Author Topic: driving a high current mosfet at 10's of khz  (Read 2854 times)

0 Members and 1 Guest are viewing this topic.

Offline lorenzop

  • Newbie
  • Posts: 4
  • Country: us
driving a high current mosfet at 10's of khz
« on: February 15, 2017, 05:40:46 PM »
I'm attempting to design my first buck converter, and as I tend to go overboard with my projects, this one is a bit excessive too. I have some information and pictures here http://blog.poixson.com/?page=1000w The transformer has an independent pair of 10 volt secondaries. I used to use this unregulated with the secondaries in parallel, but wasn't happy with the voltage drop at high current draw. My plan is to connect the secondaries in series to get 20-22 volts and regulate it down to 12 or 14.5 with a buck converter, 3 phase at around 25-40khz. Normal operation should see around 50-70 amps from the transformer and 100+ amps output from the power supply. I also hope to keep it fan-less, but I have yet to see how much heat will be produced under various loads.

My question.. how much current will the mosfet gates draw at this frequency, and will the driver chip I planned to use handle it? The mosfets I have are IRF3205 and planned to drive them with IR2104 which I already have. As this is my first buck converter, I'll likely make mistakes until I get it right. Driving the gates, I've come up with figures around 2 amps for each. This seems much higher than it should be. Each driver chip is also driving 2 mosfet gates, for the high/low side. I don't see anything in the datasheet for the driver which says how much current it can handle. I'm assuming I should supply 12 volts to the driver chip, which will then pump charge it slightly higher than that for the high side gate. Will this work as I intend, or am I doomed to fail? Do I have any better options?
 

Offline james_s

  • Super Contributor
  • ***
  • Posts: 2772
  • Country: us
Re: driving a high current mosfet at 10's of khz
« Reply #1 on: February 15, 2017, 05:53:51 PM »
If this is your first buck converter, you might consider starting with something a bit smaller and simpler until you get the hang of it. It is not trivial and it's really easy to let out the magic smoke, especially when dealing with the sort of currents you're talking about. Since the mosfet gates are capacitive the current to drive them can be higher than you might expect, increasing as the switching frequency increases. You want to make sure they are being driven hard too so they switch rapidly and don't spend much time in the linear region where dissipation is high.

Do you have a specific use in mind for this thing? If you just want 12VDC with lots of Amps, you can't beat hot swap server power supplies. They're dirt cheap on ebay, the surplus market is flooded with them from decommissioned servers. They are typically 12V at 30-100A and had a load share pin so they can be ganged in parallel, or with minor modifications they can run in series. I bought four 675W units a while back for $8 each including postage, I use them to power the chargers for my RC airplanes.
 

Offline DBecker

  • Regular Contributor
  • *
  • Posts: 179
  • Country: us
Re: driving a high current mosfet at 10's of khz
« Reply #2 on: February 15, 2017, 06:08:33 PM »
Those are modest currents and voltages, and the frequency isn't too challenging.  You'll still be learning a lot about the physical design and unexpected sources of power consumption.

You'll find that you won't want to switch the MOSFETs as quickly as possible.  You'll need to tune the shut-off time to control the inductive spike, and independently tune the turn-on gate current to deal with the Miller effect.

The current through the gate driver is pretty easy to calculate.  Look at the gate charge at the target gate voltage and drain voltage, multiply it by the frequency.  A high frequency results in high current and high losses

The power dissipation of the gate driver is a bit more challenging.  In general you want a low resistance gate driver so that the power is being dissipated through the gate resistors rather in the gate driver.
 

Online tautech

  • Super Contributor
  • ***
  • Posts: 9925
  • Country: nz
    • Taupaki Technologies Ltd.
Re: driving a high current mosfet at 10's of khz
« Reply #3 on: February 15, 2017, 06:11:55 PM »
To further emphasise what others have said: Hammer the gates.  ;)
Avid Rabid Hobbyist & NZ Siglent Distributor
 

Offline lorenzop

  • Newbie
  • Posts: 4
  • Country: us
Re: driving a high current mosfet at 10's of khz
« Reply #4 on: February 15, 2017, 06:18:03 PM »
Thanks for the suggestions. It's a general purpose power supply, I sometimes use it to power large car audio amps, or whatever other use I might come across. I'm putting the time and effort and money into the project more-so for the learning experience. If I'm able to sort out this design and get it working how I want, smaller buck converters should be no problem for me. I do have a smaller project where I'm building a mppt charge controller for solar power, but it's off-season and not much sunlight for testing, so that is set aside until spring.

What might my other options be.. Should I consider using mosfets with a lower current rating, and hence a lower gate capacitance? Originally I was thinking, rather than using a driver chip, maybe I could use a pair of smaller mosfets to drive the gate of the switching mosfet to a hard high or ground, but I haven't come across anyone else doing something like this yet.
 

Offline DBecker

  • Regular Contributor
  • *
  • Posts: 179
  • Country: us
Re: driving a high current mosfet at 10's of khz
« Reply #5 on: February 16, 2017, 04:18:24 AM »
To further emphasise what others have said: Hammer the gates.  ;)

With high current, don't plan on hammering the gates.  The inductive spikes you get will be nasty.

When I designed a controller for a series DC motor we had to slow down the gate drive to keep the inductive spikes below 20V on a 200 amp battery current.  Part of the challenge was because of the large loop area of the battery cables.  As part of the solution we used several different types of capacitors in the capacitor banks.  The diode reverse recovery time might have been an issue, but that was much easier to estimate.

When you get into physically large structures you also have to be concerned about overshoot on the gate voltage with a hard gate drive.  This is easy enough to measure and control once  you know it's a risk.
 

Online blueskull

  • Supporter
  • ****
  • Posts: 7764
  • Country: cn
  • Power Electronics PhD Candidate
Re: driving a high current mosfet at 10's of khz
« Reply #6 on: February 16, 2017, 04:46:19 AM »
The way I will do it is to build a separate gate driver board with isolated power supply, signal isolator and actual gate driver. This board is to be connected to gate as close as possible.
Then, using common cables like ribbon cable or whatever you fell like to connect the gate driver board to your control board.
On one hand, it solved high side isolation problems, on the other hand, it solved gate driving inductance issues, while it also solves ground noise issues.
SIGSEGV is inevitable if you try to talk more than you know. If I say gibberish, keep in mind that my license plate is SIGSEGV.
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 9318
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: driving a high current mosfet at 10's of khz
« Reply #7 on: February 16, 2017, 04:47:58 AM »
Do you have an inductor chosen for this yet?

