz, just at a different point on the curve.
The control layer is moving at the same pace: Omron’s C6 servo has just reached European distribution, and the CompactLogix 5390 pushes a single controller to 128 motion axes. Neither changes the power stage in your cabinet, but both change what you specify around it.
The parts that do not change when the switching stage does
We do not stock EPC devices or evaluation boards. What we do stock is the hardware around a motor-drive bench and a working cabinet, and none of it cares whether the bridge is silicon or GaN:
| Part | Role |
|---|---|
| Schneider Electric ABL8RPS24100 | 240 W, 24 V supply for the controller, encoder and interface rail |
| Schneider Electric LC1G1854KUEN | 305 A four-pole contactor for the AC feed to a bench supply, not a DC bus switch |
| Schneider Electric ATV320U07N4C | 0.75 kW drive — the incumbent class a joint drive is measured against |
| Schneider Electric ATV320U15N4B | 1.5 kW drive — next frame size, same control platform |
| Schneider Electric ATV310HU75N4E | 7.5 kW drive — the plant-side workhorse for pumps and conveyors |
| Siemens 6SL3243-0BB30-1PA3 | G120 control unit — the control side of a Siemens drive |
| Siemens 6SE7016-0TP50-Z | SIMOVERT MASTERDRIVES unit for lines that still need legacy spares |
You can browse the rest of the drive and control stock, or read the rest of our industry news.
What EPC has not disclosed
- Efficiency curves against switching frequency — the release gives none.
- The test conditions behind the 40 ARMS passive-cooling claim, and which board it refers to.
- How long the 45 ARMS and 40 ARMS peaks may be held, and at what duty cycle.
- EMI pre-compliance data at 100 kHz with realistic motor-cable lengths.
- Controller adapter pricing, which is a real line item next to a $690 board.
- Any qualification or lifetime data at 100 kHz beyond the standard FET qualification.
Bottom line
The EPC91124 and EPC91125 are the 100 V, discrete-FET siblings of the 40 V EPC91121, in the same 79 × 80 mm envelope, and that is the useful part of the announcement: two current classes, one fixture, 100 kHz available on both. For a team building a battery-powered servo axis, $690 plus an adapter is a cheap way to find out where the thermal and EMC limits really are.
For everyone else, the announcement is a marker rather than a product. The 100 V ceiling keeps it out of 400 V cabinets, the missing controller and certification keep it out of production machines, and the continuous ratings are lower than the headlines suggest. Read the numbers with the peak-versus-continuous distinction in mind and the platform is exactly what EPC says it is: an evaluation board for high-frequency motion design.
Frequently asked questions
What did EPC release on 27 October 2026?
EPC released the EPC91124 and EPC91125, two 100 V three-phase BLDC motor-drive evaluation boards for 18 V to 80 V buses. They support PWM switching up to 100 kHz, deliver 35 ARMS and 30 ARMS of continuous phase current, share a 79 × 80 mm footprint, and are priced at $690 and $680.
What is the difference between the EPC91124 and EPC91125?
The power devices. The EPC91124 uses the EPC2361, a 100 V, 0.75 mΩ FET in a 3 × 5 mm package, and is rated 35 ARMS continuous and 45 ARMS peak. The EPC91125 uses the EPC2367, a 100 V, 1.2 mΩ FET in a 3.3 × 3.3 mm package, rated 30 ARMS continuous and 40 ARMS peak. The rest of the board is the same.
Can these boards run from a 400 V DC bus?
No. Both boards use 100 V GaN devices and are specified for an 18 V to 80 V bus. A 400 V bus would destroy the FETs on the first switching event. Mains-fed servo drives are a different voltage class and a different design.
Do I need a separate microcontroller?
Yes. The boards contain the power stage, gate drivers, sensing and housekeeping supplies, but no MCU. EPC sells controller adapters separately: EPC9147A for Microchip dsPIC33, EPC9147B for Texas Instruments C2000, EPC9147C for ST STM32 and EPC9147D for Renesas RAxT.
Is 45 ARMS a continuous rating?
No. The continuous ratings are 35 ARMS for the EPC91124 and 30 ARMS for the EPC91125. The 45 ARMS and 40 ARMS figures are peak output currents, and EPC has not published how long those peaks may be sustained.
Will GaN inverter boards replace the drives in my cabinet?
Not as a retrofit. These are open evaluation boards without an enclosure, mains rectifier, fieldbus, safety functions or certification, and their 80 V bus limit excludes 400 V drives entirely. GaN changes the switching stage inside new designs; it does not convert an installed drive.
Sources
- EPC press release, 27 October 2026: EPC Expands 100 V GaN Motor Drive Portfolio
- EPC91124 product page and quick start guide
- EPC91125 product page and quick start guide
- EPC2361 datasheet summary: 100 V, 133 A, 0.75 mΩ
- EPC2367 datasheet summary: 100 V, 101 A, 1.2 mΩ
Autronexa is an independent supplier of industrial automation spare parts. We are not affiliated with EPC, Schneider Electric or Siemens, and we do not stock EPC devices or evaluation boards. Specifications are quoted from the sources above; verify them against the current datasheet before any design, procurement or safety decision.
Replacing a drive, a control unit or the hardware around a test bench? Send us the nameplate data and we will check stock and lead time — contact the team here.
