Gate Driver - Miller Effect

[rentner]

Hey guys. Coming with a new problem to you: I want to drive a P Channel Mosfet on the High side, which works fine with level Shifting. The Problem is, that I get the unwanted Miller effect. To make it less powerfull, I added an extra Rail of 15V, so it is VCC+15V. That will drive the gate Negative. That means, it stops the Miller effec just a bit more. Gate swings around -15 and +12V. Well, if the low side Mosfet turns on, the Voltage on the Mosfets Drain drops quickly to VSS and the Miller Capacity loads the gate to at least 2-4V sometimes. The Gate Driver with Emitter Follower quickly turns the mosfet off again, but not quick Enought. As soon, as the Voltage rises, the Current rises to up to 100A or more. Well, Isn't it possible, to use an output stage like an Emitter Follower, build with ultra low RdsOn, that keeps the voltage off or on, only when it is wanted?


Let's pretend, I am using a P Channel Mosfet, that has a low Gate Inductance and low Gate Resistance.

How to drive the Gate with that? The mosfets need to work like a real emitter follower, just much faster. Also It is not wanted, that the mosfets short out each other, if they switch off to slow. Could someone help me with that?

So, I have VCC with 150V, VCC+15V which means 165V and only, if it realy is required: VCC-12V.





Most important: The High side Driver MUST be built of discrete Components ONLY! The goal is a Class D Amp with only the following parts:

BJT, Diodes, Resistors, Inductors, Capacitors, Mosfets. NO IC's! That is the goal.



It also must work at very low and very high pulse widths! The Dead Time is already preset, so there is no need to add any dead time. Switching frequency is around 70khz. The High Side Mosfet is a P-Channel Mosfet, because I did not get it working at all with an N-Channel mosfet. The Low Side Mosfet is N-Channel, which realy does not matter at all, since this is stable.



Now please help me. I want to get a Class D Amp, only built of discrete Components, not to inefficient. At least 85% is a must. Thank you very much!

Edit: Just found out, that my Low Side Mosfet also is a bit sensitive to the Miller effect...

[Simon]

You are not making much sense, what exactly are you trying to achieve ?

[rentner]

Just a Class D Amp with positive and negative rail. If Mosfet 1 Turns off, there is a small delay, before Mosfet 2 Turns on. If Mosfet 2 Turns on, that means automatically, that each mosfet gets a rapid change in Voltage - from VSS to VCC and the other way. That rapid change of voltage at the Drain turns back on the mosfet for a short ammount of time. The The Miller Effect is so strong, that no matter how strong the driver is, it seems to always turn on, when the Drain Voltage changes rapidly.



Maybe the High Side Driver is just shit. Who knows... It does not work properly.

[T3sl4co1l]

Schematic?

[rentner]

The output stage is in the attachment. A lot of testing and combining was done. I still cannot supress the Miller effect...

[fcb]

Rebuild your drivers and pre-drivers with some degree of bias - they will switch faster.

Make the drivers out of little FET's also.

[zapta]

Do you drive the P chan via a 1K resistor (font is too small in the picutre)?  You want to drive higher current for fast switching, otherwise the time constant with the gate capacitance will be too slow. Same goes for the gate resistor of the N chan.

You have the two stages of transistors to drive high current, right? If so, why do you limit it with a high resistor?

(I am not a power mosfet expert, just suggested the other day a Fundamental Friday on this topic).

Edit: what P Chan transistor do you use?

[fcb]

I didn't read the 1K value on the schematic (final gate drive) - you need to be thinking like 10R and designing the driver to pull/push an amp or two out of the gate.

[A Hellene]

You are constantly talking about the Miller Effect and Gate Drivers. Do you really know what the Miller Effect is, what is it caused from and how can you deal with it?

The Miller Effect was initially found in the triode vacuum tubes (the 'valves') and affects the MOS transistors in the same exactly way (since the MOSses are voltage-driven devices also), while in the BJT transistors it is called 'storage charge' since the BJT transistors are voltage-driven devices also, despite of any formal teachings of that Educational Institutions that 'leave no one behind'...