Low frequencies need large inductors.

I don't recommend "hammering" the gates, or running much frequency at all (maybe only 10kHz, despite the audible noise it will produce).

All that wiring looks quite high impedance, compared to what load you want to run on it.  That gives plenty of room for inductive flyback spikes when the transistors switch off.  You will need that energy to be absorbed by the transistors, by switching off slowly enough.

Impedance, meaning, the impedance at AC frequencies.  It's quite low resistance at DC, of course.  But an equivalent inductance of only 0.1uH (which comes from a single loop of just a few inches of wire) will blow out your transistors if they switch off in less than, oh, 0.2 microsecond?

To avoid this, you must have supply bypass capacitance, and the buck diodes, very close to the transistors.  That keeps the switching loop short.  To keep the loop to a low impedance, you must construct the supply and switching terminals from parallel plates, very wide and very close together: this makes a low impedance parallel plane, which will have low equivalent inductance.

You've chosen a somewhat regrettable introduction here, because the current is vastly higher than what is convenient for easy wiring and easy switching.  If you had went for a 1kW power supply that runs from 320VDC (rectified 240V mains), the switching impedance would be much higher (i.e., on the order of 300V / 3A = 100 ohms), so the wiring would be much easier to route out.  Of course, high voltages have their obvious drawbacks, too, so you're probably better off learning this transistor-popping lesson, rather than a heart-popping one. :P

Tim
Seven Transistor Labs, LLC
Electronic Design, from Concept to Layout.
Need engineering assistance? Drop me a message!
 

Offline MagicSmoker

  • Frequent Contributor
  • **
  • Posts: 752
  • Country: us
Re: driving a high current mosfet at 10's of khz
« Reply #8 on: February 16, 2017, 06:49:56 AM »
My question.. how much current will the mosfet gates draw at this frequency, and will the driver chip I planned to use handle it? The mosfets I have are IRF3205 and planned to drive them with IR2104 which I already have.

If the MOSFET datasheet gives you the total gate charge spec, Qg, (usually in nanoColoumbs (nC)), then you can estimate* peak gate current required by dividing Qg by transition time in ns. E.g., IRF3205 has a Qg of 146nC, so to switch it in 300ns you would need about 0.5Apk from the driver.

IR2104 is a pretty weak driver, btw. I really don't like bootstrap/charge-pump type high side drivers, but if I had to use one - and they are certainly attractive in this kind of application - then I'd go with something capable of at least 1A peak and use the gate resistor to set the transition speed (rather than rely on current-limited slewing in the driver IC).

Looking at your blog I have to concur with Tim, et al., that you definitely don't want to try to set any speed records here driving your MOSFETs. Go for maybe 10kHz switching frequency and a transition time in the 300ns+ range. A few other random tips:

1) Do NOT use super/ultra capacitors in switchmode converters because they have very high ESR and therefore extremely poor tolerance of ripple.

2) To help ensure dynamic current sharing among several paralleled MOSFETs it is best to use a single driver IC with individual gate resistors; next best is to use several or individual driver ICs and gate resistors; absolute worst is to simply tie all the MOSFET gates together and drive them with a single resistor/driver.

3) The more inductance (read: long wires) between the input capacitor and the MOSFET drain(s) in a buck converter, the higher the voltage spike produced at turn-off for a given switching speed/current according to the equation: Vspike = Lstray * (dI/dt), where Lstray is around 20nH/inch. For example, if each MOSFET draws 20A from the input capacitor with a switching time of 100ns and a stray inductance of 100nH then the spike produced will be 20V, and note that this adds to the input voltage.

4) Similarly, the wiring between the MOSFET source and freewheeling diode (FWD) cathode should be as short as possible (this is called the switching node, because it is where current rapidly switches back and forth between the MOSFET and FWD). However, the wiring from this node to the buck inductor is much less critical (it does have a high dV/dt, though, so can be a copious radiator of EMI).

5) You can use a single buck inductor, or you can use individual buck inductors (ie - join the output of the buck converters together at the output capacitor). If you use individual buck inductors the you can then stagger their switching (run the same duty cycle for each) so that the ripple currents on both the input and output partially cancel out. This is probably a bit advanced for you right now, but you'll almost certainly find it easier/cheaper to use individual inductors for each MOSFET/FWD anyway.

6) Do not bother with synchronous rectification (ie - using a MOSFET driven in antiphase for the FWD). SR is most helpful when the switch duty cycle and/or output voltage is rather low (e.g. - duty less than 30% or so and/or output voltage of 5V or less).


* - note that the Qg spec assumes that the gate is driven by a specific voltage, usually 10V. If you were to drive the gate with, say, 15V then Qg would go up 50%.
« Last Edit: February 16, 2017, 08:19:47 AM by MagicSmoker »
 

Offline james_s

  • Super Contributor
  • ***
  • Posts: 2772
  • Country: us
Re: driving a high current mosfet at 10's of khz
« Reply #9 on: February 16, 2017, 08:15:23 AM »
I built a few solid state Tesla coils in the 500-800W range and they needed fairly beefy gate drivers but then they also run in the 200-300kHz range and the voltage was much higher and current lower. I used transformer isolated gate drive on those and it worked well, after some experimenting with the gate resistors I got reasonably nice gate drive waveforms. With a buck converter you may have issues if you were to try transformer gate drive though, it can be done but you can't get too close to 100% duty cycle. I do like the isolation it provides though, if the mosfets blow up it doesn't cascade and kill the rest of the circuit.
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 9318
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: driving a high current mosfet at 10's of khz
« Reply #10 on: February 16, 2017, 10:24:08 AM »
I do like the isolation it provides though, if the mosfets blow up it doesn't cascade and kill the rest of the circuit.

I prefer adding protection circuitry so the transistors don't blow up in the first place. :)

Tim
Seven Transistor Labs, LLC
Electronic Design, from Concept to Layout.
Need engineering assistance? Drop me a message!
 

Offline james_s

  • Super Contributor
  • ***
  • Posts: 2772
  • Country: us
Re: driving a high current mosfet at 10's of khz
« Reply #11 on: February 16, 2017, 11:58:42 AM »
Well, yeah, that is preferable, but things don't always work out so well despite the best of intentions. Building stuff like Tesla coils is a bit like building race cars, you're often pushing parts a bit harder than they were intended and sometimes stuff breaks. Also power electronics is a hobby rather than a career so while I consider myself reasonably competent, I'm not claiming to be an expert. Designing a good reliable switchmode converter is easier than it used to be but it's not trivial.

