Doc EXTREME 36 FETS controller

[ZapPat]

Hi Doc!

[...]
I also know that it's the Hi side that generate more heat so puting directly the mosfet to it without insulator would be just better... the low side can still survive by leaving them with standard insulator method.
Doc

It's actually the low side FETs that will heat up more, specially at low PWM duty cycles (when at lower speeds). While there is one FET output doing PWM, remember that there is always one other phase being held to ground (batt-) too. So you get PWM losses on one low side FET, plus conduction losses on another at any one time.

Luke, Any idea on leaving longer legs on the fets to increase a little bit their resistance and better match them in parallel??
I mean alot of high power amplifier are using low ohm resistor between the rail and collector and emitter to compensate for the gain difference and make a better equilibrium on the current share...
Could leaving the fet leg like 2-3mm longer could help that... or cold it also decrease performance on another way by doing that... any pros and con ??
Doc

Adding power path resistance doesn't help with paralleled FETs as it would with transistors. The nature of the beasts are different, the FETs having current sharing problems mostly during dynamic transition times (when they change state with each PWM period), vs the transistors having sharing problems mostly during static times (conduction). I have read that having higher FET source inductance does help share current during the switching events, but it is also generaly not good to have inductance in the switching path as it raises PWM losses. And I don't think 2-3mm longer leads for these pretty slow switching ebike controllers will change much.

Anyways, to avoid having too much uneven current sharing with FETs you would want to...
- Place all paralleled FETs thermaly close together - a very direct heat path from FET tab to FET tab as you are suggesting is good
- Use FETs of the same production batch
- Make sure the paralleled FET's power paths and gate drive paths are all electricaly symetrical as possible (no control...)
- Use individual gate drive resistors (no control...)
- Avoid pushing your FETs into avalanche (which makes current sharing much harder yet) by not using them too close to their max voltage rating.

Hope this helps!

Pat

 

[liveforphysics]

It's the inductive path that effects how current is shared as well.

If you remember the cell tester I built, a linear array of FETs with the drain tabs soldered directly to heavy copper bus bars, under the IR camera, we could see which FETs were getting used, and which were just along for the ride. It was 8x TO-247 fets in a row, and the FET closest to the outlet was like 40degF hotter than the fet on the other end of the 1/4" thick copper bus, which given the excellent heat transfer of copper, tells you the FETs at the beginning of the linear array aren't doing jack-squat. I'm confident I could have done an array with only 4 fets arranged in a circular inductively balanced configuration and arrived with a more robust end product.

Linear arrays like the brushed controller shown by Marko in the pics above are very simple to build, very pleasing to the eye, and very pathetic in current sharing function.

 

[Doctorbass]

Thanks for these great informations guys. The use of copper would be good but it's also very expensive and L bar in copper are like they dont really exist... and thay would make the controloler 4 pounds heavier so... i'll stay with aluminum L bar and will try to maximise the surface contact with the bar and the case.

anyway.. if i was able to push the 18 fets controller to up to 20kW burst at 262A burst so with a 36 fets i could expect at least 35kW burst...

The new 24 and 36 fets pcb desing show a gate resistor for each mosfet and a very symetrical traces so it's good new.

Also my case desing make the fets closer to the case than the original 36 fets controlelr case wich is too high.. The L Bar i will use is 1-1/4" per side compare to the original bar that is around 2 inch and the total effective dissipation surface of the case near the top is better than the original big case.

Doc

 

[texaspyro]

- Use individual gate drive resistors (no control...)

Actually you are best using a single gate drive resistor. Or if oscillation is a problem, a common resistor feeding individual resistors around 1/10 the resistance of the common resistor.

Matching for Rds on FETs from the same batch does not usually help (but certainly does not hurt). I bought 100 IFRP2907's for my CD welders. They all had 3000 +/- 100 micro ohm resistance at full on. I parallel 18 of them... and can switch 20,000 amps.

Matching for gate threshold voltage is very helpful. At 4V the Rds was anywhere from .004 ohms to over 1 megohm.

See http://www.irf.com/technical-info/appnotes/para.pdf for some very good (but rather heavy duty) info.

When you mount the FETs don't trim the leads. They do help get heat out of the package.

