MOSFET gate PCB design question

[Futterama]

Hi forum,

I'm designing a 200A brushless controller and have read a fair bit on MOSFET gate driving.

I'm paralleling SMD devices and have read application notes about gate oscillations and how to avoid it. But I'm in doubt regarding the gate ferrite bead and gate resistor. Should those be placed as close as possible to the gate pin on the MOSFET or is it more important to route the gate path away from the high current paths (source-drain)?

I have two options:

1. Put the ferrite bead and if needed, the gate resistor real close to each MOSFET gate, but this will place the ferrite bead and resistor parallel to the high current path, increasing the coupling to the gate.

2. Route the gate trace away from the high current plane (using a 10mm short wire as "via") to another board where the gate driver is located and then locate the ferrite bead and gate resistor here with a ground plane between the high current paths and the gate drive circuitry.

See the board layout below. I started placing the ferrite bead next to the high current path, thinking it should be as close to the gate as possible.

Also, I have been reading a bit on MOSFET matching when paralleling devices, but it seemed to me that matching was only done when the MOSFETs should operate in their linear region. Is it a good idea to match the MOSFETs when paralleling devices in a PWM configuration or is that not necessary?

Thanks.

 

[nieles]

yes, mosfet matching is still important. because you want the fets to turn on all at the same time.
if the mosfets are not mached, and one will turn on earlier/faster than the other parallel mosfets.
that mosfet will take the full phase current until the other mosfets are turned on.

 

[zombiess]

SMD devices = fail, they can't be heat sinked well.. To-247 or to-264. Two to-247 devices in parallel will do 200a with headroom left for safety. Assume worst case values and run math calcs at different temperatures. I like to go with 100c as my max operating temp initially.

As for gate drive layout is absolutely critical. Having power supply traces overlap each other 100% helps to reduce stray inductance. Figuring out a good layout is a mofo. Try to minimize the loop area. I spent around 4 months working with a pro to get my layout to where it is now. I am still building so no real world data yet. It is posted on here in the tech section.

You Absolutely should match paralleled FETs and parallel as few as possible to get to your power level.

Use a good gate driver that incorporates safety to protect the MOSFETs. If you want to go high end use isolated power supplies on each phase segment, full isolation requires 7 of them.

Watch your propagation delay as it will effectively limit your maximum switching frequency.

Good low ESR caps directly between high side drain and low side source.

Work out all the math and document it, you will need it for reference later!

I will post more as I can think of the basics. I am noon but have had some very solid guidance.

Read lots of app notes. I read them for around 1.5 years (and still do) before I received A Lot of help from HighHopes and Bigmoose.

 

[zombiess]

Thought of some more tips.

Be careful with component selection. Anything in the gate drive loop should be as low inductance and fast as possible, caps, diodes, resistors (go smd). If you use zeners to protect anything read their data sheets carefully and look at how the breakdown voltage drifts with temp and how quickly they can react.

I am not sure if their are any books on this black art, but it is as much art as EE work. 10x harder than I thought it would be. It looks so easy at first but it is unbelievably complex.

Always know the state of your gate under all conditions or possible failure modes. It is a good practice to install gate to source resistors on every fet so it will always be in a known state. This is important especially when applying power in the turn on sequence. Something such as a high current spike could induce enough stray inductance in the gate loop partially or fully turning on FETs. I have read about and actually experienced this on a Xie Chang controller once when I powered it on and forgot to connect the hall sensors. The instant I powered on the low voltage / drive section along with connecting the batteries I lost 8 FETs in an entire phase (technically only a few died but I replaced all of them). Somewhere around 5k is a good value. I went with 4.75k on my design because I anticipate using a 5 ohm primary gate resistor with 0.5 ohm gate resistors placed at each fet as close as I could get them to the gate.

In the zip file I posted of my Cadillac driver setup I included an open office document which explains every component and its function. I also included some of the math to justify the parts values. I believe the gate to source resistor.

You will have to make trade offs in your design to reach what ever design limitations are imposed by your self or due to physical constraints.

What I am saying is just general advice I am regurgitating, but it has been drilled into me because it is important. As I said am an amateur at this so do your due diligence. I am sure some of the more knowledgeable design guys will chime in on this thread.

 

[Futterama]

zombiess, thanks for posting.

