Skip to content

IGBT vs. MOSFET

Single-sentence summary. At the project's operating point (50 V per bridge, 5 kHz switching, ≈ 400 W total), every loss term favors MOSFETs. The choice was unambiguous and quantitative — not a preference call.

The classical crossover

For a power switch in an inverter, the relevant figures of merit are conduction loss (the loss while the switch is fully on, carrying current) and switching loss (the loss during the transition between on and off, each PWM period). The two technologies make opposite trade-offs:

MOSFET IGBT
On-state behavior Resistive: VDS ≈ I × RDS(on) Voltage-source: VCE(sat) ≈ constant (≈ 1.4 V regardless of current)
Conduction loss at low I Very low (quadratic with current) Constant V × I (linear)
Conduction loss at high I High (RDS(on) grows with temperature; loss scales I²) Bounded by VCE(sat)
Switching transition Fast, clean Slower; tail current adds switching-off energy
Switching loss at low fsw Negligible Significant (tail current loss × frequency)
Switching loss at high fsw Significant Dominant
Body diode Built in (free-wheeling) None — external freewheeling diode required

The crossover point — where IGBTs start to win — is somewhere around 250 V VDS and 20 kHz, depending on the exact part. Below those numbers, MOSFETs are better on every term that matters.

This project's numbers

Parameter Value
VDC per bridge 50 V
Switching frequency 5 kHz (PSC default)
Output current (RMS) ≈ 5 A
Output current (peak) ≈ 7 A
Output power ≈ 400 W

At 50 V / 5 kHz, we're solidly in MOSFET territory. The math:

Conduction loss per device

Using the as-built IRFB4110 (RDS(on) = 4.5 mΩ at VGS = 10 V):

Pcond,MOSFET = I2 × RDS(on) = 52 × 0.0045 ≈ 0.11 W per device

For comparison, a typical 60 V IGBT at I = 5 A:

Pcond,IGBT = VCE(sat) × I = 1.4 × 5 = 7.0 W per device

That's a 63× difference in conduction loss alone. With 4 devices per bridge × 2 bridges = 8 devices in the system, the totals are:

  • MOSFET conduction loss: 0.88 W system-wide
  • IGBT conduction loss: 56 W system-wide

Switching loss per device

The IRFB4110 has QG(tot) ≈ 150 nC. With the project's 22 Ω gate-series resistor and 15 V drive voltage, the gate transition takes ≈ 220 ns.

Per-switching-event loss (turn-on + turn-off, half each at 25 V × 7 A peak per bridge):

Psw,MOSFET ≈ ½ × VDS × Ipeak × trise+fall × fsw ≈ 0.5 × 25 × 7 × 440e-9 × 5000 ≈ 0.19 W per device

For an IGBT in the same role, the tail current adds a large additional loss term (typically 5–10 µJ per switching event):

Psw,IGBT ≈ Esw × fsw ≈ 8e-6 × 5000 = 40 mW from switching transitions + ≈ 5 W tail current loss = 5.04 W per device

System totals:

  • MOSFET switching loss: 1.5 W system-wide
  • IGBT switching loss: 40 W system-wide

Combined

MOSFET total IGBT total
Conduction loss 0.88 W 56 W
Switching loss 1.5 W 40 W
Power-stage loss ≈ 2.4 W ≈ 96 W
Implied efficiency (at 400 W) ≈ 99.4 % ≈ 80 %

(The IGBT efficiency is so bad because both axes — conduction and switching — are in its worst-case region. At 400 V / 50 kHz, the picture inverts.)

The IRFB4110 specifically

Build Guide v3.1 specified IRFZ44N (55 V VDSS, 49 A, 17.5 mΩ RDS(on)). The as-built v4 hardware uses IRFB4110 (100 V VDSS, 180 A, 4.5 mΩ RDS(on)) instead. Three reasons for the swap:

  1. VDSS headroom. IRFZ44N at 55 V VDSS for a 50 V bus + switching spikes is uncomfortably tight. The 1.5KE62A TVS clamps at 84.5 V — well above the IRFZ44N's VDSS, meaning a TVS firing event would still cause MOSFET damage. The IRFB4110 at 100 V VDSS sits comfortably above the TVS clamp, so the protection chain actually protects.
  2. RDS(on) headroom. 4.5 mΩ vs. 17.5 mΩ is a 4× improvement in conduction loss. At 5 A this is the difference between 0.11 W and 0.44 W per device — small in absolute terms, but it means the MOSFETs run cool enough to skip forced air on a bench prototype.
  3. Gate charge tradeoff. The IRFB4110's higher current rating comes with ~2× the gate charge of the IRFZ44N (150 nC vs. ~67 nC). With the same TLP250 + 22 Ω gate resistor, transitions take ~2× longer. The firmware compensates with 3 µs dead time (PWM_DEAD_TIME_DTG = TIM_DTG_3US_AT_64MHZ, BDTR.DTG = 0xA0), up from the IRFZ44N's 2 µs. This is documented in the firmware CHANGELOG and Build Guide v4.0 §6.

The IRFZ44N → IRFB4110 substitution is one of the project's more important course corrections — it eliminated a latent BOM hazard (the TVS / VDSS mismatch), gave better thermal margin, and only cost a slightly longer dead time.

What to remember

  • For 50 V / 5 kHz inverters, the answer is always MOSFET. No exceptions worth discussing.
  • The crossover where IGBTs start to win is around 250 V VDS + 20 kHz, but in practice modern Si and SiC MOSFETs keep pushing that frontier higher.
  • Check VDSS against your TVS clamp voltage. If TVS clamps above MOSFET breakdown, the protection chain is broken.
  • When swapping MOSFETs, recompute the gate-charge → switching-time → dead-time chain. The firmware's dead-time value is the most likely thing to need updating.

Related: Build Guide v4.0 §4 Power stage, Modulators (for the dead-time constants), and the iteration-3 narrative where the MOSFET swap actually happened.