Totem pole PFC design and real time validation

Key Takeaways

• A totem pole PFC removes diode bridge losses, and the saving holds only when the controller manages the role swap at each AC zero crossing.
• Zero crossing distortion is a timing and sequencing problem, solved with blanking, a ramped restart and careful polarity sensing.
• Real-time testing with switching-level resolution lets you repeat ugly line conditions against your actual controller before hardware is at risk.

A totem pole PFC earns its efficiency gain only when its control holds steady through the AC zero crossing.

Power supplies for servers, chargers and industrial drives all start with a power factor correction stage, and the losses there scale with every kilowatt behind it. Data centres alone consumed about 4.4% of total U.S. electricity in 2023, so each tenth of a point recovered at the front end matters. Removing the diode bridge with wide-bandgap switches recovers more than a tenth, and it hands your controller a harder job.

That job sits in a window a few hundred microseconds wide, twice per line cycle. The switches trade roles, the duty cycle swings from one extreme to the other, and the polarity measurement is at its least reliable. Treat the totem pole converter as a control problem first and a topology second. You’ll get the most from it when the zero crossing is designed and tested on purpose.

What a totem pole PFC does differently from a boost PFC

The main difference between a totem pole PFC and a conventional boost PFC is the diode bridge. A boost PFC rectifies the line through four diodes and then shapes current with one switch. A totem pole PFC removes the bridge and uses two half bridge legs to rectify and boost in a single stage.

Consider a 3.3 kW stage on a 230 V line as a generic illustration. It draws roughly 14 A RMS, and a bridge always has two diodes conducting. At about 1 V per diode, that’s close to 26 W of heat, or 0.8% of throughput. The top efficiency tier for server supplies requires 96% efficiency at half load, which leaves 4% for every loss in the unit.

The bridge alone would spend a fifth of that budget. A totem pole stage replaces it with one fast leg switching at high frequency and one slow leg switching at line frequency.

“What you give up is simplicity, because rectification now depends on your firmware.”

Why wide-bandgap switches make continuous conduction operation practical

Wide-bandgap switches make a totem pole converter practical at kilowatt levels because they have little or no reverse recovery charge. Continuous conduction forces one fast leg switch to turn on while the other still carries inductor current through its reverse path. A silicon MOSFET body diode stores too much charge for that commutation to be efficient or safe.

Picture the fast leg at 65 kHz with 20 A flowing in the inductor. With a silicon superjunction part, the body diode’s recovery current adds to the load current every cycle, and the resulting loss and voltage stress grow with frequency. A gallium nitride transistor has no body diode to recover, and a silicon carbide MOSFET recovers a small fraction of the charge. The price is edge speed. Faster edges shorten the time your controller has to get each transition right.

How totem pole PFC control handles each half cycle

Totem pole PFC control uses the same outer voltage loop and inner current loop as a boost PFC, plus a polarity signal that assigns switch roles. During the positive half cycle, one fast leg switch acts as the boost switch and its partner acts as the synchronous rectifier. The roles reverse for the negative half cycle.

Take the positive half cycle. The low side fast switch charges the inductor, the high side fast switch releases that energy to the bus, and the low side slow switch stays on to return current to the line. Once polarity flips, the high side fast switch becomes the boost switch and the high side slow switch carries the return path.

The duty cycle follows the line. An active switch needs a duty cycle near 100% when the input voltage is close to zero. So at each crossing, the switch that was nearly always on becomes the rectifier and must be nearly always off.

Nothing in a boost PFC resembles that step. Your current loop and your modulator have to agree on the instant it happens, and they depend on a polarity reading taken when the signal is smallest.

What causes zero crossing distortion in a totem pole converter

Zero crossing distortion in a totem pole converter comes from the role swap happening at the wrong instant or too abruptly. The inductor briefly sees the full bus voltage and current spikes. The result is a notch or spike on the line current twice per cycle.

Assume a 400 V bus and a 200 µH inductor as an illustration. If the slow leg flips while the fast leg still runs a duty cycle near 100%, the bus sits across the inductor and current ramps at 2 A every microsecond. A polarity detector that is 5 µs late produces a 10 A spike on a waveform that should be passing through zero.

Sensing makes it worse. Measurement offset, filter delay and noise from fast switching edges all move the detected crossing away from the true one, and the current loop integrator arrives holding a value from the previous half cycle. Each spike raises harmonic distortion and stresses the electromagnetic interference filter.

Design choices that shrink the zero crossing current spike

The zero crossing spike shrinks when the transition is treated as a sequenced event. Stop the fast leg just before the crossing, wait a short blanking interval, turn the slow leg over, and then restart the fast leg with a duty cycle that ramps from narrow pulses to its commanded value.

A common sequence blanks all four gates for a few tens of microseconds around the detected crossing. Blanking that’s too long leaves a flat spot in the current, and blanking that’s too short brings the spike back, so the window is a tuning decision you’ll revisit for every line voltage and load.

How to validate a totem pole converter in real time

You validate a totem pole converter in real time when you run your actual controller against a simulated power stage that resolves individual switching events. Hardware-in-the-loop testing lets you repeat the zero crossing thousands of times under faults and line disturbances too risky to stage on a prototype.

Resolution decides how much the test is worth. At 100 kHz, one switching period lasts 10 µs, and a model stepping at 1 µs sees it in only 10 slices. A 100 ns dead time vanishes at that step size, and a coarse totem pole PFC simulation hides exactly the behaviour you came to see. OPAL-RT runs the converter model on FPGA-based simulators so totem pole switching is modelled through the zero crossing in real time.

A useful test plan covers the cases that bench work tends to skip.

• Line voltage sags and swells that land on a crossing
• Frequency steps between 47 Hz and 63 Hz
• Flat-topped line voltage with repeated crossings
• Light load operation where current becomes discontinuous
• Sensor offset and delay injected into the polarity signal

Each case repeats identically after every firmware change, so a blanking window tuned on Monday can be regression tested before Friday’s build.

Where disciplined zero crossing testing pays off over a program

Disciplined zero crossing testing pays off because the failures it prevents are the expensive kind. A current spike found in simulation costs a firmware change. The same spike found during compliance testing or in a failed field unit costs a board revision, a schedule slip and some of your team’s confidence.

“A current spike found in simulation costs a firmware change.”

Teams that struggle with this topology rarely picked the wrong switches. More often they tuned the crossing on one prototype at one line voltage, then met a weak grid or a cold start that moved the timing. What separates a dependable product from a fragile one is how many crossings your controller has seen before it ships, and how ugly you made them.

Put the controller in front of a model that’s honest about switching, break the crossing on purpose, and keep the evidence. OPAL-RT builds real-time simulation for that kind of work, so the most fragile few hundred microseconds of your line cycle get tested as thoroughly as everything else.