 

[tostino]

Doc, you think that with good water cooling that it would matter how much surface area is touching the case? It seems to me, that good water cooling could do the job with a few passes of pipe soldered to the aluminium bar without any attachment to a metal case.

 

[Doctorbass]

Doc, you think that with good water cooling that it would matter how much surface area is touching the case? It seems to me, that good water cooling could do the job with a few passes of pipe soldered to the aluminium bar without any attachment to a metal case.

Water cooling seem a good idea.. but like i said to Steveo, it is also creating additional malfunction possibility and more complex system. If i would use water cooling i would certainly use copper but again it's adding ALOT of weight to the controller and it is more difficult to get the part that fit in the controller unbless you solder copper tube and things like that like Arlo did...

I have two 36 fets boards so i could try it with the 4115 and watercolling for the next project

Doc's controller inventory:

-Crystalyte 35A 72V modified to 100V 100A.... (BLOWN.... I mean BLASTED)

-Infineon 18 fets controller with copper bar inside for cooling... Still working great!!

-Infineon 18 fets No copper inside (BLOWN... BLASTED... VAPORISED)

-6 fets controller for testing ( Working)

-Kelly controller 120V 220A BLDC ( working)

-36 fets controller ( in progress)

 

[Sam_Q]

Okay let me get this straight... if we keep the alu bars isolated from each other we don't have to use kapton tape between the fets and the alu bar?

Please tell me that's true!!

Possibly... not sure..

Well.. using one bar for the positive rail could work.. since the pin 2 of each fets are conducting current exactly like the back of the mosfet and the entire conductive part.

The problem is conducting heat and current.. and the heat sink compound is not conductive material and would insulate... just the screw that fix hold the mosfet to the bar could really conduct...

but for the goal of not using any insulator between the 2 bars and the mosfets to better conduct heat and just put heatsink compound it would work... but you MUS ensure that none of these 2 bar are conudcting with the case!!

Doc

this is a tottaly ignorant question but how about copper grease?

 

[ZapPat]

- Use individual gate drive resistors

Actually you are best using a single gate drive resistor. Or if oscillation is a problem, a common resistor feeding individual resistors around 1/10 the resistance of the common resistor.

I think there's still some debate about the best things to do to avoid dynamic current sharing unbalance when using paralleled FETs. I am of the opinion that it would not be good to use a single gate drive resistor - not only because of oscillation dangers, but also because one common resistor would degrade current sharing. If all the FET gates are bound to the same voltage, any threshold voltage mismatch between devices will make the lowest Vth device take the current for even longer because the lower Vth FETs will hold the other's gates below their own Vth for longer. I also can't see the point of combining a common and individual resistors - why use the common one at all? Here's two citations from IR AN-941 to illustrate my point:

"Figure 2 shows that when using paralleled devices, a low impedance path is generated that may be prone to parasitic self oscillations. This is analyzed in greater detail in Ref [1]. Individual gate resistors provide the necessary damping and gate decoupling to prevent oscillations."
"The first potential difficulty is that if we apply a common drive signal to all gates in a parallel group, then the first device to turn ON—the one with the lowest threshold voltage—will tend to slow the rise of voltage on the gates of the others, and further delay the turn-ON of these devices."

I already had the other app note from IR you posted (para.pdf), but will scan it again for different/complementary info since it's been a while I was into reading about this design step... maybe I'll find something I missed or forgot about it that document about the advantage of using a common gate resistor?

Matching for gate threshold voltage is very helpful. At 4V the Rds was anywhere from .004 ohms to over 1 megohm.

Matching for Rds on FETs from the same batch does not usually help (but certainly does not hurt). I bought 100 IFRP2907's for my CD welders. They all had 3000 +/- 100 micro ohm resistance at full on. I parallel 18 of them... and can switch 20,000 amps.

Both good points, because matching for Rds does not help at all with the dynamic (state changing) current sharing issue, whereas screening and matching Vth for each FET group would certainly be the most important thing to do.


Pat

 

[ZapPat]

Well.. using one bar for the positive rail could work.. since the pin 2 of each fets are conducting current exactly like the back of the mosfet and the entire conductive part.

The problem is conducting heat and current.. and the heat sink compound is not conductive material and would insulate... just the screw that fix hold the mosfet to the bar could really conduct...