Also I'll mention that I've encountered plenty of cascading failures in supposedly professionally designed equipment that was used for the purpose it was designed for and still blew up.
« Last Edit: February 16, 2017, 12:00:50 PM by james_s »
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 9318
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: driving a high current mosfet at 10's of khz
« Reply #12 on: February 16, 2017, 04:47:00 PM »
Well, yeah, that is preferable, but things don't always work out so well despite the best of intentions. Building stuff like Tesla coils is a bit like building race cars, you're often pushing parts a bit harder than they were intended and sometimes stuff breaks. Also power electronics is a hobby rather than a career so while I consider myself reasonably competent, I'm not claiming to be an expert. Designing a good reliable switchmode converter is easier than it used to be but it's not trivial.

Also I'll mention that I've encountered plenty of cascading failures in supposedly professionally designed equipment that was used for the purpose it was designed for and still blew up.

Why inspect dams when they're always going to erode and collapse?  (To take a reductio absurdum from the headlines.)

Tim
Seven Transistor Labs, LLC
Electronic Design, from Concept to Layout.
Need engineering assistance? Drop me a message!
 

Offline lorenzop

  • Newbie
  • Posts: 4
  • Country: us
Re: driving a high current mosfet at 10's of khz
« Reply #13 on: February 21, 2017, 02:40:47 AM »
I appreciate all the valuable information.

blueskull: Thanks for the suggestion, I think I will do that. I'll make a seperate driver board that snaps ontop the switching board with a few headers like a shield. This will make it easier to work on the driver circuit and move it away from the heatsink.

T3sl4co1l: In this picture I made new connections from the transformer to the rectifier diodes. This wire is much better than the old insulated wire I was using. I'm upgrading from the old 10 gauge copper to 6 gauge soldered together along almost the entire length. The other connections should all be shorter than these pictured.

http://blog.poixson.com/images/IMG_20170203_163330.png

I was planning to use 2 sided 4oz copper boards for the switching board, with the high current traces mirrored to both sides. Would it be worth losing half the trace thickness to put a ground plane on one side? I do expect to pop many innocent mosfets until I get it right, then I hope it'll be a solid piece of equipment.

http://poixson.com/_files/4cc6240e-1abf-4070-b9f2-ebb7b1784f4c.jpg

This picture is one of the inductor cores I got for this. I wrapped 15 turns of 6 or 8 gauge and got (I think it was) 150uH, a bit better than my requirement. In the picture, I was just playing with it, proving the inductor works, capturing the flyback current into the capacitor. Another side note, I now have the eevblog branded meter, and love using it. I may need to find better wire for the inductor. Magnet wire this thick isn't as easy to find.

I was planning to do synchronous rectification (and have my worries about it), but it's easy enough to put that on the back burner and disable it in software. I'll design the boards with the traces to possibly add this at a later point.

The driver chips I have are from another project. Until recently, I wasn't sure if I could use these for this project. I'll do some looking around online and find a much better driver for this project. I'm also looking for better power mosfets with a lower gate capacitance.

MagicSmoker: I do have super capacitors I plan to use. My thinking was the electrolytic capacitors would handle smoothing the output, then further charge the super capacitors to help prevent over/under-shoot. the super capacitors should be well suited to the slower surges.

10khz isn't usable since the power supply will sometimes be used for audio equipment, though I'm wondering if I'll have any electrical noise problems. I do plan to have plenty of big and small capacitors in strategic spaces, and possibly a few zeners. I wasn't planning to use optoisolators, but maybe I should, to protect the mcu.

I'll likely dedicate one driver to each mosfet, with a relatively low value resistor on the gate. This will take some trial and error. I plan to have 2 or 3 large inductors, out of phase, but making this signal is a challenge of its own. I might use an arduino due and manually manipulate the registers for the pwm features of the chip. Easier said than done. Are there any special chips I could possibly use for this? Something like a feature-full pwm generator with I2C or something?
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 9318
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: driving a high current mosfet at 10's of khz
« Reply #14 on: February 21, 2017, 08:38:39 AM »
I appreciate all the valuable information.

blueskull: Thanks for the suggestion, I think I will do that. I'll make a seperate driver board that snaps ontop the switching board with a few headers like a shield. This will make it easier to work on the driver circuit and move it away from the heatsink.

Mind that connecting gate drive through two adjacent pins in a header, adds at least 5nH of stray inductance there.

Be very careful about what you think ground is.  Ground is a matter of definition, and you will find many locations to define it on this circuit.  The connected copper, that's ground at DC, will NOT be ground at AC.  You will most likely find several volts peak, measured between different points of that ground.

These are more reasons to keep switching speed low.

Quote
I was planning to use 2 sided 4oz copper boards for the switching board, with the high current traces mirrored to both sides. Would it be worth losing half the trace thickness to put a ground plane on one side? I do expect to pop many innocent mosfets until I get it right, then I hope it'll be a solid piece of equipment.

That's good enough material, if you use four or six transistors in total (for synchronous).

Quote
This picture is one of the inductor cores I got for this. I wrapped 15 turns of 6 or 8 gauge and got (I think it was) 150uH, a bit better than my requirement. In the picture, I was just playing with it, proving the inductor works, capturing the flyback current into the capacitor. Another side note, I now have the eevblog branded meter, and love using it. I may need to find better wire for the inductor. Magnet wire this thick isn't as easy to find.

Protip: Double up a piece of wire: now its AWG equivalent is 3 lower.  Keep going, and, say, 8 strands of #20 wire is effectively a single piece of #11, but has the flexibility of the #20.

Stranded wire is okay for inductors too, though ideally the strands should be insulated from each other for lower losses (litz cable).  This can be alleviated by using a relatively large inductor, so the current ripple is small (under 20%, say).

150uH for 15 turns is A_L = 0.67 uH/t^2, which is suspiciously high.  It's not ferrite, but it's not powdered iron.  It may be MPP.  In any case, permeability is certainly too high for storing the amount of energy you need -- it'll saturate too easily.

Speaking of, you need around 20uH and 100A rating (at 10kHz).  If you did 12 turns on a T300-52D, you'd get 23uH (at 0A, dropping to about 11uH at 100A), and core losses just about manageable (~10W).  A slightly smaller core could probably be chosen, out of a more suitable material (mix #30 would be best, or Kool-Mu, or a gapped ferrite core).

Quote
I was planning to do synchronous rectification (and have my worries about it), but it's easy enough to put that on the back burner and disable it in software. I'll design the boards with the traces to possibly add this at a later point.