In the vacuum tubes the workaround was the introduction of a second (or more) grid(s) between the driving grid and the anode, and in the transistors is a better (harder) drive, which can easily reach the current of a few (or more!) Amps peak; we are talking of real-world Amperes of gate drive! Now, each additional driver stage will introduce a certain propagation delay of its own, which *will* interfere with the power transistor drive pattern, and it will probably introduce cross-conduction problems, which can be really catastrophic.

Now, regarding your schematic, it is not really a good idea to bias an N-Channel MOSFET gate with negative voltages of unknown amplitude.

I a few words, if you want a really solid solution you should use dedicated MOSFET drivers that the manufacturer(s) have already characterised in any aspect, in order for you to adapt the power transistors drive patterns.


-George

[fcb]

Most important: The High side Driver MUST be built of discrete Components ONLY! The goal is a Class D Amp with only the following parts:

BJT, Diodes, Resistors, Inductors, Capacitors, Mosfets. NO IC's! That is the goal.

[A Hellene]

Yes, I've read that.

My criticism is towards the instructor's demands, since it seems to be a school assignment:

BJT, Diodes, Resistors, Inductors, Capacitors, Mosfets. NO IC's! That is the goal.

If it is not, there is no way to beat ready-made specialised parts with discreet ones, for that purpose.


-George

[rentner]

So, is my guess correct, that the change in Voltage at Drain is the reason?


And why not set the gate voltage to -12V? Isn't the Gate able to withstand -20V theoretically? At least that would give me some room for the Miller effect - It needs quite "some power", to load the gate. Or Does this help absolutely nothing?


What to do now? Do I need to increase the switching current or decrease (which propably gives me no room for any efficiency )



What is it, what a factory made gate driver does different? Also, How do I actually use Mosfets for the Driver? I have never seen this in practice and can't even imagine that, since Mosfets are voltage controlled - I cannot just build something like an Emitter follower.

[fcb]

Yes, I've read that.

My criticism is towards the instructor's demands, since it seems to be a school assignment:

BJT, Diodes, Resistors, Inductors, Capacitors, Mosfets. NO IC's! That is the goal.

If it is not, there is no way to beat ready-made specialised parts with discreet ones, for that purpose.


-George

Possibly, or perhaps they want to do it for the hell of it. Like those people that build 4 bit microprocessors out of NAND gates.

[fcb]

What is it, what a factory made gate driver does different? Also, How do I actually use Mosfets for the Driver? I have never seen this in practice and can't even imagine that, since Mosfets are voltage controlled - I cannot just build something like an Emitter follower.

You'll find that the headline spec. on commercial MOSFET drivers tends to be the drive current - to make your design (MOSFET) efficient, you need to switch it ON and OFF fast - so you need to push/pull plenty of current into the Gate as fast as possible (with reason/design/etc... yada yada).

In my experience, you can use pull-up resistors and a small fet (like 2N7002) to pull-down if you want to go at medium speeds with medium MOSFETs (like BUZ11), but class-D tends to be quite fast (100KHz+) and your MOSFET will be beefier (more Gate).

[Zero999]

Why use a bipolar 140V supply?

It would be better to use a single supply and a full bridge. If a filterless modulation scheme is used, the output filter can be eliminated, which should also increase the efficiency, as well as reduce the size of the design. The disadvantage is the modulation is more complex.
http://pdfserv.maximintegrated.com/en/an/AN3977.pdf
http://www.ti.com/lit/an/snaa034a/snaa034a.pdf

[A Hellene]

And why not set the gate voltage to -12V? Isn't the Gate able to withstand -20V theoretically? At least that would give me some room for the Miller effect - It needs quite "some power", to load the gate. Or Does this help absolutely nothing?

You cannot find even one decent design, where the gate is driven below the source voltage to set the transistor OFF. Vgs_th is not equal to Vs; in high voltage power MOSFETs Vgs_th is equal to 2..6V above Vs, so there is nothing to gain by driving Vg below Vs. On the contrary, if you did that you would have to overcome that additional negative charge of the gate, every single time you would need to set the transistor ON.

You could use gate drive transformers for fast switching of high voltage power MOSFETs, in order to make your instructor happy. But you cannot, because the gate drive transformers are no good for duty-cycles above 50%; in Class-D power amplifiers the output stages duty cycle can easily go beyond the 1..99% 'marginal' figures.

What is it, what a factory made gate driver does different?

