Driving a high current MOSFET for primary stage of battery inverter

[theoldwizard1]

Anyone who has looked at a typical DC to AC inverter would notice the the input section is almost identical to a switchmode PS, except that the output drives one or more transformers that step up the voltage to over 200VDC, after rectification (higher for an inverter that must output 240VAC).  Typically, these transformers are driven by multiple low/medium current MOSFETs in parallel.  The problem with this design is that if one MOSFET "dies" it will usually take all of the rest with it.  (Footnote :  A few inverters will forego the high voltage section by driving a LARGE step up transformer)

My thought was, instead of using the multiple parallel MOSFET, why not use one BIG one, say Id >200A ?  The problem is, these "bad boys" typically have a Vgs of 20V.  Is there some kind of ready made voltage doubler, MOSFET driver that could handle this ?

[David Hess]

The use multiple parallel MOSFETs to keep the source inductance low because of the high currents and to achieve a lower junction to ambient thermal resistance.

[RobK_NL]

these "bad boys" typically have a Vgs of 20V. 

Yeah, a maximum V_GS of 20V.

The relevant parameter is V_GS(th), which is typically in the 4-5V range. Driving them with 10V will open them up nicely.

[jbb]

Typically, a number of smaller MOSFETs would be cheaper / easier to source than a couple of big chunky ones.

Also, don’t trust the current ratings of a TO220 or TO247 datasheeet. Look at your losses and thermal resistance instead.

[theoldwizard1]

Also, don’t trust the current ratings of a TO220 or TO247 datasheeet. Look at your losses and thermal resistance instead.

Chassis mount !

I learned a lot about TO220 and TO247 packages.  The screw mount is a joke.  The best you can do is a spring mount, not on the tab but directly on the body.

[T3sl4co1l]

Transistors over 50A are in the domain of diminishing returns.  Or a bit more at higher voltages, but not much more because size goes up as well.

The dual problems are speed and stray equivalent inductance.

Stray inductance is the low frequency manifestation of signals propagating at the speed of light.  The longer a device is, the more inductance it has.

MOSFETs are most useful when they are fast.  Switching in, say, 100ns is slow, especially these days.  Many cases, space is at a premium, forcing ever higher switching frequencies (because smaller capacitors and inductors can be used).  A switching edge of 100ns isn't really "switching" at all, if the frequency is, say, 1MHz or more!

To go from conducting a full, say, 50A, to zero, in some amount of time, causes a momentary voltage drop along all the conductors in the path.  In particular, the source and drain terminals of the MOSFET.

Typical figures are around 10nH for a TO-220 or TO-247 device, 5nH for a D/D2PAK, less than 2nH for a PDSON; and maybe 10-100nH for large industrial modules (depending on size and design).

The inductor equation is: V = L * dI/dt

If L = 10nH, dI = 50A and dt = 100ns (assuming dI/dt is a linear ramp -- fortunately, not terrifically far off in practice!), then Vpk = 5V.  That's a sizable fraction of Vgs(on), meaning you can't turn off a TO-220 much faster than this (the voltage drop, in the source pin, opposes turn-off -- it generates negative feedback).

The voltage drop on the drain pin, mainly due to whatever is connected to it (because traces have length, and therefore inductance, too), causes voltage overshoot.  5V may not sound like much, but the stray inductance can easily be many times larger, and if the source inductance is made small (by using a more compact transistor, or one which provides Kelvin terminals), the transistor can switch faster as well, generating much more peak voltage in the process.

A carelesss SMT layout, using PDSON packages (~2nH per device), and having ~10nH in the switching loop, can switch in 20ns pretty easily, but generate peaks of 25V in the process.  If you started with 40V transistors (say for a 12V automotive application), what you thought started out as a 3.3:1 safety overhead has suddenly vanished!

So what do we do?  We connect transistors in parallel, so that the inductances go down in parallel.

Even better, we might wire multiple inverters in parallel, with independent controls for each channel, so each transistor pair need only handle, say, 20A (and they can do it with lots of safety headroom), while the total delivers, say, 100A or whatever (from 5 channels in parallel).  Knock-on benefits include ripple cancellation, by phase shifting each channel.  This is what PC motherboards do, with upwards of 8 phases!

Downside, the phase shift method may not be applicable for everything (a single phase AC output needs to draw its load current then and there), but even so, the switching edges can still be spread out slightly to help reduce high frequency noise.

Tim

[theoldwizard1]

Transistors over 50A are in the domain of diminishing returns.  Or a bit more at higher voltages, but not much more because size goes up as well.

