Relationship between RdsON and Gate capacitance

[raptor1956]

I've been doing some Peltier module testing using a testing rig I built that includes a circuit to control power to the Peltier module and another to control power to a heating circuit that consists of a couple 5 ohm 10W power resistors.  The rig is driven by a Teensy 3.6 micro-controller that's limited to 3.3V so between the P-channel MOSFETS I'm using to do the final drive I have an NPN transistor for each circuit to isolate the Teensy 3.6 from the 12-16V the MOSFETS see as well as provide more current to drive the gates of the MOSFETS.  I'm using a 1.5kohm pull-up resistor at the gate so that when the NPN output is off the gate is pulled high turning off the MOSFET. 

While running I noticed the power values at the various duty-cycle setpoints appeared to be high so I hooked up the scope and confirmed the MOSFET output duty-cycle to be about 2.4% greater than intended and about 2% higher than the duty-cycle from the Teensy.  The on-time slope was within 1uS but the off-time was more like 13.5uS.  I figured that the gate capacitance was the likely culprit and when I added a second 1.5kohm resistor in parallel with the original pull-up, effectively halving the value, the off-time dropped to about 6.8uS.  This certainly pointed to the gate capacitance as the problem.  On turn on the NPN transistor appears more than capable of charging the gate capacitor within about 1uS but the 1.5kohm resistor is too limiting to shut it off as quickly.  A 100ohm resistor would do the trick as far as speed is concerned, but it would need to handle about 2.5W -- not practical.

OK, with that out of the way the question posed in the title comes to mind -- namely, do low RdsON MOSFETS have higher gate capacitance as a rule and if so is that do to the gate area being greater as a means of lowering the RdsON value?  My brief perusing of a few data sheets seem to suggest that MOSFETS with low RdsON do in fact tend to have higher gate capacitance than do MOSFETS with higher RdsON values.  Again, if this is down to the effective plate area of the gates then this would make sense if low RdsON MOSFETS have larger gate areas.

Now, assuming I have this correct it would appear that low RdsON MOSFETS are a great choice for loads that do not switch ON/OFF very frequently but that for circuits that need to switch frequently, such as a PWM output, that a MOSFET with a somewhat higher RdsON value and lower gate capacitance would be much preferred.  In fact, even with a higher RdsON value if the device turns ON/OFF much quicker it could well dissipate less heat than a lower RdsON device that stays in between ON/OFF longer and therefore operates at higher resistance.

The components used were:

NPN = ZTX618
P-Channel = IPB120P04P4L-03

The Ciss of the MOSFET is listed as typically 11380pF with a max of 15000pF which is much higher than an IRF4905PbF whos Ciss is listed as typically 3400pF.  Interestingly, the 6.8uS turn-off of the IPB120P04P4L-03 when using a pull-up of 750ohm results in a single time constant of 6uS so I calculate the gate capacitance at 8000pF -- lower than the specifications.


Brian

[T3sl4co1l]

Rds(on) goes down and Qg goes up by putting more MOSFET cells in parallel.

So yeah, a ~kohm pullup will take an electrical eternity to turn off.

Tim

[Someone]

Now, assuming I have this correct it would appear that low RdsON MOSFETS are a great choice for loads that do not switch ON/OFF very frequently but that for circuits that need to switch frequently, such as a PWM output, that a MOSFET with a somewhat higher RdsON value and lower gate capacitance would be much preferred.  In fact, even with a higher RdsON value if the device turns ON/OFF much quicker it could well dissipate less heat than a lower RdsON device that stays in between ON/OFF longer and therefore operates at higher resistance.

This is a classic engineering tradeoff, even switching supplies where you are toggling the gate at 10's or 100's of kHz still have an optimisation of on resistance to gate characteristics (total gate charge mostly) which then also flows into the choice of operating frequency and inductor sizing.

The good news for you is that Peltiers are very slow systems so you can afford to turn them on and off relatively gently, seems your homework assignment is to put up the calculations for losses in switching for a range of different MOSFETs and toggle rates and compare this to the static losses in the RdsON and find the minima.

