P-Channel MOSFET Gate Driver, 15kHz PWM

Thread Starter

newothegreat

Joined May 3, 2012
26
I am designing (rather redesigning) an H-Bridge circuit to drive TEC modules, and I am having trouble coming up with a suitable gate drive circuit for the P-Channel MOSFETs on the high side. I've found some integrated gate drivers for the N-Channels from Diodes, Inc, p/n ZXGD300xE6 (x = 1, 2... 6) that seems like they will work fairly well, but there is no equivalent for driving P-Channels. The P-channels have a source voltage of 24V, so I can't just drive the gates to ground because Vgs max is 20V, so I need to somehow vary the gate voltage between 24V (Vs) and about 12V. Here are the specs of my circuit:
PWM_f = 15.625 kHz
PWM_res = 6-bit
smallest PWM "on" pulse length = 1.00us (1 / 15.625kHz) / (2^6)
Vsupply = 24V
Imax (TEC) = 12A
Pchannel gate capacitance = 1200pF
Pchannel max gate charge = 63nC
PWM controller = PIC @5V
PWM voltage = 5V

Since the smallest pulse length is 1us, I'd like to be able to fully turn on and turn off the MOSFET within 200ns, which means my driver needs to be able to sink and source about 300mA (63nC/200ns). The previous circuit had a small signal n-channel mosfet (2n7000) with two 10k resistor in series connected between the drain of the 2n7000 and the 24V rail. This was way too slow since the turn on time was around 10us and the turn off time was like 20us.
It seems like I need some sort of "totem-pole" output stage on the driver in order to properly charge and discharge the gate, but I haven't seen an circuits that don't discharge the gate directly to ground, which, in my circuit's case, would exceed the Vgs max rating. Is there a simple way to discharge the gate into 12V somehow? (indepenent 12V regulator?, zener between totem pole negative rail and ground, etc...?)

I'm sure some people want to tell me to use an N-Channel as a high side switch, but I am looking for information for using a P-channel.

The P-Channel part number is IRF5305PBF.

Thank you,
Owen
 

Attachments

John P

Joined Oct 14, 2008
2,025
You can set up a "pseudo-ground" at some suitable level between 24V and real Gnd. Just do it with a voltage regulator. Obviously, you'll then have to translate between control signals referenced to true Gnd and your intermediate level, but that's usually simple.
 

takao21203

Joined Apr 28, 2012
3,702
20K is indeed a bit high for such a gate drive.

I use a 200 Ohms/800 Ohms for ~50V (p-CH), that in turn is grounded by a smaller BJT (not 2n3904 small signal).

I use 3x 1W resistors in parallel, and after 2 months, they are pretty much carbonized (the coloring). The resistor engraving is visible.

Also works pretty equally for a power PNP. I can swap the MOSFET with a PNP directly.

Don't know if this gives you 200 nsec. I don't even think. I use that for max. 150 KHz or so.
 

Thread Starter

newothegreat

Joined May 3, 2012
26
Thanks guys. I think I'm going to create a "pseudo-ground" like John_P suggested. Using such low value resistors just to create a gate voltage will probably waste too much power/board space. Is it really just as easy as slapping on a 7812 with output caps on the 24V rail? Would there be a way to implement the 12V pseudo ground with a zener? The only thing I am worried about with using a voltage regulator would be the transient response, since it needs to be able to supply hundreds of milliamps for very short ns scale pulses.
 

takao21203

Joined Apr 28, 2012
3,702
I saw a zener used in a linear MOSFET supply. Again you'd need a resistor here I think towards ground. The 7812 also is a hidden resistor.

My circuit is 300W class and maybe 3W wasted- 1%. On lower loads the duty cycle is not that much and they don't heat up as much. Something I can live with, having a large 500VA toroid.

Having a 36V transformer and 32V LEDs + a PNP BJT, I can raise the voltage to 100% and the LEDs will actually not burn or heat critically (also dropping several volts in the cable and inside the storage coils).

Such circuits are often a question of dimensioning, which must be adjusted carefully.
 
Last edited:

Thread Starter

newothegreat

Joined May 3, 2012
26
Yeah those are good points, takao. I guess what I'm really concerned about is board space. Since this is a redesign, it's gotta go on the original footprint and I'm very quickly running out of room. I don't really have space for a couple of big toasty 1W resistors, especially since I would need to give them some room to themselves to cool off.
This circuit could potentially use duty cycles close to 100%, so if I were using two resistors in series with a BJT, I would be constantly dissipating that 3W, whereas if I used an intermediate supply at 12V or so, I could only dissipate that 3W during the 200ns I charge up or discharge the gate every 64us.

Nothing wrong with your circuit, I just don't think it will work for my purposes, unless you can convince me otherwise.
 

John P

Joined Oct 14, 2008
2,025
Your FET gate might need 300mA of current, but only briefly. With a good-sized filter, a small regulator should be enough. You also need to plan for which way the current will flow--would it be going from 24V to the 12V level, or from 12V to Gnd? If it's the former, then I think you actually need a -12V regulator, where your pseudo-ground would be at (24 minus 12) volts.
 

takao21203

Joined Apr 28, 2012
3,702
Yeah those are good points, takao. I guess what I'm really concerned about is board space. Since this is a redesign, it's gotta go on the original footprint and I'm very quickly running out of room. I don't really have space for a couple of big toasty 1W resistors, especially since I would need to give them some room to themselves to cool off.
This circuit could potentially use duty cycles close to 100%, so if I were using two resistors in series with a BJT, I would be constantly dissipating that 3W, whereas if I used an intermediate supply at 12V or so, I could only dissipate that 3W during the 200ns I charge up or discharge the gate every 64us.