Thanks for that excellent explanation !

As I stated in the original post, I am trying to understand the trade off in designing a DC to AC inverter.  Inexpensive ones seem to take the 2 stage approach with the first stage being multiple parallel MOSFETs driving one or typically more small/medium sized transformer.  It is very common for these drive MOSFET to blow even with what appears to be adequately large heat sinks an forced air cooling.

Many of the more expensive "domestic" designs forego the first stage and just drive a LARGE transformer via a Class D amplifier.  These seem to be more "robust" or at least last a lot longer.

In the first case, the MOSFETs are switching at >100KHz while in the second case it is a 50/60Hz sine wave.

[T3sl4co1l]

Blown converters are usually due to lack of control.  What IC do you see on the primary side, TL494?  SG3524?  Yeah, there's your problem.

Tim

[theoldwizard1]

Blown converters are usually due to lack of control.  What IC do you see on the primary side, TL494?  SG3524?  Yeah, there's your problem.

Why are these inferior and what would be better ?

[David Hess]

Blown converters are usually due to lack of control.  What IC do you see on the primary side, TL494?  SG3524?  Yeah, there's your problem.

Why are these inferior and what would be better ?

More modern current mode switching regulator controllers include cycle-by-cycle current limiting as a side effect of current mode control which both maintains flux balance in a transformer and protects the output transistors.  Current mode control also simplifies frequency compensation and improves response.

[theoldwizard1]

More modern current mode switching regulator controllers include cycle-by-cycle current limiting as a side effect of current mode control which both maintains flux balance in a transformer and protects the output transistors.  Current mode control also simplifies frequency compensation and improves response.

Okay, I understand that (although you did not reference a part), but how does this really help the situation ?

With multiple MOSFETs in parallel. if one fails the switching regulator can not sense that, it is only seeing the TOTAL current which the remaining MOSFETs try to maintain.

The only "safe" solution would be to over specify the MOSFETs by a wide margin (100%?) so that if one fails the others can carry the additional load.  Careful thermal monitoring should be able to detect that before additional MOSFETs fail.

[jbb]

With multiple MOSFETs in parallel. if one fails the switching regulator can not sense that, it is only seeing the TOTAL current which the remaining MOSFETs try to maintain.

Yes, generally the paralleled devices are treated as one, large switch.  If you want to treat them individually, you end up with multiple channels, interleaved:

Even better, we might wire multiple inverters in parallel, with independent controls for each channel, so each transistor pair need only handle, say, 20A (and they can do it with lots of safety headroom), while the total delivers, say, 100A or whatever (from 5 channels in parallel).  Knock-on benefits include ripple cancellation, by phase shifting each channel.  This is what PC motherboards do, with upwards of 8 phases!

Downside, the phase shift method may not be applicable for everything (a single phase AC output needs to draw its load current then and there), but even so, the switching edges can still be spread out slightly to help reduce high frequency noise.

The only "safe" solution would be to over specify the MOSFETs by a wide margin (100%?) so that if one fails the others can carry the additional load.  Careful thermal monitoring should be able to detect that before additional MOSFETs fail.

Well, not quite.  Silicon devices tend to fail short circuit, causing large overcorrects.  Often these overcorrects will blow up the complimentary switches. Sometimes these overcurrents are large enough to blow the bond wires off the failed device, leading to an 'open circuit' failure.

[T3sl4co1l]

Blown converters are usually due to lack of control.  What IC do you see on the primary side, TL494?  SG3524?  Yeah, there's your problem.

Why are these inferior and what would be better ?

The controller isn't necessarily bad, in and of itself, but it is almost always used in a bad way.

The standard example, used for car amps, automotive inverters and such, is a TL494 (or similar) driving a wad of transistors at fixed PWM, period.  No control whatsoever, no regulation, no current limiting, nothing.  The only thing that allows it to even start up at all, is a generous soft start (PWM rises slowly), which drops inrush power across the transformer's leakage inductance, in turn dissipating that power in damping resistance (if provided) or transistor switching loss.

The most apparent characteristic, looking at such a design, is this: no filter inductor.


The next step up isn't any better: a voltage mode controller.  This typically has voltage feedback, varying PWM to regulate the output voltage.  This has two problems: 1. no current control, obviously; 2. compensation is a pain, because the LC filter introduces two poles into the loop (basically, the LC circuit tends to resonate, pushing the loop towards instability).