[raptor1956]

Now, assuming I have this correct it would appear that low RdsON MOSFETS are a great choice for loads that do not switch ON/OFF very frequently but that for circuits that need to switch frequently, such as a PWM output, that a MOSFET with a somewhat higher RdsON value and lower gate capacitance would be much preferred.  In fact, even with a higher RdsON value if the device turns ON/OFF much quicker it could well dissipate less heat than a lower RdsON device that stays in between ON/OFF longer and therefore operates at higher resistance.

This is a classic engineering tradeoff, even switching supplies where you are toggling the gate at 10's or 100's of kHz still have an optimisation of on resistance to gate characteristics (total gate charge mostly) which then also flows into the choice of operating frequency and inductor sizing.

The good news for you is that Peltiers are very slow systems so you can afford to turn them on and off relatively gently, seems your homework assignment is to put up the calculations for losses in switching for a range of different MOSFETs and toggle rates and compare this to the static losses in the RdsON and find the minima.

I'm not switching the Peltier ON/OFF using PWM -- I'm pertty much turning it on and leaving it on at rated voltage.  It's the heater circuit that's switched using PWM so that I can vary the heating power during the testing.  So, a low RdsON is just fine for the Peltier circuit but not so good for the heater circuit. 

I need to learn more about gate drivers -- a transistor to turn it on and another to turn it off could provide hundreds of ma or even more than 1000ma to charge/discharge the gate faster.  In my testing rig it's not a huge deal and I can live with the slower turn-off by compensating for the roughly 2.4% extra power using 1.5kohm or 1.4% using 750ohm pull-ups.  Not a big deal for this setup but it's a good thing the PWM frequency is only about 480Hz.


Brian

[Someone]

Now, assuming I have this correct it would appear that low RdsON MOSFETS are a great choice for loads that do not switch ON/OFF very frequently but that for circuits that need to switch frequently, such as a PWM output, that a MOSFET with a somewhat higher RdsON value and lower gate capacitance would be much preferred.  In fact, even with a higher RdsON value if the device turns ON/OFF much quicker it could well dissipate less heat than a lower RdsON device that stays in between ON/OFF longer and therefore operates at higher resistance.

This is a classic engineering tradeoff, even switching supplies where you are toggling the gate at 10's or 100's of kHz still have an optimisation of on resistance to gate characteristics (total gate charge mostly) which then also flows into the choice of operating frequency and inductor sizing.

The good news for you is that Peltiers are very slow systems so you can afford to turn them on and off relatively gently, seems your homework assignment is to put up the calculations for losses in switching for a range of different MOSFETs and toggle rates and compare this to the static losses in the RdsON and find the minima.

I'm not switching the Peltier ON/OFF using PWM -- I'm pertty much turning it on and leaving it on at rated voltage.  It's the heater circuit that's switched using PWM so that I can vary the heating power during the testing.  So, a low RdsON is just fine for the Peltier circuit but not so good for the heater circuit. 

I need to learn more about gate drivers -- a transistor to turn it on and another to turn it off could provide hundreds of ma or even more than 1000ma to charge/discharge the gate faster.  In my testing rig it's not a huge deal and I can live with the slower turn-off by compensating for the roughly 2.4% extra power using 1.5kohm or 1.4% using 750ohm pull-ups.  Not a big deal for this setup but it's a good thing the PWM frequency is only about 480Hz.


Brian

The power resistors will be similarly slow for all the thermal mass around them, the error (if you insist on using open loop control) is proportional to the period of the PWM so you can slow it down 10x or 100x to reduce the effects of the slow edges by that amount. You need to explore the design space and start looking at what the limitations and optimisations are.

[raptor1956]

Now, assuming I have this correct it would appear that low RdsON MOSFETS are a great choice for loads that do not switch ON/OFF very frequently but that for circuits that need to switch frequently, such as a PWM output, that a MOSFET with a somewhat higher RdsON value and lower gate capacitance would be much preferred.  In fact, even with a higher RdsON value if the device turns ON/OFF much quicker it could well dissipate less heat than a lower RdsON device that stays in between ON/OFF longer and therefore operates at higher resistance.