Try beating the output current/ propagation delay/ delay matching/ output rise & fall times/ etc. specifications of a FAN7190/ISL2111/LM5101A/etc. or any better gate driver, by using discreet components!

Like those people that build 4 bit microprocessors out of NAND gates.

Nowadays, they are not that smart; neither do they need them to be that smart...


-George

[rentner]

I don't know how, but I did it! It works now and that quite efficient. 93% with ideal filter - an efficient filter should give me close to 90% efficiency, which is good enought for me. And driving the Gate below Source is not as efficient, but only wastes about 0,5W at my 70khz. I am getting quite a clean sine wave even at 20khz. Man, I am quite happy with that. I might show some results, If I can take the time to build it in real.


Only negative thing with discrete components is the modulation to "only" 90% of the Supply Voltage. So a 40V Dual Supply gives my about 35V Pk. That is good enought, I would say. I don't want to risk clipping, so I will not make the gain any higher.



I don't know, why some people do not like the idea of discrete components... Well, you can only learn from that. Gather experiences.

Just one more thing: I increased the Gate Drive Current to 2A Peak. It gives me a very clean signal.

[zapta]

OP, you are not allowed to use off the shelf mosfet drivers but you can learn from them. Take a look at this list of drivers

http://www.analog.com/en/power-management/mosfet-drivers/products/index.html

It's all about the current that they can source/sick. The miller effect introduces a large (equivalent) capacitance so if you want to switch fast you need high current.  If you switch slow, your transistors spend more time with partial conductivity (both current and voltage are not close to zero) which results in heat dissipation and reduced efficiency. 

Another issue is current flowing directly between the two transistors, bypassing the load. A break before make timing may address this, don't know, other may comment.

[fcb]

Only negative thing with discrete components is the modulation to "only" 90% of the Supply Voltage. So a 40V Dual Supply gives my about 35V Pk. That is good enought, I would say. I don't want to risk clipping, so I will not make the gain any higher.

Great & well-done! Can you publish your tweaked circuit on this thread - will be interesting to see the changes.

Also, you can make a basic charge pump using your 70KHz clock, this can be used to give your driver circuitry a bit more swing above/below  the main rails, so you might be able to get a bit more output from your final stage.

[T3sl4co1l]

Why are the series gate resistors unequal?  (1 and 10 ohms)

Miller effect (drain-gate feedback) is limited by the gate drive impedance.  If you cannot get the impedance low enough, you have to reduce dV/dt instead, for example with a snubber.

I have a simulation using discrete devices which claims better performance than any gate drive IC I have ever seen datasheets of (let alone for sale anywhere); of course in practice, that'll come down by a few nanoseconds, but nonetheless with careful design a discrete driver offers competitive performance.  (It does NOT offer competitive current consumption or PCB area however!)

Tim

[Zero999]

I don't know how, but I did it! It works now and that quite efficient. 93% with ideal filter - an efficient filter should give me close to 90% efficiency, which is good enought for me.

I doubt that. The losses are likely to be higher in a real filter.

[rentner]

I don't know how, but I did it! It works now and that quite efficient. 93% with ideal filter - an efficient filter should give me close to 90% efficiency, which is good enought for me.

I doubt that. The losses are likely to be higher in a real filter.

Because the Filters choosen in commercialy available products almost all of the times are cheap filters. Caps with high resistance and also the Inductors are loosing a lot with a bad core.

[envisionelec]

Just a Class D Amp with positive and negative rail. If Mosfet 1 Turns off, there is a small delay, before Mosfet 2 Turns on. If Mosfet 2 Turns on, that means automatically, that each mosfet gets a rapid change in Voltage - from VSS to VCC and the other way. That rapid change of voltage at the Drain turns back on the mosfet for a short ammount of time. The The Miller Effect is so strong, that no matter how strong the driver is, it seems to always turn on, when the Drain Voltage changes rapidly.



Maybe the High Side Driver is just shit. Who knows... It does not work properly.

This is the result of high dv/dt. You must suppress the energy that reverse biases the drive circuit. It's trying to return energy back to the rails which turns on the FET. Please research RCD snubbers. See figure 11a, page 2-13 in the following document. Hint: L1 is a single-turn through a ferrite core which comprises L2.

http://www.ti.com/lit/an/slup100/slup100.pdf

EDIT: Oh...you fixed it already. Well...this is good information for others...