Fast(est) shoot-through protection

Hello,

I want to design a shoot-through (and short-circuit) protection for a Mosfet full-bridge.

Currently, I have a LEM current transducer on the positive rail (DC-link) and I'm reading its output every 100us using an ADC (for power monitoring purpose only).

Anyway, I don't think that polling interval is short enough to save the Mosfets on a short-circuit/shoot-through condition.

As the current transducer has a detection time of around 7us, I thought of using an analog comparator (op amp) to continuously check the transducer output voltage against a manual-adjusted threshold (maximum allowed current).

That way, it could disable the Mosfet drivers faster than any software detection (ADC/MCU).

Here comes my question: is that delay (7us) short enough to protect the power Mosfets?

Maybe using a shunt resistor would be faster but: (1) I prefer a non-invasive solution and (2) the Mosfet current is quite large (100A) thus I'll need a shunt resistor with a very low value (1mOhm?) so any parasitic resistance (pcb/wires) will affect the current readings.

Thanks in advance for any suggestion.

Here comes my question: is that delay (7us) short enough to protect the power Mosfets?

Click to expand...

I think that totally depends upon your MOSFETs, so as to how long they can withstand the direct short. Other factor affecting the protection aspect during shoot through can be the dissipation at that time due to 2I²R_ds(on) and whether the thermal resistance is low enough to absorb the heat in that short span and dissipate it quick enough to save the MOSFETs.

Protection time also varies with environmental conditions:
For shoot through at startup, probably the differential temperature between the MOSFETs and heatsinks will be high hence, the heatsinks may dissipate power and gives the devices time till the shoot through protection kicks in.

If shoot through occurs during running conditions, the differential temperature between junction and heatsink will be low, hence it may not dissipate it as quick as required and may thus cause failure.

Maybe using a shunt resistor would be faster

Click to expand...

You cannot keep the respose time so low that it reacts to high di/dt probably caused during startup of a non-inductive load (if present). Adding a time constant will save you from false triggering, and hence I think that it would be ideal to use a LEM sensor

7us is likely much too slow to save the FETs from a short circuit condition. A current transformer is probably your best option for detecting the extremely high di/dt pulses resulting from shoot through.

Besides studying MOSFET SOA (safe operation area) you should also determine the expectable short circuit current and current rise time. Shoot-through protection (protecting against instant short with "zero" rise time) is hard to achieve for low voltage FETs, you have a slightly better chance to turn of a high voltage FET before it's destroyed by the high short circuit current.

Gate drivers have often interlock means so that shoot through scenarios caused by simple cases of erroneous control signals can be avoided. Unfortunately there are cases like fast toggling gate signals or MOSFET self oscillations that can put the FETs into linear operation with high power dissipation but aren't necessarily detected by overcurrent sense circuits.

Load shorts can be well managed with current sensors if the output circuit has a certain series inductance.

I forgot to mention that I"m using a soft-start procedure and the PWM signals already have dead time insertion thus there"s no much to worry about shoot-through during normal operation.

My only concerns are about unpredictable situations, when the PWM generator (MCU) fails to output the correct modulation scheme (due to a software glitch or a latch-up) or when a h-bridge switch get damaged (fails short) hence no dead time could save the complementary switch from dying (shoot-through).

I'm using six Mosfets (in parallel) for every switch so there will be a chain reaction if one of them dies (short) unexpectedly.

So do I need a reaction time shorter than (let's say) a microsecond for an effective protection? Is there some kind of di/dt parameter (for a Mosfet) to take into account when designing such a reaction time?

Oh, I forgot about current transformers (as I want to avoid invasive detection methods and I thought transformers are not the fastest devices anyway).

Is there a 100A current transformer available or should I manufacture one? What about their efficiency (power loss during operation)? Now that you mention, I think I prefer a current transformer over a shunt resistor due to its isolated output.

EDIT:

@FvM, I've just seen your reply. You've actually summarised all those unexpected failure actions I was afraid of.

What's the MOSFET Vds range?

If internal circuit failures are your main concern, desaturation detection may be more effective than overcurrent detection.

Vds parameter is 200V and the h-bridge rail voltage is 48V (40-60V, actually).

I've just encountered a strange MCU behaviour due to a transient voltage on its power supply (due to a loose wire connection) and all Mosfets of one bridge's leg have died in that instant.

Maybe there was a PWM pulse latch-up or the transformer core saturation due to unbalanced parasitic PWM signals.

I had a separate circuit for dead time (300ns) insertion and complementary signal generation thus a shoot-through should have been avoided.

I had four Mosfets in parallel for every switch, each one rated at 130A. The load was moderated (10A) at that moment thus there was no thermal issue.