I have considered using non-SMD MOSFETs as I have seen many designs in here that don't use SMD MOSFETs. But all the RC controllers I have seen has been using SMD MOSFETs, even those in the 200A range. They have something like 8 or 10 MOSFETs in parallel. I have read an application note where heat sinking in SMD devices are discussed, and the point was that the heat is generated in the source connection inside the MOSFET, so the source pins will need the most heatsinking. This is done by soldering the source pad to a good thermal conductor, usually a multilayer PCB, but I was planning on soldering them directly to a custom milled bus bar which is in thermal (and not electrical) contact with the primary heat sink.
The devices I plan to use can handle 108A each (package limited) and I plan to parallel 6 of them. They have a RDSon of 1.9mOhm each so 6 devices will generate 12.66 watts of heat from resistive losses at 200A. This doesn't seem like much to me.

Regarding MOSFET matching, is it the same procedure as for those guys who use them in their linear region? Like this:
http://www.diamondstar.de/transistor_matching_mosfet.jpg

I will be using half-bridge bootstrap drivers which monitors the gate voltage.

I guess it should be enough to add gate-source resistors to the low side MOSFETs, because if they are controlled during power-up, the MOSFETs can't short out even if the high side MOSFETs turn on. Adding gate-source resistors to the high side MOSFETs will also add to the bootstrap capacitor discharge when the high side MOSFETs are on.

 

[Futterama]

zombiess, do you have a thread on your controller you can post a link to?

Regarding my gate design, can you give me your opinion, which is more important? Routing the gate path away from the drain as close to the gate pin as possible to minimize coupling or adding the gate resistor and ferrite bead as close to the MOSFET gate as possible even though this will place them close to the drain current path?

 

[Honk]

The devices I plan to use can handle 108A each (package limited) and I plan to parallel 6 of them.
They have a RDSon of 1.9mOhm each so 6 devices will generate 12.66 watts of heat from resistive losses at 200A.

Hehehe!
You must be talking about FDMS86500DC from Fairchild with Dual Cool case for easy heatsinking.
Yes, they handle 108A max but this is not realistic in real life. The 200amps of your design when divided by 6 fets = 33.3amps each. This is a fair load
if you mount a top heatsink with active air flow cooling. But if your controller uses high speed switching as power limit then the power loss can double that.
Make sure you design the cooling for twice the powerloss you expect.

I'm actually planing two 250amp sine wave three-phase controllers of my own. They operate at 40KHz and use the same basic PCB.
One unit will use 12+12pcs FDMS86500DC for each channel.
The other will use 6+6pcs of FDP020N06B.

Why two, well it's simply good to have another controller if one breaks down. And I can just as well test different
mosfet configurations to see what solution works the best.

 

[zombiess]

Watch your spec sheets and have you reality meter out. TO-263 devices aka D2PAK are a TO-220 with no heat sink tab so they suffer a disadvantage. I am looking at moving to TO-247 or larger devices to simplify design. Some of the packages allow bus bar to be bolted directly to the FET tabs and they have massive heat sink contact area with good Tj/Tc C/W ratings.

TO-220's can run around 35-50A each in switching environments reliably with good thermal management. TO-263's are going to be a lot less unless you are talking short term burst rating since there are very few ways to heat sink them well.

Don't forget to account for body diode losses in your thermal calculations. Diode losses at low PWM can be higher than switching and conductive. You must have a good thermal and amperage safety margin.

Just remember, more parallel devices = more things to go wrong. You also hit diminishing returns.

What is the justification for high freq PWM, what is your intended load?

 

[Futterama]

Honk, yes, it's the FDMS86500DC I'm planning on using. The Dual Cool devices have slightly better specs than the non-Dual Cool FDMS86500L devices.

I don't know what you mean by using high speed switching as power limit, can you explain or point me to some reading?

I'm using a PC CPU cooler with modified base. It should be able to sink 70W. Compaired to the heatsink on a 200A RC ESC, the CPU cooler is many times larger.

I'm not planning to use more than 16 or 32kHz switching frequency.

Have you been thinking about the PCB layout around the FDMS86500DC MOSFETs? I would very much like to hear your ideas.