The usual hack is actually exemplified nicely here:
https://www.eevblog.com/forum/repair/tl494-based-82v-lifepo4-charger-bad-part-or-suspect-substitute/msg1447211/#msg1447211
This has current and voltage mode operation; they are cooperative, in that whichever one is closer to its setting, takes over.  This isn't full current mode operation, but it's certainly something.  It should be robust to short-circuiting the output.  (The automotive version would have these changes: 1. TL494 powered from +12V input; 2. MOSFETs instead of BJTs, and push-pull instead of half-bridge; 3. current sense in the primary side ground return, allowing primary side protection; and, any other changes to suit the desired output voltage(s), for example probably using a FWB rectifier instead of FWCT as shown.)

Preferably, current mode would be implemented by using the internal feedback loop exclusively for current regulation, so that, no matter what current the circuit is commanded to deliver, it delivers only that much, under full control at all times.  An external op-amp is added onto this (since, although the TL494 has two error amps inside, they aren't independent -- their outputs are wired-OR'd), and it controls the current command in order to regulate output voltage (or whatever other quantity is being controlled, ultimately).

TL494 isn't really suitable for peak current mode control: you must provide an external flip-flop, that is set by the oscillator (somehow*), and reset by a comparator sensing switch / inductor current.

*The very similar SG3524 enables this with its "OSC" output, which outputs pulses.

So you need to add three chips (latch, comparator, voltage error amp), and the '494 is basically an oscillator, might as well use a 555, or just more logic (indeed, I've used CD4047 for this).

One modernish chip that can do all of this in just 8 pins: UCC3808.  Newer controllers are often overlooked because they're more expensive, but going for the $0.05 TL494 is a false economy when you need reliability and performance.

Tim

[floobydust]

+1 Thanks for the pointing out the UCC3803.

TL494, I don't know how that old dog is still so prolific. I recall the SG3524 has a hardware bug and was not reliable, that's why the SG3525 came out.

[T3sl4co1l]

Hmm, interesting. Anything specific?

3525, and TL598, are totem pole output versions.

Also, there's a UC3524/5 version which is improved (faster, stronger output drive; and I think something about better supporting latch logic?).  And a mess of other UC(C) controllers for various purposes, though they're mostly on the obscure/expensive side.

Tim

[floobydust]

Tim, you make me dust off my noggin.

There is a "bug" in the 3524 "... a common problem ... was that any noise or ringing on the output of the error amplifier would effect multiple crossings of the oscillator ramp signal resulting in multiple pulsing at the comparator's output. The [newer] SG3525's latch terminates the output pulse with the first signal from the comparator, insuring that there can be only a single pulse per period..." SGS-Thomson App Note AN250/1188

The problem I encounter with SG3525 is the output stage cannot supply high currents for most MOSFET loads. The IC does not have a separate GND pin for the output stage and the current spikes get into the IC and it re-triggers in the middle of a PWM cycle. So I always add a gate driver.

[T3sl4co1l]

Ah, not so much a bug as an oversight; the CMP pin is indeed just the PWM comparator input.  You shouldn't have noise there with proper compensation, but introduced noise, that can happen.

Which also means.. you can tweak the CMP pin up to terminate a pulse and have it latch, which means you only need the current comparator to do peak current mode.

Tim

[David Hess]

There is a "bug" in the 3524 "... a common problem ... was that any noise or ringing on the output of the error amplifier would effect multiple crossings of the oscillator ramp signal resulting in multiple pulsing at the comparator's output. The [newer] SG3525's latch terminates the output pulse with the first signal from the comparator, insuring that there can be only a single pulse per period..." SGS-Thomson App Note AN250/1188

The TL494 has the same "bug" which is why the second error amplifier cannot be used to implement cycle-by-cycle current limiting.

[theoldwizard1]

3525, and TL598, are totem pole output versions.

I am a retired automotive electronics (powertrain) controls engineer.  I only has one electronics class in college and I suspect it was before some of you were born.  Having said that I can just barely keep up with the discussion, but you do not have to dumb it down any further.  I "learn" by example, reviewing sample implementations.

The 3525 has 1 error amplifier and the TL598 has 2.  However, I have not found one example circuit that make use of these amps !  Any one have a link ?

[David Hess]

The 3525 has 1 error amplifier and the TL598 has 2.  However, I have not found one example circuit that make use of these amps !  Any one have a link ?

Except for the output stage, the TL598 is identical to the TL594 so the same application circuits apply.

Designing Switching Voltage Regulators With the TL494