I've always thought that once a member of a parallel Mosfet group fails short, it should protect (by-pass) the other Mosfets within its group. I wonder why it didn't actually happened.. all Mosfets from a group were dead short (D-S).

Of course, the fuse (120A) has tripped once the smoke has disappeared..


PS: I'm talking about a LF transformer based sine wave inverter.

Presumed you can rely on complementary gate control and dead time generation, you should a least have an output overcurrent recognition.

My take on this is that during power up and power down (or severe brownout) in the PWM driver logic, conditions may not always be well defined.

The whole thing needs to be very carefully thought through, especially when multiple independent supply rails are involved.

FvM mentions using a gate interlock system, and that is my preferred solution also.

I have had success with using opto isolator gate driver chips such as HCPL3120 and connecting the LEDs in direct inverse parallel.
It is then simply impossible for both LEDs to be be on together at the same time.
Its a very simple and robust solution.

A separate logic driver is connected to each end of the LED combination through a current limiting resistor.
Each driver is fed with normal alternately phased PWM with appropriate dead time.
If both ever try to go on together, there will be zero volts across both LEDs.

If you are really after these remote possibilities, then I'd say
you want something like a latched overcurrent protection
(after all, a failed PWM or failed MOSFET isn't going to get
better) and maybe you can make use of existing current
control feedback with a comparator and latch, use the
other MOSFET driver input (?) for a disable port (many
have an A, B/ pair and often use only one).

However protection fast enough to be useful, and
sensitive enough to run directly off current sense
resistor / winding, may be more nuisance than help.

I think a values analysis might be in order; given that
the piece has already failed, what do you gain by
"passivating" its failure for the indefinite future? And then,
how much do you stand to gain and therefore how hard
and how expensively should you try?

And if that is indeed the goal, what's wrong with a
simple fuse?

A simple fuse is actually a pretty good suggestion.
But it may need some help.

A fuse robust enough to carry full load current reliably may not blow fast enough under sudden fault conditions to save the mosfets.
What may help is providing enough deliberate inductance in the load circuit to slow down the rate of rapid current rise (under fault conditions) to something that may be manageable.
This may need to bee an air cored inductor.

Then using an over current crowbar circuit to blow the main supply fuse as quickly as possible, independent of the mosfets.

Something quite crude such as a large SCR and current transformer can be surprisingly fast acting and effective.

@Warpspeed:

I'm also using optocouplers for every switch (high-side/low-side) and a separate circuit (logic gates) to insert the dead time and to generate the complementary PWM signal for every h-bridge leg.

That's it, there should be no shoot-through condition whatsoever (theoretically).

There is an exception though: when one switch fail short, and the complementary one it's still driven by the PWM signal. This is where my protection lacks.

I guess I have to design a simple and reliable circuit to detect if the complementary switch is shorted before driving the actual switch.

Anyway, there comes those built-in freewheeling diodes that keep the switch almost shorted during dead time period.

The only difference is that you read a negative voltage (1-2V) between Mosfet drain and source when the freewheeling diode is conducting while a shorted switch should read 0V or a small _positive_ voltage between those two terminals.

But how to reliable detect this at that switching frequency??

@dick_freebird:

Yes, a latching protection is what I'm looking for. I don't mind if there's a false triggering now and then, that bunch of Mosfets is more pricey than a small power outage.

The problem with fuses is their slow reacting time; you really can't have a fuse to operate normally at 100A then burn in an instant at 200A or so.

I had a 120A fuse that tripped after the Mosfets went on fire. I guess its only purpose is to manually disconnect the inverter from the battery.

Warpspeed said:

My take on this is that during power up and power down (or severe brownout) in the PWM driver logic, conditions may not always be well defined.
I have had success with using opto isolator gate driver chips such as HCPL3120 and connecting the LEDs in direct inverse parallel.
It is then simply impossible for both LEDs to be be on together at the same time.
Its a very simple and robust solution.

Click to expand...

Clever, very clever.

This depends on the voltage of the DC link to my mind. I my experience a LEM is not fast enough to be able to detect the current, as their bandwidth tends to be about 100kHz, which comes out as being 10us. Vds desaturation as implemented in various HP/Avago gate drive ICs will always give you the best protection, although you will have to do a round of pulse tests whereby you use a DSO and a signal generator and actually subject the circuit to abuse, including shoot through. That is the best and frankly only way to have any confidence in an inverter. I used this approach on a 500kW inverter I designed for someone a couple of years ago.

Have a look on the Semikron site for info on pulse testing. I also mentioned pulse testing in another of my posts a few months ago from memory. Look for 'industrial grade inverter' in my posts.