What are you going to use for diodes?  30V schottky would be ideal, but I wouldn't be comfortable even with 40V at this supply voltage.  Each step up in voltage rating, Vf goes up too, and you'll very quickly get too much to handle.

They also don't play well together, because of tempco.

Taken together, all these above reasons suggest you should design it as several independent channels wired in parallel.  This has the bonus that you can drive the channels out of phase, so their ripple currents overlap, reducing total ripple.

Note that multiple channels cannot simply be driven at the same PWM, because the inductor current is a free variable.  You need one current feedback loop per channel, and you servo all channels to set the desired output voltage.

This servo control takes place at the same speed as the output filter capacitor, so you should use only the required amount of filter capacitance, no more.  There is no need or want of supercaps in this power supply.

Quote
10khz isn't usable since the power supply will sometimes be used for audio equipment, though I'm wondering if I'll have any electrical noise problems.

Meanwell makes off-the-shelf power supplies that can be used for these, and they are much cheaper than the labor you will spend here.

And they've already passed FCC and CE!

And if you're worried about learning something in this project, well, guess what you get?  A complete power supply, rated for all the current you wanted, that works!  You can open it up and inspect it, taking note of everything: how and where the traces are routed, what components they chose, what magnetics they used (well, you'd have to unwind them to know it down to the turn, so maybe not that in depth), and perhaps come to some conclusions about why they did what they did!

Learning from others' successes and failures is much cheaper, and much less frustrating, than failing on your own.  :-+

Quote
I'll likely dedicate one driver to each mosfet, with a relatively low value resistor on the gate.

Ignoring advice already?   :scared:

Quote
This will take some trial and error. I plan to have 2 or 3 large inductors, out of phase, but making this signal is a challenge of its own. I might use an arduino due and manually manipulate the registers for the pwm features of the chip. Easier said than done. Are there any special chips I could possibly use for this? Something like a feature-full pwm generator with I2C or something?

First of all,
0. Arduino, just no.
1. Doing it with an MCU (certainly not with Arduino) is a significant challenge.  It's best left to the professionals.  Very few of whom choose to go this route, because it's hard, and easy integrated solutions exist, that are better in all ways.
2. Doing it at low frequency, and using stupendously excessive inductance (maybe more like 60uH), gives you more time to respond in the update loop.  At 10kHz, you even have a chance to acquire one sample -- which has to be inductor current -- with the ADC.  Assuming you can time everything consistently, which is doubtful within Arduino libraries.
3. Driving synchronous transistors (if/when you choose to go this route) from two separate PWM generators --, uh, so what happens if you start one not-quite-synchronously with the other?  How do you ensure they never overlap?  (Which I suspect is actually wholly impossible within Arduino, is not possible to guarantee in C, and may be just barely possible with a really stupid hack in assembly, but only as long as you don't need to do anything else while you're setting it up.)
4. And why would you want to use digital control in the first place?  It's terrible for so very many reasons; are you blind to these reasons?  Do you not see any alternative?
5. There are many, excellent, stupendous alternatives, and they're easier!  First and foremost, build a circuit like this:



This generates a PWM signal. From nothing. No MCU, no programming, no toolchain, it just IS.

Vary R5 or C1 to control frequency.  "CV In" sets the PWM fraction.  You can tie it to a potentiometer if you like, but that's dumb -- let an op-amp do it for you!

The PWM output drives a gate driver, which has to be a bootstrap type, because you have a high-side switch.

CV In is driven by an op-amp, which itself has another control voltage at its input.  It controls PWM% to regulate inductor current.

This is important.  You NEVER just control PWM.  You must control current first, because current in the inductor is a free variable.  If you drive PWM blindly, then inductor current can surge to any value it likes; as soon as that ticks past what the transistors can handle, PSST--BANG!  Death every time.  There is only one way to avoid this: sense current.

Since you have a transformer (or battery?) isolated source, the current sense can be a ground side shunt resistor.  That gives a feedback voltage, perfect for the op-amp to work with.

Outside of that, you set the CURRENT% control voltage with one final op-amp.  This senses output voltage, and adjusts it to match a reference voltage (which can finally come from a potentiometer, or if you must, an MCU DAC output).

All of this circuitry fits easily on a solderless breadboard (except for the high power stuff), doesn't need high speed layout techniques, and controls the power circuitry with perfect current control and precise voltage regulation.  There is no need to have a nasty MCU's grubby fingers all inside there, but if you want additional control methods, you can add a digital disable to the PWM output (just AND gate it), you can sense the output frequency, you can sense the control voltages (voltage setpoint, current setpoint and PWM setpoint), you can sense the feedback voltages (output voltage, inductor current), all of that.

Best of all, since the MCU doesn't have any critical timing responsibilities, you can program it as dirty as you like.  Arduino?  Sure, go right ahead!  You'll never blow a transistor again.

Tim
« Last Edit: February 21, 2017, 08:42:26 AM by T3sl4co1l »
Seven Transistor Labs, LLC
Electronic Design, from Concept to Layout.
Need engineering assistance? Drop me a message!
 

Online blueskull

  • Supporter
  • ****
  • Posts: 7764
  • Country: cn
  • Power Electronics PhD Candidate
Re: driving a high current mosfet at 10's of khz
« Reply #15 on: February 21, 2017, 11:10:02 AM »
This will take some trial and error. I plan to have 2 or 3 large inductors, out of phase, but making this signal is a challenge of its own. I might use an arduino due and manually manipulate the registers for the pwm features of the chip. Easier said than done. Are there any special chips I could possibly use for this? Something like a feature-full pwm generator with I2C or something?

As Tim has said, don't use Arduino, even without AnalogWrite library. Everything in Arduino framework has terrible efficiency. ADC, PWM, you name it.
In fact, for digital PWM with feedback control, I would use TI's F28377S, it has very fast ADC with anti-noise differential input, should you use isolated SDM ADC, it also has interface for these.
It has the best PWM generator on market, 150ps resolution, synchronize with dead-time, kill signal, multi-phase, phase shift, multiple PWM collaboration, you name it.
What's more, it comes in a $29.99 package that has debugger and digital isolator built in. TI's toolchain is also free of charge (CCS6.x).
I know a few people building multi-ten-kW converters controlled by one F28377S, and I also use it in my 1 MW (1*10^6W!) energy recirculation power module testing converter, which is my PhD dissertation.
« Last Edit: February 22, 2017, 07:03:54 AM by blueskull »
SIGSEGV is inevitable if you try to talk more than you know. If I say gibberish, keep in mind that my license plate is SIGSEGV.
 