This is a classic engineering tradeoff, even switching supplies where you are toggling the gate at 10's or 100's of kHz still have an optimisation of on resistance to gate characteristics (total gate charge mostly) which then also flows into the choice of operating frequency and inductor sizing.

The good news for you is that Peltiers are very slow systems so you can afford to turn them on and off relatively gently, seems your homework assignment is to put up the calculations for losses in switching for a range of different MOSFETs and toggle rates and compare this to the static losses in the RdsON and find the minima.

I'm not switching the Peltier ON/OFF using PWM -- I'm pertty much turning it on and leaving it on at rated voltage.  It's the heater circuit that's switched using PWM so that I can vary the heating power during the testing.  So, a low RdsON is just fine for the Peltier circuit but not so good for the heater circuit. 

I need to learn more about gate drivers -- a transistor to turn it on and another to turn it off could provide hundreds of ma or even more than 1000ma to charge/discharge the gate faster.  In my testing rig it's not a huge deal and I can live with the slower turn-off by compensating for the roughly 2.4% extra power using 1.5kohm or 1.4% using 750ohm pull-ups.  Not a big deal for this setup but it's a good thing the PWM frequency is only about 480Hz.


Brian

The power resistors will be similarly slow for all the thermal mass around them, the error (if you insist on using open loop control) is proportional to the period of the PWM so you can slow it down 10x or 100x to reduce the effects of the slow edges by that amount. You need to explore the design space and start looking at what the limitations and optimisations are.

Yes, as slow as the response is I could easily slow the PWM frequency down by a factor of 100 without inducing any problems and reduce the effect of the slow turn-off by 100X.  I'm pretty sure there is some register in the Teensy 3.6 that permits changing the PWM frequency -- I'll check that and report back. 

Since I have no experience with IC based gate drivers I would like some advise on an appropriate high side (P-channel) gate driver.


Brian

[jbb]

You could try this. The MOSFET turns on via Q1, D1 and Rg.

When you want to turn the MOSFET off, the 10k pull-up turns on Q2 and turns the MOSFET off again.

[raptor1956]

You could try this. The MOSFET turns on via Q1, D1 and Rg.

When you want to turn the MOSFET off, the 10k pull-up turns on Q2 and turns the MOSFET off again.

Thanks jbb, could you tell me the reason for D2?

What about using an op-amp/comparitor driver with threshold midway between ground and Vcc?


Brian

[pigrew]

Somewhat roughly for long-channel devices, R_DSon is inversely proportional to W/L * C_ox .  The W and L are the length and width of your MOSFET's gate area, and C_ox is the capacitance-density (F/m^2). With a particular process, the length will be determined by your desired drain->source breakdown voltage, so in answering your question, one can assume assume L is fixed. C_ox is also fairly well fixed. The gate thickness is usually reduced to the point that either gate leakage is too high or the gate->S,D breakdown voltages are too low.

The main variable you get to tune is the device width, which will be inversely proportional to your R_DSon. Your gate capacitance is W*L*C_ox. Based on these relationships, the drain to source resistance is proportional to the gate capacitance.

These relationships should be very close to accurate for large power devices, but don't well model short devices (perhaps under 0.1um gate lengths in silicon).

[jbb]

D2 is to clamp the drain voltage of the MOSFET to approx 0V at turn off. Wiresound reistors have inductance, and if you try to switch off the current instantly you get a big voltage spike (aka inductive kickback).

You could try something with a comparator,  but if you’re going to drop a chip it should be a gate driver instead.

[batteksystem]

A transistor to turn it on and a transistor to turn it off is the classic totem pole gate driving. If you match a npn and pnp transistor then you only need one signal to do the control.

[xani]

I'm not switching the Peltier ON/OFF using PWM -- I'm pertty much turning it on and leaving it on at rated voltage.  It's the heater circuit that's switched using PWM so that I can vary the heating power during the testing.  So, a low RdsON is just fine for the Peltier circuit but not so good for the heater circuit. 