I've read some documentation about the desaturation fault detection but it seems to be hard to implement for Mosfet based low voltage h-bridges (the difference in Vds voltage for steady-state operation and fault (short) condition being very small).

It just crossed my mind a more simplified variant of such a circuit. What if I'll design a Mosfet based switch (very high current rated, by using 4-6 power Mosfets in parallel) and insert that switch between the h-bridge and the ground rail?

This way, I'll have to design a single desaturation detection circuit (for this single switch). Being ground referenced, it should be easier to read the Vds voltage during conduction state. Moreover, that switch will stay in conduction state (no switching) during normal operation thus the perturbation should be minimal.

Knowing the Rdson and the rated h-bridge current, I could calculate a safe operating range for that Vds.

Using an analog comparator (for faster reaction time) I could generate an early alarm for the MCU and/or I could disable the switch, to protect the whole h-bridge.

The only downside is the switch conduction losses but by paralleling a bunch of them I could reduce the losses to an acceptable amount.

Do you find it a feasible idea?

Are there any downsides for the proposed circuit?

I'm only interested in sensing the peak current using this "electronic fuse". I've did some quick math on this: the four Mosfets have a 10mo Rdson each hence the resulting conduction resistance would be (roughly) 2.5mo.

If I set the trip point at 200A, that's a Vds of 500mV. It should be easier to detect such a voltage threshold using an usual instrumental op amp (comparator).

As the Mosfets will only stay in conduction mode, there should be no parasitic voltage across Vds thus the filtering would me minimal (to enhance the reacting time).

Any comments are highly appreciated.

- - - Updated - - -

There could pe a problem at start-up, if the load (the inverter) is short-circuited.

As I have to add a delay in Vds detection while the switch turns on (to allow Vds to go down from h-bridge dc-link voltage to ground), there could be a masked overcurrent condition.

What's the Mosfet behaviour when it's turned on in a short-circuit condition (i.e. a Mosfet placed right across the power supply terminals)?

There is a voltage sag before Vds is stabilizing at Rdson * Id value?

I guess I'll beter test for a short-circuited load before turning on the electronic fuse (by injecting a constant current into load and read the output voltage).

What about a shunt in the +DC connection, before the de-coupling capacitors? What is your normal working current and DC link voltage?

You are experiencing now that it's difficult to implement "desaturation" (BJT/IGBT terminology) detection with low voltage MOSFETs, as mentioned before. A feasible implementation would probably go for a higher threshold voltage than 0.5 V and fast detection (e.g. 5 to 10 µs). Did you check the SOA for a safe threshold voltage?

I wonder why are you focusing on load shorts now? Load overcurrent protection, if it's intended to keep safe current limits for the load will most likely require a separate current sensor.

brushhead said:

What about a shunt in the +DC connection, before the de-coupling capacitors? What is your normal working current and DC link voltage?

Click to expand...

The DC link voltage is 48VDC (40-60V, actually) and the normal working current (rms) is around 100A.

In my current implementation, every h-bridge switch has four 200V/130A/10mo Mosfets in parallel, but I'm interested in making an adjustable electronic fuse.

FvM said:

A feasible implementation would probably go for a higher threshold voltage than 0.5 V and fast detection (e.g. 5 to 10 µs). Did you check the SOA for a safe threshold voltage?

Click to expand...

As the electronic fuse's Mosfets are operating in continuous conduction mode, could it be that hard to detect a 500mV Vds peak voltage?

FvM said:

I wonder why are you focusing on load shorts now? Load overcurrent protection, if it's intended to keep safe current limits for the load will most likely require a separate current sensor.

Click to expand...

I'm interested in load short condition because I'd like to design a fail-proof electronic fuse.

As mentioned above, a current sensor (Hall based, from my experience) might be too slow. I still believe that a large bandwidth/high slew rate comparator could better detect a Vds threshold of a Mosfet in a steady conduction mode.

- - - Updated - - -

To make myself clear: I'm not talking about "desaturation" detection of a switching Mosfet but the Vds peak detection of a Mosfet in a steady conduction mode.

- - - Updated - - -

brushhead said:

What about a shunt in the +DC connection, before the de-coupling capacitors?

Click to expand...

I want to be able to disconnect the load in a short-circuit condition hence my preference for an "active" shunt (e-fuse).

To make myself clear: I'm not talking about "desaturation" detection of a switching Mosfet but the Vds peak detection of a Mosfet in a steady conduction mode.

Click to expand...

Don't see a difference. Usual "desaturation" circuits are covering both cases.

There are hall sensors with < 10 µs (even down to 1 or 2 µs) response time.