The following users thanked this post: jbb

Offline MagicSmoker

  • Frequent Contributor
  • **
  • Posts: 752
  • Country: us
Re: driving a high current mosfet at 10's of khz
« Reply #16 on: February 22, 2017, 03:41:56 AM »
...
This picture is one of the inductor cores I got for this. I wrapped 15 turns of 6 or 8 gauge and got (I think it was) 150uH, a bit better than my requirement...

I don't recognize the color code for that core, but the permeability seems a bit high for this application.

MagicSmoker: I do have super capacitors I plan to use. My thinking was the electrolytic capacitors would handle smoothing the output, then further charge the super capacitors to help prevent over/under-shoot. the super capacitors should be well suited to the slower surges.

Your thinking isn't incorrect, it's just that supercapacitors have no business being so close to the output of a switchmode converter; they have very little tolerance of AC ripple because of their relatively high ESR.

10khz isn't usable since the power supply will sometimes be used for audio equipment, though I'm wondering if I'll have any electrical noise problems.

A 10kHz switching frequency might not be ideal, but it is certainly still usable because any audio amplifier design worth a crap should still have decent power supply rejection ratio at 10kHz. But I agree that it is altogether best to not inject unnecessary sources of noise, so the easy solution - which you will likely have to employ anyway - is to use a multiphase, interleaved buck. For example, if you have 2 channels running 180 degrees out of phase then the ripple in the output will both be greatly reduced (potentially nulled out complete at 50% duty) and doubled in frequency. There are also controller ICs that handle more than 2 channels (Linear Technology's "Polyphase" series come to mind), though watch out for the ones intended for "VRM" (CPU core voltage) applications, as they tend to be difficult to adapt to any other use (for example, they often rely on sensing the voltage drop across the inductor to measure current, and so have a very hard limit on the allowed input voltage, usually 12V).

So assuming 10kHz and 2 channels each delivering 50A with a ripple current ratio of 40% (ie - 20A peak to peak ripple), you'll need ~22uH of inductance per channel. An appropriate core would be Magnetics, Inc. Kool-Mu 77737, with 13 turns on it, but there are lots of possibilities here. Even cheap iron powder will do fine at this low a frequency.

I do plan to have plenty of big and small capacitors in strategic spaces, and possibly a few zeners. I wasn't planning to use optoisolators, but maybe I should, to protect the mcu.

Don't forget the retro-encabulator while you're at it...  :o

I'll likely dedicate one driver to each mosfet, with a relatively low value resistor on the gate. This will take some trial and error. I plan to have 2 or 3 large inductors, out of phase, but making this signal is a challenge of its own. I might use an arduino due and manually manipulate the registers for the pwm features of the chip. Easier said than done. Are there any special chips I could possibly use for this? Something like a feature-full pwm generator with I2C or something?

Skip the Arduino... seriously. You simply can't guarantee PWM latency or interrupt response time with these platforms. Totally inappropriate for switchmode power conversion. Also, you are trying to learn 2 fairly difficult things at the same time (your first SMPS and first coding project to control it... a recipe for lots of blown MOSFETs if there ever was one).

I would look at National's - now TI - LM5032. It's a good little interleaved controller IC that you can set the max duty cycle, which is a very important feature if using a transformer-isolated or bootstrap type gate driver circuit.

 

Offline lorenzop

  • Newbie
  • Posts: 4
  • Country: us
Re: driving a high current mosfet at 10's of khz
« Reply #17 on: February 22, 2017, 04:49:24 AM »
Quote
Mind that connecting gate drive through two adjacent pins in a header, adds at least 5nH of stray inductance there.

Good to know, I didn't think of this. I'll just use a local ground pin between each. In this case, would it be better to connect all the ground pins through on both boards, or might it be better to leave one end disconnected?

Quote
Double up a piece of wire: now its AWG equivalent is 3 lower.  Keep going, and, say, 8 strands of #20 wire is effectively a single piece of #11, but has the flexibility of the #20.

Interesting, thanks for the tip. By that rule, my 4 strands of 6 AWG are equivalent to roughly 0 AWG. My guess was 0/2, being generous. I'm still happy with it, plenty of copper.

Quote
Stranded wire is okay for inductors too, though ideally the strands should be insulated from each other for lower losses

I like that idea. It might be a pain to wind, but I could use a few 10 AWG magnet wire.

Quote
150uH for 15 turns is A_L = 0.67 uH/t^2, which is suspiciously high.  It's not ferrite, but it's not powdered iron.  It may be MPP.  In any case, permeability is certainly too high for storing the amount of energy you need -- it'll saturate too easily.

I'm using a cheap LCR meter, so the value of the inductor is basically an educated guess. It's probably closer to 100uH at the most, which is around what I need for 20-30khz by my own math. It's supposed to be a ferrite core, though I don't remember if the manufacturer name was on the paper it came with. The schottkey diodes I have are 45v, which I hoped would be ok for a 20v input. Forward-voltage of 0.55

Quote
Speaking of, you need around 20uH and 100A rating (at 10kHz).  If you did 12 turns on a T300-52D, you'd get 23uH (at 0A, dropping to about 11uH at 100A), and core losses just about manageable (~10W).  A slightly smaller core could probably be chosen, out of a more suitable material (mix #30 would be best, or Kool-Mu, or a gapped ferrite core).

I think the T300-52D toroid you suggested is similar in size to the 3 I have. My green ones have a 39mm center diameter rather than 49mm. I've been doing a lot of reading the last months, but there are still many important things I don't understand. It seems to me having more inductance than needed wouldn't hurt anything much, aside from the physical size and the small additional resistance of the extra wire length. I'm probably missing something here, but I can't think what it is.

Quote
Taken together, all these above reasons suggest you should design it as several independent channels wired in parallel.  This has the bonus that you can drive the channels out of phase, so their ripple currents overlap, reducing total ripple.

Yup, this is what I was planning. I can fit 2 or 3 parallel channels inside the power supply case. With this much current, I imagine having the extra phases greatly lowers output ripple, and lets me deal with lower currents per channel. When I do the math, I divide the current by the number of phases, and usually give it a generous bit more for safety margin, depending on what I'm calculating. My ultimate hope is this thing will withstand a dead short of the output, or possibly experiment with DC welding.

Quote
Note that multiple channels cannot simply be driven at the same PWM, because the inductor current is a free variable.  You need one current feedback loop per channel, and you servo all channels to set the desired output voltage.