I have been reading about sine-wave control and space vector modulation, and I must admit, I don't get it. The explanations must have been on a level higher than my understanding, because I was missing the basic info on how to even create a sine-wave. And all that vector talk is just confusing.

zombiess, the Tj/Tc C/W rating for the FDMS86500DC is 1.0 to the bottom drain. This seems ok to me, even though it is not as low as the FDP020N06B TO-220 device. I wonder what the Tc/Tbusbar C/W rating would be between soldering the case to a copper busbar compaired to bolting it to the bus bar.

Regarding body diode losses, I'm planning on using synchronous rectification, but I have been reading an application note where I think they meant that they put a schottky diode parallel to the body diode because the body diode has a slow reverse recovery, and this would improve something, can't remember all of it now. Something about the schottky diode conducting faster than the body diode so less voltage would be build up before the synchronous rectification would take over. Does this make sense?

I'm open for switching frequency suggestions. Would it be possible to match the frequency to the specific motor? I think I read something about saturating the motor windings due to low PWM frequency was bad.

zombiess, I also found your thread on the power stage for Lebowski's controller, but I haven't had the chance to study your PCB layout yet. I don't see how it would be possible to put the power traces above each other with Power 56 SMD style devices as you suggested.

 

[Honk]

What is the justification for high freq PWM, what is your intended load?

I need approx 40KHz to block idle currents through the Y-winding of a low inductance heavy duty BLDC (43V 10KW).
If any lower then the motor whould become very inefficient at low power. I might go for 30KHz but this need further tests.
Btw, my mosfets will be very well cooled by fans and heatsink. Last but not least, proper & fast gate control together with
thick and sturdy ultralow resistance copper bars to/from source-drain ensure equal current through the parallel fets.

 

[Futterama]

We are actually completely offtopic. There has been no attempt at answering my question about what is more important, place the gate resistor/ferrite bead close the MOSFET but next to the high current path, of route the gate path away from the high current path and place the gate resistor/ferrite bead a bit (10mm) away from the MOSFET.

 

[bigmoose]

I would put the individual gate resistors/beads close to the package.

 

[Honk]

I don't know what you mean by using high speed switching as power limit, can you explain or point me to some reading?

Thats simply adjusting the on-time by duty-cycling the fets when the power limit is hit, usually sensed by a low ohm resistor as current shunt.

I'm using a PC CPU cooler with modified base. It should be able to sink 70W. Compaired to the heatsink on a 200A RC ESC, the CPU cooler is many times larger.

I'm using CPU coolers as well, great little suckers with big potential.

I'm not planning to use more than 16 or 32kHz switching frequency.

That might be fine but I don't know anything how you designed your controller, so I can't give any usuable feedback.

Have you been thinking about the PCB layout around the FDMS86500DC MOSFETs? I would very much like to hear your ideas.

The FDMS86500DC design is not ready for release yet, mostly due to lack of time.

I have been reading about sine-wave control and space vector modulation, and I must admit, I don't get it.

There's a lot of IC's out there doing all the fun. Fairchild have a nice one. FCM8201. Check it out.

Regarding body diode losses, I'm planning on using synchronous rectification, but I have been reading an application note where I think they meant that they put a schottky diode parallel to the body diode.

Personally I never use parallel schottkys, instead I closely match the H-bridge for perfection without any cross conduction.
That's so much better and efficient. In industrial massproduction they tend to use schottkys as it is less struggle to make it all work.

I'm open for switching frequency suggestions.

It strongly depends on the inductance but it's enough with 16-25KHz for most motors. My own motor is speciell though, takes higher frequency.

 

[Honk]

We are actually completely offtopic. There has been no attempt at answering my question about what is more important, place the gate resistor/ferrite bead close the MOSFET but next to the high current path, of route the gate path away from the high current path and place the gate resistor/ferrite bead a bit (10mm) away from the MOSFET.

I would use a 0402 footprint close to the gate, then test IRL what seems to work the best on that specific design.
Another important little tip is to mount a 600W transildiode at the gate to prevent a mosfet breakdown conducting backwards and detroying the whole controller. Should be fitted as: Gate - Resistor - Transil - Gatedriver.

 

[Futterama]

I would put the individual gate resistors/beads close to the package.

Thanks!