Nothing wrong with your circuit, I just don't think it will work for my purposes, unless you can convince me otherwise.
It is really a question of dimensioning. The current you need to clamp correspondends to the required on/off times.

It is possible without resistors, but you need a PNP (or p-ch) in the upper branch, and a NPN to the artificial rail. As well you need to take care pretty well about the dead time.

This is so complicated, they make driver ICs for such a purpose.

Resistors and clamping with a BJT is simpler in terms of designing, modifying, and circuit building. Might not be the best approach for a small PCB.

You'd have to do the maths what resistors you need for 15 KHz. Why 200nS? Pretty low for 15 KHz only, of course I understand you want a sharp response.

If you'd encounter that a large 5W aluminium resistor is the proper thing, a driver IC is better. I have not built this in detail, but I think it is possible to do it with additional BTJs and an artificial rail.

As for my circuit, an aluminum resistor would be the least of my worries. Also having a large VGA cooler, accomodating 3 semiconductors (the 7805 for voltage reference, the flyback diode and the large BJT/MOSFET. Quite tricky t place them so the cooler plane aligns evenly. I had largely to deal with such mechanical problems.

PCB estate and layout in the end also are kind of mechanical problems.
 

takao21203

Joined Apr 28, 2012
3,702
I looked again the details, what you call NEW CIRCUIT?? already outlines the circuit with 2 additional BJTs. The thing is, what I believe is BJTs don't neccessarily have equal ON/OFF response. So if you don't accomodate a dead time, you'll create a short, possibly just creating heat (12V), possibly destroying the BJTs.

If you have an oscilloscope, you could observe the waveforms in the test circuit.

Eventually a smaller resistor to limit the power that is shortened due to lack of dead time is already enough. Also depends on the BJTs.
 

Thread Starter

newothegreat

Joined May 3, 2012
26
The reason I spec'd 200ns is because, though the PWM freq is at 15kHz, the resolution is 6-bit (well originally 8-bit, but that's another modification I'm going to make) so the shortest pulse would 1us. I felt like 200ns on and 200ns off was a reasonable spec since that means the FET would spend 400ns of its 1000ns pulse in transition. I suppose I could create a "minimum duty cycle/pulse length" condition in the code and get by with a longer transition time, but create some numbers from somewhere when I created this thread.

Oh I have plenty to deal with in terms of layout. The four MOSFETs and four flybacks already take up over 3 sq inches of my 8 sq inches I have to work with. After you put on the PIC, miscellaneous circuitry, the beefy 15A connectors and the various signal connectors, I can barely fit anything else.

I'll probably end up making that artificial 12V rail, and treat it as ground for one of those integrated gate driver. The problem I was having, was that I couldn't find a gate driver that worked on P-channel FETs when Vs > 20V. If someone knows of one, please tell me. And if any semiconductor companies are reading this thread, you have your next project.
 

takao21203

Joined Apr 28, 2012
3,702
The reason I spec'd 200ns is because, though the PWM freq is at 15kHz, the resolution is 6-bit (well originally 8-bit, but that's another modification I'm going to make) so the shortest pulse would 1us. I felt like 200ns on and 200ns off was a reasonable spec since that means the FET would spend 400ns of its 1000ns pulse in transition. I suppose I could create a "minimum duty cycle/pulse length" condition in the code and get by with a longer transition time, but create some numbers from somewhere when I created this thread.

Oh I have plenty to deal with in terms of layout. The four MOSFETs and four flybacks already take up over 3 sq inches of my 8 sq inches I have to work with. After you put on the PIC, miscellaneous circuitry, the beefy 15A connectors and the various signal connectors, I can barely fit anything else.

I'll probably end up making that artificial 12V rail, and treat it as ground for one of those integrated gate driver. The problem I was having, was that I couldn't find a gate driver that worked on P-channel FETs when Vs > 20V. If someone knows of one, please tell me. And if any semiconductor companies are reading this thread, you have your next project.
You can invert the polarity eventually with some trick.

I am not a university expert, only built some circuits for learning and for research. I am somehow limited to deal with power electronics, say I would not really want to produce a circuit like you are designing now. Even if I am maybe capable to do it.

In other words, I have plenty BJTs here of all sizes, MOSFETs, and I even ordered gate driver ICs (but never actually used them in the end).

My main piece of work was a TL494 circuit from an old Thailand schematic, and it is actually used now and works quite well.

I am rather into microcontroller firmware.

So don't take offense my poor advice others will pick up eventually.
 

takao21203

Joined Apr 28, 2012
3,702
Maybe you could mix the analogue feedback and the digital PWM signal.

Is the operation point a preset, or is it changing from a circuit input?

Since this is for a motor do you have the RPM in a register?

Do you want to control it for a certain RPM and do you have a load variation so must maintain a regulation feedback for that?

Just some ideas when I think about the circuit.

I mean if you only want to preset via the 6bit value, you can maybe mix it with analogue and get higher resultion. Hard to say without to know the details.

Or add analogue dead time control.
 

Thread Starter

newothegreat

Joined May 3, 2012
26
The H-Bridge is driving a Peltier module and it's using a thermistor as the temperature feedback source. The thermistor voltage is read by a PIC and the firmware does all the PID calculations, along with deadtime control and PWM output. That stuff was already developed by the time I was given the circuit, thankfully. Thanks for the input though.
 

takao21203

Joined Apr 28, 2012
3,702
A peltier module? Isn't that a heat transport component?

In this case you don't need 15 KHz, reduce it to much lower value- done with the 200nS problem.
 
Top