This, I did not know. I've done a bunch of reading on MPPT charge controllers for solar panels, where the duty cycle is constantly adjusting to match the peak efficiency. I was under the impression this would have a static duty cycle, which I'd pre-calculate and/or fine tune then not need to change. Would this be dependent on the fluctuation of the input voltage, or does it somehow relate to the output current? BTW, I do intend for this to function in discontinuous mode as well.

I found a schematic for a meanwell RSP-1000. It's a different type of power supply than what I'm building, but I'm sure I'll learn something from it.

Quote
Quote
I'll likely dedicate one driver to each mosfet, with a relatively low value resistor on the gate.
Ignoring advice already?

Not sure what you mean. Isn't that what you suggested, one driver per mosfet and a gate resistor?

Quote
Arduino

Things aren't as bad as you're thinking. I'm more programmer than I am electronics engineer. I have a good understanding of what makes the arduino slow, but the trade-off is, it's easy and convenient to work with. I've already done some research into the subject. The arduino due has a Sam3X Arm processor with some handy pwm features. I'm not yet sure if it can generate 3 phase signals, but I'm pretty sure with some confusing manipulation of the registers, it can generate 2 phase signals for both the source and free-wheeling mosfets and a dead-time between them. I'm also not sure yet if this can handle discontinuous mode, but I'll save that for another time, if I ever try adding the synchronous circuit parts.

I'm not fixated on using the arduino. This wasn't my first choice, but after I failed to find a purpose-made chip which would do what I want, I had found the arduino due could be a possible option. Do you know of any buck controller chips with the features I want and can handle the current? I assume I'd use external driver chips with the controller, but I haven't found how to match these up, or which controller I could use.

As I said, a noisy psu wouldn't be acceptable, so I'll go to the added effort to figure out running at least 20khz. Once I have enough together to be able to start load testing, I'll surely try lower frequencies as well, but I'm unsure how audible it might be.

Quote
This generates a PWM signal. From nothing. No MCU, no programming, no toolchain, it just IS.

Interesting circuit there, but I've never looked at an op-amp oscillator before. It seems to be a whole other animal from what I'm used to. I may try playing with this circuit on a bread board, but it does have some limits. It's single phase/channel, and doesn't support a synchronous free-wheeling mosfet.

If I have multiple phases/channels, how can I go about measuring inductor current? Might a non-invasive approach be possible? Otherwise, I'm guessing a shunt and an op-amp?


blueskull: Thank you for the suggestion, you have my attention with that kit. I've looked at it, but it's far to much information for me to comment much just yet. I'll read it over, and I might get one to play with. your power converter thing sounds very interesting from what little you said about it. do you have any information online about it?


MagicSmoker:

Quote
Your thinking isn't incorrect, it's just that supercapacitors have no business being so close to the output of a switchmode converter; they have very little tolerance of AC ripple because of their relatively high ESR.

I've done a lot of reading on inductors, but as I've said, I still have much to learn. what I don't understand is why to much permeability would be a bad thing. My thinking is DC will pass relatively freely once the inductor is sufficiently saturated. Is the problem related to affecting the AC of the inductor? If that's the case, what I don't understand is how less permeability could help that situation.

Quote
I would look at National's - now TI - LM5032. It's a good little interleaved controller IC that you can set the max duty cycle, which is a very important feature if using a transformer-isolated or bootstrap type gate driver circuit.

That, or something like it, might be just what I need. As I said, the arduino due wasn't my first choice, but simply a possible solution. I'll look over the datasheet more for that and see what others like it I can find.

I almost forgot about the retro-encabulator! A good source of prefabulated amulite is so hard to come by.

Final thought for the moment: I realized today, I need a switch mode psu book.
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 9318
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: driving a high current mosfet at 10's of khz
« Reply #18 on: February 22, 2017, 07:01:52 AM »
Good to know, I didn't think of this. I'll just use a local ground pin between each. In this case, would it be better to connect all the ground pins through on both boards, or might it be better to leave one end disconnected?

Try to ground in one area, and bring all signals through that area.

Gate drive signals will probably take priority, and the other signals (supplies, current and voltage feedback?) will tag along.

Every other pin ground is a good strategy, or all one side on a two-row header (which works the same way if you plug in a ribbon cable -- every other ground, that is).

Not that I would recommend using any ribbon extension here, but keep that in mind when you need some data on a cable. :)

Quote
I'm using a cheap LCR meter, so the value of the inductor is basically an educated guess. It's probably closer to 100uH at the most, which is around what I need for 20-30khz by my own math. It's supposed to be a ferrite core, though I don't remember if the manufacturer name was on the paper it came with.

Ungapped ferrite is the wrong stuff here.  No energy storage.

Energy is stored in air; a high permeability core shorts out the air gap.  This is wonderful news for transformers, which should have as little air as possible!  It works less well for inductors, that need to store energy.  (The distinction between 'transformer' and 'inductor' is mixed; you can have a transformer that intentionally stores energy -- better called a coupled inductor -- and you can have an inductor, that happens to have multiple windings, where the intent is a high series impedance -- a common mode choke, better called a transformer.  Anyway...)

We'd simply make inductors in air, if it were practical; but, as it turns out, it takes a lot of copper to make good inductors in air!  We can save copper by bringing the field in tighter -- shorting out some of the magnetic field path length.  But not all, because we need that precious space to store energy.  The effect is, we take a large and otherwise lossy coil, and concentrate its field down into a much smaller space, increasing the energy density in that space (the air gap) considerably.  This increases the inductance for the same resistance, thus giving more Q.  But too little space, and total energy storage drops (and also, core losses begin to dominate).

So the exact permeability that's desirable for an inductor, varies with materials, frequency and other constraints.  But it's usually in the 20-60 range.

Solid ferrites are usually in the 800-10000 range, so they're way off.  :-\

In short: lower permeability --> more energy storage and more copper losses.  Higher permeability --> lower energy storage and more core losses.  Inbetween lies an optimum.

Quote
I think the T300-52D toroid you suggested is similar in size to the 3 I have. My green ones have a 39mm center diameter rather than 49mm. I've been doing a lot of reading the last months, but there are still many important things I don't understand. It seems to me having more inductance than needed wouldn't hurt anything much, aside from the physical size and the small additional resistance of the extra wire length. I'm probably missing something here, but I can't think what it is.

You can calculate, by inspection, how much flux you need per cycle: if you are doing 24 to 12V conversion (ignoring losses), then PWM will hover around 50%, and a 10kHz 50% square wave spends half the time at 24V and half the time at 0V, while the other end of the inductor averages 12V even.  The change in flux is the flux of one half cycle, or 12V * 50us = 600uVs.

