Skip to content

Iteration 1 — single dual-bridge layout, IRFZ44N, IPD LS-PWM

Breadboard IRFZ44N test rig — iteration-1-era component characterisation
Breadboard IRFZ44N test from the iteration-1 era — characterising the original MOSFET choice before any PCB was fabricated. The VDSS-headroom problem the table below describes was visible on this rig before the first board was ordered.

Status: superseded. Not for fabrication. KiCad files for iteration 1 are not preserved in the repo; the closest survivor is the v3.1 BOM at hardware/legacy/iteration-3/CHB_BOM_v3_1.xlsx.

What was attempted

A single PCB that hosted both H-bridges of the 5-level cascade, with:

  • 8× IRFZ44N MOSFETs (4 per bridge) in TO-220 packages, per the Build Guide v3.1 BOM.
  • TLP250 optical gate drivers on every gate (this part was right from the start).
  • IPD Level-Shifted Carrier PWM at 500 Hz, generated by the STM32 Nucleo-F303RE.
  • Three independent MCP3201 ADCs (DC1, DC2, current), each with its own MISO line back to the controller (the firmware's original "3-MISO" assumption).

The intent was the simplest-possible 5-level inverter — one board, one bring-up, and the most-textbook modulation strategy.

What failed

Failure What happened Why
MOSFET VDSS too tight IRFZ44N VDSS = 55 V; bench supplies at 50 V plus switching spikes brought peak VDS uncomfortably close to breakdown. At a 50 V bus the headroom is barely 10 % nominal; spikes from gate-loop parasitics ate that margin.
TVS / VDSS mismatch 1.5KE62A TVS clamps at 84.5 V — above IRFZ44N VDSS. A TVS firing event would still damage the MOSFETs. Latent BOM hazard. The TVS was sized for a higher-VDSS MOSFET than what was specified.
Bridge-1 thermal asymmetry With IPD LS-PWM, Bridge 1 always carried the inner-band switching (±1 step). One of Bridge 1's MOSFETs ran measurably warmer than its counterparts. The IPD mapping puts the most-switched bridge in a fixed role for the whole fundamental period. See PSC vs. LSPWM.
3-MISO firmware assumption The firmware read three MCP3201 MISO lines in parallel during the SPI scan, assuming each ADC had its own dedicated return. This worked on the iteration-1 layout — the issue surfaced later when the modular per-bridge isolation made the MISO scheme harder to maintain.

What was learned

  • Voltage headroom is not optional. A 10 % margin above nominal bus isn't enough when gate-loop parasitics add tens of volts of ringing. Iteration 4 raised VDSS headroom to ≈ 2×.
  • The protection chain has to actually protect. If the TVS clamps above the MOSFET VDSS, a TVS event is still a MOSFET kill. Check the clamp voltage against the MOSFET rating before ordering.
  • Modulator choice has thermal consequences. Symmetric (PSC) vs. asymmetric (IPD) modulation directly translates to whether the bridges run at the same temperature. This wasn't obvious at design time; the bench measurements made it concrete.

What changed for iteration 2

  • Schematic split into modular sheets (the start of the Highside_cell / Lowside_cell / driver_cell decomposition that survives in the iteration-4 KiCad).
  • Gate-drive routing redrawn around bootstrap timing concerns — leading directly into the bootstrap fundamentals discussion in iteration 2.
First MOSFET test light — gate-drive verification
Earliest gate-drive sanity check — a single MOSFET driving an indicator light, confirming the TLP250 isolation chain works end-to-end before scaling to a full H-bridge.

The IRFZ44N and the IPD LS-PWM choice were both kept, on the theory that they were "good enough" and the layout was the bigger risk. Both assumptions were reversed by iteration 4.