Now, I have trouble selecting the ferrite bead. I only have access to a 12MHz scope so I cannot measure the fast oscillation on the gate so I can only guess which material in the ferrite beads I should select. What impedance at 100MHz should I aim for? The gate drive current is 3-4A, do I need to select a ferrite bead rated for this current? The current is not constant as it is in a power supply so if the rating is only based on heat dissapation, I'm thinking a 1A ferrite bead should be able to handle 3-4A gate drive current. But then there is the thing about saturating the ferrite core, if the gate drive current has saturated it, the high frequency oscillation can pass through it easier - right?

Looking at Digikey and their selection of stock SMD ferrite beads with a current rating of at least 4A and not bigger than 0805, the impedance at 100MHz is around 100-120Ω. Is this enough?

 

[Honk]

Don't worry about ratings or heating in the gate bead. The rise/fall time is very fast, approx 10-20ns with a properly designed driver circuit.
The affect on the bead is minuscule. More important is the bead characteristics.

If your going to measure any gate oscillations you need a probe with very short current path. You can't use a regular prope as this will induce false oscillations.
You can rebuild a regular probe though, replace the long ground wire by a 5mm short wire from probe tip ground. It should sit parallel to the signal tip.
Now you have a short path probe suitable for the jobb. Go lend a 100MHz scope

 

[bigmoose]

I would use individual 2.7 ohm gate resistors, give or take. Only go to a bead if you have demonstrated high frequency issues.

 

[Futterama]

Honk, if I can use a ferrite bead rated for 200mA, it will have a DC resistance of 1Ω. Would this allow me to use the ferrite bead internal resistance as the gate resistor? It would also have a higher impedance at 100MHz which I guess would remove even more of the oscillations.

Bigmoose, the problem is I cannot demonstrate high frequency issues without a better scope, and I just don't have the cash for a better scope right now, thats why I recently got the 12MHz I have now. Since the price of the bead is something like $1 for 50pcs, I would just add them as a precaution.

 

[Honk]

Yes, you can probably use the ferrite bead as "gate fuse".
Myself I prefer using size 0402 1R as gate resistor and it fuses really well if needed.
Size 0402 works beautifully and has really protected my prototype designs at several occasions.
If oscillation is a problem then experiment with other values, resistive or inductive, use what ever you find best.

 

[HighHopes]

most appnotes on subject of gate drive will focus on minimizing the gate drive loop area, stray inductance (or any inductance for that matter). this is because the switching voltage and rise/fall time of the gate current is high .. so mathetmaically you have high dV/dt and high dI/dt in the gate drive. so any kind of inductance or capacitance will start to cause problems via equations:

I = C * dv/dt
V = L * di/dt

so when i see that you have a separate gate drive board and are using wires to connect to your mofset board i think "this gate drive loop area will be large". then i read you are going to add ferrite beads and i think "oh no... he is purposely adding inductance with no definable purpose why".

so i think maybe you are going down the wrong path and perhaps do not fully understand the nature of the beast. that's OK, you have to start somewhere with education.

for motor drives, in my humble opinion, gate drivers are the most critical part (closed loop current control is the most complex part). if you want to learn how this is done with more clarity you might try focusing your appnote selection from those suppliers that regularly deal with higher voltage higher current applications. when you talk about 200A you are not talking about a toy.. 200A is serious. you should read appnotes from companies like IXYS, Infineon, PowereX, Sensitron.

 

[Njay]

All the times I read (2 or 3) about the ferrite beads on the gates, they mention putting the bead on (around) the gate *leg* of the MOSFET; so I guess they want it pretty close .

It is a good practice to install gate to drain resistors on every fet so it will always be in a known state.

I guess you meant "gate to source"?...

I've been thinking about putting a 2 pin male header in place of the driver-gate resistor, so I can build a kind of a "jumper plug" with an SMD resistor and easily try several values.

 

[HighHopes]

All the times I read (2 or 3) about the ferrite beads on the gates

where did you read this? i would like to read it also so i can understand the context in which it is written

 

[HighHopes]

All the times I read (2 or 3) about the ferrite beads on the gates

where did you read this? i would like to read it also so i can understand the context in which it is written. i couldn't tell you how many hundreds of documents i've read on the subject and don't ever recall adding ferrite beads in a gate driver for 200A application.

 

[Njay]

I don't have any reference on hand, but I also didn't mean it was for a 200A application; just "MOSFET gate ferrite". If I cross again with one of those refs I'll post it here.

 

[Arlo1]

Excited to see this progress we need more building DIY controllers.