The inductor needs to handle at least this much flux without saturating (otherwise the inductance drops sharply, and switch current skyrockets... another excellent reason to have current mode control!).

But more than that, we can calculate how much flux the inductor absorbs at DC!

Inductance is simply the conversion factor between current and flux.
L = Phi / I == V*s/A == ohm*s
(Which should make sense when you reflect on the sinusoid reactance formula: X_L = 2*pi*F*L.)

If L = 20uH and I = 100A, then Phi = 2000 Vs.

Which means it takes about three cycles to get there, depending on how hard the controller is working to get there.  (And of course, longer if the inductance is even larger.)

If inductance drops suddenly, you can see that adding more flux (applying more positive voltage for more time) causes current to increase rapidly.  This is a crude definition of saturation.

Quote
This, I did not know. I've done a bunch of reading on MPPT charge controllers for solar panels, where the duty cycle is constantly adjusting to match the peak efficiency. I was under the impression this would have a static duty cycle, which I'd pre-calculate and/or fine tune then not need to change. Would this be dependent on the fluctuation of the input voltage, or does it somehow relate to the output current? BTW, I do intend for this to function in discontinuous mode as well.

Yes, yes and yes.

If you want an output voltage that's simply a ratio of the input, you could set a fixed PWM.  With fixed PWM and no exceptions, a hard start (input voltage step, output current step, short circuit, etc.) will destroy the transistor -- simply put, the circuit is trying to be a constant voltage source (in the same sense as a transformer at AC), and drawing fault current from it, well, draws fault current from it, whether from input or load inrush or fault.

The sense in which a transformer is a constant voltage source, is that the mains is a very low impedance (fractional ohm at line frequency), and so has good regulation (small change in voltage for a large change in load current).  Here, regulation would be limited by Rds(on), diode drop and inductor DCR, just as a transformer's regulation is limited by its parameters (DCR and leakage inductance).

Indeed, in a useful sense, such a circuit is a "DC transformer".

And, as such, there is no line regulation: the output is always proportional to the input.  Input ripple, line variations, whatever, it all goes straight through.

So, what differentiates a practical regulated SMPS is the PWM control, which is constantly adjusting to minimize these errors.

Moreover, an SMPS with the two stage design I described (inner current loop, outer voltage loop) is completely immune to fault conditions, as long as the circuit is designed to operate continuously at the maximum possible current setpoint (i.e., when the output of the voltage error amp is saturated).  Under fault conditions, voltage drops, so the voltage error amp commands full current; and, the inner current loop simply does its job delivering full current!  The inductor is never out of control, and the transistors are always switching inside their design limits.

Quote
I found a schematic for a meanwell RSP-1000. It's a different type of power supply than what I'm building, but I'm sure I'll learn something from it.

Ah, cool!  I imagine it has a lot of additional features, too: PFC, mains isolation, special-purpose controller ICs (they make integrated PFC and bridge controllers, these days!), and maybe an auxiliary supply to run it all.

If the schematic is messy, you'll have to study it to tease those apart... but you should come to recognize these things in time. :)

Quote
Quote
I'll likely dedicate one driver to each mosfet, with a relatively low value resistor on the gate.
Ignoring advice already?

Not sure what you mean. Isn't that what you suggested, one driver per mosfet and a gate resistor?[/quote]

Well, the one-driver-per is good, but low value resistors mean fast transistors and big transients...

A single variable, like a resistor value, isn't always an ancillary aspect of the design; often, it's critical and speaks volumes! :)

Do take the time to build scale models.  Start with easy, say, 10 or 20W.  You can even do that on the breadboard.  You'll see all the common errors: bad wiring, common mode noise, nasty switching, maybe even some toasted transistors.

Then solder it together over a ground plane (learn "Manhattan" or "dead bug" style construction, or pick up board-making; or splurge on some custom protoboards!), and see how things change (hopefully, for the better? :) ).

Then, build another one, that's some times bigger.  Double, quadruple, whatever.  Do 20A instead of 2A.  See how much more exaggerated the transients are, and what circuit dimensions they relate to.  Follow the ground-return paths to identify and isolate ground loop voltages, and common mode noise.  (This relates to how you'll connect the controller board -- I think you'll find grouping the grounds around the transistor gate drive connections is best -- but do set up some tests and prove, positive or negative, why that is, and by how much!)

Then make the final version that's, about double that size again, and make three of 'em.

Hook up your controller(s) and you've got three phases of output, from three phases of development (ha?). :)

And, ideally, no toasted (high current) transistors!

Quote
Things aren't as bad as you're thinking. I'm more programmer than I am electronics engineer. I have a good understanding of what makes the arduino slow, but the trade-off is, it's easy and convenient to work with. I've already done some research into the subject. The arduino due has a Sam3X Arm processor with some handy pwm features. I'm not yet sure if it can generate 3 phase signals, but I'm pretty sure with some confusing manipulation of the registers, it can generate 2 phase signals for both the source and free-wheeling mosfets and a dead-time between them. I'm also not sure yet if this can handle discontinuous mode, but I'll save that for another time, if I ever try adding the synchronous circuit parts.

Oh, good :-+ someone who knows what registers are, and why Arduino sucks. :)

Yes, Due is better than the original (AVR) at least.  Dunno if it's still quite suitable, but it's faster, and has more capable hardware, that's for sure.  (I'd put that MCU in the domain of "an expert can pull it off", following my earlier categorization. :) )

Quote
Interesting circuit there, but I've never looked at an op-amp oscillator before. It seems to be a whole other animal from what I'm used to. I may try playing with this circuit on a bread board, but it does have some limits. It's single phase/channel, and doesn't support a synchronous free-wheeling mosfet.

If I have multiple phases/channels, how can I go about measuring inductor current? Might a non-invasive approach be possible? Otherwise, I'm guessing a shunt and an op-amp?

1. Synchronous: in CCM, just add an inverter (or use an inverting gate driver).  You may want to tweak the waveforms with some RC filtering, to get the delays just right (ideally, you don't want more than 10s of nanoseconds of shoot-through or dead-time, but dead-time is very much preferable to shoot-through in the conventional implementation of this circuit!).

2. Synchronous DCM.  This would be trickier.  From a discrete design, I'd probably use the PWM signal to clock a type-D flip/flop, which gets reset when a comparator (on the inductor current) reads current < 0 (or a small amount above zero, to account for propagation delay).

