H-Bridge Resonance Driver IGBT / MOSFET Power Rating

Dunno if I fully understand your specs, but in series resonance your piezo voltage will be dramatically higher than the H-bridge voltage, with a ratio up to the Q. For example, an 80V H-bridge will deliver 160V p-p to the resonant load, which with a modest Q = 12, can see 1kV peak, roughly speaking. You can measure your resonant Q at low power levels with a signal generator, to see what your required input voltage will be.

If you are in exact resonance the series R-L-C looks like a resistance and it's an easy job for the H-bridge to handle, but when slightly off resonance the load looks either capacitive or inductive. Your design needs to be able to handle severe off-resonance conditions, even if only for a short while.

The off-resonance capacitive case greatly increases the current the MOSFETs must handle. It's helpful if your H-bridge supply source has a current limit and the voltage can sink down while you get the system in tune.

In the inductive case, currents flow during the deadtime when the switches are all off, forcing current conduction into the MOSFET's reverse body diodes. These have a reverse-recovery time and fail to stop conducting when the current drops to zero or reverses. Finally they run out of stored charge and a rapid snap-off occurs. This causes large V = L dI/dt voltage spikes, which can damage the MOSFET gates. You can help solve this by selecting MOSFETs that have built-in soft-recovery diodes. If your H-bridge is below 100V, like the 80V example I gave, you can use Schottky diodes across each MOSFET. Naturally, you'll also be adding snubber capacitors or R-C networks to reduce the spike dV/dt.

All the above encourages a substantial over-rating of the MOSFET specs, with higher voltage and much higher current than a simple 2A current calculation. Fortunately there are excellent MOSFETs to fill your need. I suggest MOSFETs over IGBTs because a) they're faster and can support much lower deadtimes, and b) your power requirement is low enough for them to easily handle the task.

About 12 years ago I made a 500W 10kV resonant-drive system at running at 300kHz, using these techniques and an H-bridge powered from 80 volts, max. I used a UCC3895 phase-shift controller with my favorite Intersil HIP4081A gate-driver (because I like its fast speed and high gate current) and four mid-sized 33A 100V Fairchild FQP33N10 MOSFETs, in TO-220 packages with small 3W heatsinks. It's been working well ever since.

Here's a set of waveforms as it's generating 8.5kV peak (yellow), showing a 130V p-p H-bridge switch-output waveform (differential probe, in red) and a 20A p-p output-current waveform (purple) (can that be right? it should be about half that). If you'd like to see my drawings, email me and ask for RIS-514.

Winfield Hill:


You have written that 3W heat sink was enough to switch 1kW plus power using this MOSFET FQP33N10 on continuous bases at 300kHz. That sounds very interesting one. at 30kHz the period is going to increase 10 times so will it be the same or will drop the power rating drastically?

Most of the loss, which equals dissipated power in the MOSFET, is a fixed amount each switching cycle, so it's a direct function of the switching frequency. Your loss would be 10x lower at 30kHz. The excellent drive capability of the HIP4081A is in part responsible for the low switching loss. The low conduction loss is due to the 33A MOSFET.

I should mention that my application was to make a 100V to 10kV linear ramp for a mass spec application. This means the system would spend much less than a minute at full power. The longest amount of time would be to optimize the tuning adjustment at full power.

But you should be fine with 3W clip-on heatsinks for continuous operation at 50kHz or under.

I used a phase-shift controller to allow for very narrow effective on-time pulses, necessary for achieving the lowest voltage levels, and I also scaled the DC voltage along with the phase shift, for the same purpose, to easily get a full 100:1 amplitude range. But this meant the output level was proportional to the control voltage squared, so I added a square-root function inside the feedback loop. In the end my control loop was both very fast (high loop gain, high loop bandwidth) and very stable (good phase margin) without any overshoot for step commands. None of this extra stuff would be necessary for your application.