I need to learn more about gate drivers -- a transistor to turn it on and another to turn it off could provide hundreds of ma or even more than 1000ma to charge/discharge the gate faster.  In my testing rig it's not a huge deal and I can live with the slower turn-off by compensating for the roughly 2.4% extra power using 1.5kohm or 1.4% using 750ohm pull-ups.  Not a big deal for this setup but it's a good thing the PWM frequency is only about 480Hz.

Brian

You can do a 3 transistor circuit (one for level conversion + push-pull to drove mosfet) to make it faster, but yeah, mosfet drivers are cheap and easy.

OR..... just mount power transistor with the resistors, that way it will just be doing extra heating in the place you want the heat to be, as long as it doesn't drive  MOSFET temp >100 you'll be fine

[Seekonk]

A high side driver will not work unless it switches to common occasionally to do a bootstrap recharge. For low frequency I just use an opto isolator and a resistor. The driving power supply does have to be isolated. I just use a little wall wart.  For quick and simple use a  VOM1271.

[raptor1956]

Well, I found the secret formula to lower the PWM frequency and after lowering the freq from 480Hz to 75Hz the difference between Teensy drive and MOSFET output is negligible.  In case anyone is interested it's:

analogWriteFrequency(pin, Hz);

Couldn't be easier...

Also, I guess I should have realized that wire wound power resistors would have some inductance -- I should add one across them today.

Lastly, I made a mistake in calculating the gate capacitance as I reported in a prior -- I was measuring the output and not the input -- dohhh....

I measure 12uS for the time constant and with a 750ohm resistor that works out to 16000pF -- higher than specified.


Brian

[james_s]

You could always just bit bang it too, if you're controlling a Peltier you would probably never notice the difference between 1Hz (or even 0.1Hz) and 100kHz PWM in the temperature of the Peltier. Industrial style PID controllers typically pulse at about 1Hz.

[raptor1956]

There's many ways to skin a cat and although I feel no need to make any further modifications I did add the suppression diode across the wire-wound resistors.

As to another method to speed to turn-off ... by putting a zener diode between the NPN and the gate of the MOSFET so that when the NPN turns on and pulls the gate down, turning it on, the voltage at the gate will not go down as far so will not have as far to be pulled back up.  In my example the voltage is 15.4VDC and with the current configuration when the NPN turns on it will have about 0.4V drop so that the gate is -15V.  With a zener of, say, 8V the gate would only go down to -7V and even with the same pull-up resistor the turn-off would be less than half as long.  Additionally, since there is less voltage being dropped across the resistor its value could be halved resulting in an overall improvement of a factor of 4 or more.

Fun stuff...


Brian

[raptor1956]

You could always just bit bang it too, if you're controlling a Peltier you would probably never notice the difference between 1Hz (or even 0.1Hz) and 100kHz PWM in the temperature of the Peltier. Industrial style PID controllers typically pulse at about 1Hz.

Again, I turn the Peltier on and leave it on -- I'm not using PWM on the Peltier.  But, I am using PWM on the wire-wound resistors I'm using as heating elements.  But, even 1 Hz would be sufficient though I've opted for 75Hz as it is easier to see on my scope and the bench power supply displays the current and power without jitter.  A slower frequency would make reliable readings from the power supply difficult and it also makes getting stable traces on the scope problematic.

Back in the day, when I worked in the semiconductor division of IBM (East Fishkill), we had a Fluke thermocouple reader that was very accurate, had a solid state 0C reference, but a sampling rate around one complete cycle in about 20 seconds if I remember correctly.  This was connected to a four stack horizontal furnace, yes, that far back, with, I think, 5 TC's per tube or 20 in total.  So, it would read one TC per second or all 20 in 20 seconds.  Given the dynamics of the furnace and its heating element the sample rate, though slow, was fast enough so long as the power was constrained.  In my Peltier module testing rig I am similarly constraining the heating power so it does not over-power the unit as result in over/under shoot.  I need to do a bit more tweaking but at first pass the over/under shoot is 0.13C -- with additional tweaking I can halve that to 0.06C (DS18B20 at 12-bit).


Brian