Doc

Doc - don't worry about the thermal paste you use making the drain tab not conduct electricity well to the heatsink... it will conduct very well even if your compound is non-conductive. If it does end up having bad electrical conduction problems, it's because you would be using way too much compound and/or not enough mounting pressure.

Pat

 

[ZapPat]

It's the inductive path that effects how current is shared as well.

If you remember the cell tester I built, a linear array of FETs with the drain tabs soldered directly to heavy copper bus bars, under the IR camera, we could see which FETs were getting used, and which were just along for the ride. It was 8x TO-247 fets in a row, and the FET closest to the outlet was like 40degF hotter than the fet on the other end of the 1/4" thick copper bus, which given the excellent heat transfer of copper, tells you the FETs at the beginning of the linear array aren't doing jack-squat. I'm confident I could have done an array with only 4 fets arranged in a circular inductively balanced configuration and arrived with a more robust end product.

Linear arrays like the brushed controller shown by Marko in the pics above are very simple to build, very pleasing to the eye, and very pathetic in current sharing function.

Luke,

While I agree that a well done circular array of paralleled FETs would have superior current sharing when compared to a linear array... but designing an easy to build circular array that also features a symetrical gate drive layout might prove much more difficult than a linear layout. I for one one would love to see your circular design come to life, and I would also love to compare it to my "old school" linear array. :wink: I have put much thought in my linear layout, and the main power path doesn't even touch the PCB (copper bars of course) and switching times are very low. One thing is for sure - don't use that linear FET array you made (and destroyed) as an example of what linear FET arrays can do, as it is pretty "prototypish" (no insult intended). Have you worked on that circular array again since that photo last winter BTW? I have tried imagining the layout of such a beast since you posted that idea, but it looks to be quite a challenge IMO - specially when you included the gate driver section that would have to wrap around the FET array.

Pat

 

[markobetti]

It's the inductive path that effects how current is shared as well.

If you remember the cell tester I built, a linear array of FETs with the drain tabs soldered directly to heavy copper bus bars, under the IR camera, we could see which FETs were getting used, and which were just along for the ride. It was 8x TO-247 fets in a row, and the FET closest to the outlet was like 40degF hotter than the fet on the other end of the 1/4" thick copper bus, which given the excellent heat transfer of copper, tells you the FETs at the beginning of the linear array aren't doing jack-squat. I'm confident I could have done an array with only 4 fets arranged in a circular inductively balanced configuration and arrived with a more robust end product.

Linear arrays like the brushed controller shown by Marko in the pics above are very simple to build, very pleasing to the eye, and very pathetic in current sharing function.

Luke, what material would you choose then ? Aluminium ..?

 

[tostino]

I think Luke's point is that a linear array is no good. Not that copper is a bad material (which it's not).

 

[markobetti]

I think Luke's point is that a linear array is no good. Not that copper is a bad material (which it's not).

thanks

 

[tostino]

So doc, what progress have you made so far? I'm really interested to see what you end up with. My boards should be here next Monday.

 

[texaspyro]

My intuition says that individual gate drive resistors is the way to go, and I went that way for a long time. Then I tried the common gate drive approach and it did work better. It might just be my applications.

I think that if you match Vth, then a single gate drive works best (maybe with the smaller resistors feeding each gate from a common one). If you don't properly match the fets, then the individual gate resistors is probably best. Anyway, use FETs from the same batch code and spend the time to match your fets Vth. It is, by far, the single best thing that you can do.

 

[Doctorbass]

This gate driver discussion is interesting but from now i'll spend my time on thesting the actual desing with our most common DIY methods.

tostino, the controller is in progress. I should get the aluminum bar finished on monday. i might use 0.500" thick bar for the low side and 1/4 thick "L" bar for the high side. The reason is that according to some of you the low side is the one that generate the most heat so i preffre using 1/2 thick aluminum bar. that will end in a higher dt heat inertia and smooth the temp increase allowing a bit higher peak power.

I will also need to make 18 perfectly aligned holes in every of the 2 bar for the fet drain tab.

The high side bar will have flat head screw dur to the lack of space of 1/4 width room and because the high side will also make contact directly with the side panel of the case.

i'm still not decided if i make it watercooled or not.. i dont like the idea of more complex system and added weight and pump and tube and radiator....