There are synchronous rectifier controller chips available, too, though it's better to do it from within the controller -- you already have the timing signals to at least get started.

3. Multiphase.  That triangle wave oscillator circuit wouldn't do, but a polyphase oscillator will.  This could be made by starting with a higher clock frequency, using a divide-by-N counter and decoder to generate pulses, and using the pulses to trigger ramp generator circuits.  Those ramps then get compared to the PWM CV's (there are N control voltages, from N current error amps, for N independent inductors), and everything else is the same.

4. Current sense.  For a single channel, a shunt resistor works.  For more, you need to use a high side current sense, or a Hall-effect sensor.  Since the output voltage isn't crazy, I'd go for the first option.  You use a current sense amplifier to convert the small shunt voltage (referenced to the output) to a larger voltage (referenced to ground).  AD8210 is one of the best, but INA1xx and such are good, too.

Cheers,

Tim
Seven Transistor Labs, LLC
Electronic Design, from Concept to Layout.
Need engineering assistance? Drop me a message!
 

Offline julian1

  • Frequent Contributor
  • **
  • Posts: 273
  • Country: au
  • http://blog.julian1.io
Re: driving a high current mosfet at 10's of khz
« Reply #19 on: February 22, 2017, 08:10:09 AM »

...I really don't like bootstrap/charge-pump type high side drivers, 

What would be a preferred alternative? A high-side p-channel mosfet in order to avoid the source-follower configuration and the need for a bootstrap circuit? Or a single mosfet + schottky diode?

Offline julian1

  • Frequent Contributor
  • **
  • Posts: 273
  • Country: au
  • http://blog.julian1.io
Re: driving a high current mosfet at 10's of khz
« Reply #20 on: February 22, 2017, 08:34:00 AM »
Quote
I might use an arduino due and manually manipulate the registers for the pwm features of the chip. Easier said than done. Are there any special chips I could possibly use for this?

For an analog PWM controller for my own more modest project I was looking at LTC6992 as a drop-in prototyping component. Both switching frequency and PWM width are voltage controlled.

I was looking at a synchronous design, using the output of the LTC6992 with some AND-like logic perhaps with a comparator to ensure one mosfet's gate is fully discharged, before driving the other mosfet - to avoid shoot-through. The comments here about not driving the gates at 100% to avoid inductive spiking might make that kind of circuit timing more important.

Offline MagicSmoker

  • Frequent Contributor
  • **
  • Posts: 752
  • Country: us
Re: driving a high current mosfet at 10's of khz
« Reply #21 on: February 22, 2017, 08:44:29 AM »

...I really don't like bootstrap/charge-pump type high side drivers, 

What would be a preferred alternative? A high-side p-channel mosfet in order to avoid the source-follower configuration and the need for a bootstrap circuit? Or a single mosfet + schottky diode?

I prefer fully isolated gate drive that is powered by a floating supply. You get the full 0-100% duty cycle range with clearly defined and consistent high/low levels, extreme dV/dt and common mode voltage tolerance, and a physical break between the high power and control sides in case something goes pear-shaped. It costs a few USD - and some cm² of board area - but tends to be worth it above, say, the 1kW level for width-modulated converters (in contrast to, say, the phase-shifted full-bridge, which is most amenable to having each bridge leg driven by a simple two-secondary gate drive transformer).

 
The following users thanked this post: julian1

Online rx8pilot

  • Super Contributor
  • ***
  • Posts: 2721
  • Country: us
  • If you want more money, be more valuable.
Re: driving a high current mosfet at 10's of khz
« Reply #22 on: February 22, 2017, 09:04:18 AM »
I prefer fully isolated gate drive that is powered by a floating supply. You get the full 0-100% duty cycle range with clearly defined and consistent high/low levels, extreme dV/dt and common mode voltage tolerance, and a physical break between the high power and control sides in case something goes pear-shaped. It costs a few USD - and some cm² of board area - but tends to be worth it above, say, the 1kW level for width-modulated converters (in contrast to, say, the phase-shifted full-bridge, which is most amenable to having each bridge leg driven by a simple two-secondary gate drive transformer).

The charge pump drivers are generally slow. At 10Khz maybe not an issue, but that is rather slow. At those lower switching speeds, the Rds(on) is more of a factor, but low FET on resistance usually goes along with higher gate capacitance, needing a high current driver just as much as with faster switching speeds.

As for 'hammering the gate' - it is much more delicate than that. The goal is to switch as fast as possible without causing other problems. Fast switching makes every parasitic something to consider. Traces, wires, headers, etc all quickly contribute inductance that leads to out of control gate signals that ring and bounce, creating an ugly mess. PCB layout can invalidate a perfectly good schematic and simulation.

SMPS design is a humbling experience.
Thank you EEVBlog for all the support in the Keysight Test to Impress Giveaway voting!
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 9318
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: driving a high current mosfet at 10's of khz
« Reply #23 on: February 22, 2017, 09:08:45 AM »
If you opt for PMOS, this sort of circuit is an option:



The level shift isn't the most efficient, but it doesn't need very many parts, and it will tolerate 40V+ on the power rail.

Tim
Seven Transistor Labs, LLC
Electronic Design, from Concept to Layout.
Need engineering assistance? Drop me a message!
 

Online nctnico

  • Super Contributor
  • ***
  • Posts: 12943
  • Country: nl
    • NCT Developments
Re: driving a high current mosfet at 10's of khz
« Reply #24 on: February 22, 2017, 10:46:28 AM »
I prefer fully isolated gate drive that is powered by a floating supply. You get the full 0-100% duty cycle range with clearly defined and consistent high/low levels, extreme dV/dt and common mode voltage tolerance, and a physical break between the high power and control sides in case something goes pear-shaped. It costs a few USD - and some cm² of board area - but tends to be worth it above, say, the 1kW level for width-modulated converters (in contrast to, say, the phase-shifted full-bridge, which is most amenable to having each bridge leg driven by a simple two-secondary gate drive transformer).
The charge pump drivers are generally slow. At 10Khz maybe not an issue, but that is rather slow. At those lower switching speeds, the Rds(on) is more of a factor, but low FET on resistance usually goes along with higher gate capacitance, needing a high current driver just as much as with faster switching speeds.
In my experience finding the right MOSFET is very time consuming. On one hand you want a device with a low RDSon and on the other you want one with low gate capacitance while meeting SOA, current and voltage requirement. It usually takes going through tens to hundreds of parts from various manufacturers to find the one which is best suited.
There are small lies, big lies and then there is what is on the screen of your oscilloscope.
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf