half bridge drive oscillatory gate

Your question isn't clear. There are two FETs, same or different polarities, at least four diodes that could free-wheel (body+clamping diodes), and many combinations and connections possible.

Might you mean this:

. Vdd . -+- . | . +-----. . Q1 | | . ||--'-. | ^ . || exceeding the turn-on threshold by some volts as only Ciss is active.

The gate driver prevents it!

Cheers, James Arthur

. Vdd . -+- . | . +-----. . Q1 | | . ||--'-. | ^ . || where the load is flying back through D1, and Q2 is just being

Yes

The gate threshold is exceeded without the effects of Crss during reverse recovery. When Crss begins to affect the gate, it does so severely enough to turn off the fet temorarily, initiating oscillation.

That's the interesting thing. The gate drive is fairly high impedance, though assisted at the gate with pnp active pull-down.

The oscillation of interest is 4MHz, , source current with slow' rise fast reduction, with peaks exceeding the inductive load current by a factor of ~2.

The oscillation can be sustained for many microseconds, and the resulting EMI confuses nearby electronics. Very lossy and heading towards valhalla, if not disabled promptly, by a reduction of load.

RL

Okay, that's because you're driving Q2 'on', right?

Then the drive isn't stiff enough. You need a heftier driver, one that can overcome the Miller capacitance * dV/dt.

If it oscillates even with a stiff driver then the problem is likely parasitics and you need damping: a small gate resistor or a ferrite bead. That's not your problem though.

Exactly.

Best, James Arthur

That's your problem: you do not have enough channel enhancement margin at the instant of diode recovery and the FET is going high gain linear. The nonlinear diode reverse recovery I-V and a bunch of phase delays equate to oscillation. A possible fix is to reduce the high frequency linear gain with an R+C snubber from the load to GND.

The gate drive is already a combination of the two techniques you recommend. As the reverse recovery of the freewheeling diode/fet has to be handled, fast turn on with a stiff driver is not an option in the bridge. The low impedance turn-off provides stiff drive, when permitted.

Yes..... exactly the problem. If gate drive modifications are the only option, then degrading the turn-off circuit is one logical step, with efficiency implications of it's own.

RL

Probably agravated by the upper mosfet's reciprocating the behavior.

Isn't turn on, accompanied by stable gate waveform threshold plateaus also 'linear'?

An original functional circuit using STW18NB40 had performance that was only degraded at higher power levels by the introduction of assisted turn-off.

Substituting IRFP360LC into the modified circuit produced the oscillations, at very modest power levels, although the physical characteristics of the devices are very similar.

Two backwards steps.

Output snubbing, at 47pF, is probably only symbolic at present. In high voltage circuits, snubbers are easy ways to throw away power, but I can probably double it. Compared to Coss of >400pF, it's a flea-bite.

RL

But I understood the followin passage to mean the gate was only pulled _down_ hard, but not so when driven high:

If the gate is not driven hard 'high,' then why not? Normally, it should be.

It sounds like your driver is instead gently goosing the FET into its linear region, and the FET, possessing a reactive load, proceeds to oscillate.

Why? The standard approach is to just brute-force rip the charge out of that freewheeling diode, i.e., damn the torpedoes, turn Q2 on hard and fast. That's why gate driver ICs put out _amps_--to rapidly traverse the danger zone, avoiding this very problem.

How fast are you switching anyhow? You haven't mentioned any numbers. How do we know irr is a problem?

If you can't abide shoot-through currents, a ferrite bead on Q2(drain) would allow quick transitions despite D1.

Really, you have two kinds of options: to keep the FET from getting and staying linear in the first place (i.e, avoid lingering in the linear region during transitions), or to prevent it from oscillating when it does (Fred's suggestions).

That doesn't help--the load still flies back into D1, and D1 recovery still applies.

If you're worried about irr(D1), what kind of diode are you using for D1, how much freewheeling current is it passing. and how fast are the transitions (i.e., what are the trr requirements)? Might there be better diode choices? These are all things to ponder.

Best wishes, James Arthur

Thanks for the advice.

At 100KHz, >300V switching losses account for close to 3/4 of the losses in the mosfets. As faster switching would not reduce the qrr, this is considered a fixed loss, as irr simply increases linearly with di/dt.

Any modest increase in fall time that resulted would be expected to aggravate EMI, which is one of the features of this pre-existing circuit that I have been assigned to correct. As it fits into a family of others, both more and less complex and with more and less capacity, the choice of options is limited by expectations concerning pricing, volume, complexity and IP.

This is particularly the case when it comes to modifying a power train that has been demonstrated functional in the application. The assisted turn-off was intended only to assist in reducing dissipation in the driver, who's body is being down-sized in a move to SMT.

The exact topology is not as straight forward as originally outlined, but the devices involved in the oscillation refered to in the OP are accurately represented by function during the relevant period of time.

RL.

Ah hah, the motivation for the slow turn-on emerges: EMI. The plot thickens...

A ferrite bead on Q2's drain then? Q2 could switch quickly (avoiding oscillation), but at low current (and low EMI). The current ramp-up would follow, with a slew rate controlled by the bead, lowering EMI. Q2's switching losses would be low.

Two more comments: a) the main EMI problem could be from the oscillation; fast switching might be okay, b) I agree the fact that the oscillation lasts so long _does_ suggest rectification or some other interaction within the driver (even a wimpy driver should slew through the linear region, even if/while the FET oscillates) .

Best, James Arthur