But if i change my mind i will be able to glue some copper tube to the bar with some high thermal performances RTV.. like the CV-2566 or CV-2960 from Nusil... or Dow Corning 93-500.



Do you think that puting copper tube to the positive and negative rail on the pcb really help??

I mean does the heat transfer from the drain pin 2 of each fet is important enough compared to the screw tab ?

Doc

 

[tostino]

If anything Doc, I think that putting water cooling on the PCB may increase the 75 amp package limit a little, by keeping the legs at closer to room temp. There has to be at least some heat that is transferring from the fet through the legs to the pcb. I'd say it couldn't hurt lol. Even if I don't water cool it right away, I am gonna be using copper pipe to beef up my traces instead of just wire. That way I at least have it available in the future if I decide to cool it.

Edit: I do think that water cooling the screw tab is going to do WAY more than the PCB. But if you have the ability to do both, why not

 

[John in CR]

Hey Doc,
Don't you have the windings split in half on that motor? Don't you also have two 18fet controllers sitting around? I bet 1 18fet controller on each half of the windings run in parallel would be easier on the controllers due to double the resistance that one controller was seeing before. I also think two 18 fet controllers would be better than one 36 fet controller.

 

[liveforphysics]

I agree with you John.

 

[John in CR]

I agree with you John.

I wonder what the net result of double the resistance but half the BEMF would be though.

 

[amberwolf]

If the BEMF is reduced, current flow increases (for the same incoming voltage), because there is less motor-generated voltage pushing it back.

Reduced BEMF also would mean higher current flow for more of the speed range of the motor.

 

[tostino]

John, would it actually have less BEMF if you were to just split the windings up at the end? The number of turns for each phase is still the same, the wire is just smaller in effect.

Unless I don't understand exactly how Doc would split the windings on his motor...

 

[Doctorbass]

Hey Doc,
Don't you have the windings split in half on that motor? Don't you also have two 18fet controllers sitting around? I bet 1 18fet controller on each half of the windings run in parallel would be easier on the controllers due to double the resistance that one controller was seeing before. I also think two 18 fet controllers would be better than one 36 fet controller.

Nice idea... BTW.. I know someone here that will provide us DUAL MOTOR CONTROLLER sooon !! :wink: with single throttle!!


About the dual controller 18 fets and dual winding.. This is a good idea... but you forget that it need 6 wired out of that motor!! :| .. and 6x AWG 10 is a bit difficult to me to cary in an X5 axel !

The 5303/06 motor is now a pure 5303 no relay.. no split... i reconverted it as OEM... 5306 just suck!.... Luke you was right about that

Doc

 

[John in CR]

John, would it actually have less BEMF if you were to just split the windings up at the end? The number of turns for each phase is still the same, the wire is just smaller in effect.

Unless I don't understand exactly how Doc would split the windings on his motor...

Half the copper with same number of turns must be less BEMF, just like you´d get less power from it.

Amberwolf,
I´m looking at how difficult these low turn count motors are on controllers, so my point was each controller would get the benefit of double the copper resistance compared to one controller and full windings, but wouldn´t that benefit be reduced by each controller seeing only half of the BEMF that it would see with full windings.

Doc,
Sorry, I thought you still had those winding strands split. I am a bit disappointed that you didn´t get to test hill climbing affects on the controller, because that is the big problem with great low turn count motors. Hills is where the series/parallel would have paid off, lower speed up the hill, but better efficiency and easier on the controller...Low turn count motors eat controllers on hills.

Yes, if everyone tried a low turn count hubmotor driven with sufficient current, they wouldn´t use what they incorrectly believe are higher torque motors. They are just lower speed motors, and nothing replaces the great acceleration we have at 30mph or higher.

John

 

[ZapPat]

John, would it actually have less BEMF if you were to just split the windings up at the end? The number of turns for each phase is still the same, the wire is just smaller in effect.

Half the copper with same number of turns must be less BEMF, just like you´d get less power from it.
Amberwolf,
I´m looking at how difficult these low turn count motors are on controllers, so my point was each controller would get the benefit of double the copper resistance compared to one controller and full windings, but wouldn´t that benefit be reduced by each controller seeing only half of the BEMF that it would see with full windings.
John

Just to clear this up, tostino is right that if you do split the windings at the ends and thus leave the same number of turns with just less copper (more resistance), then you do get the exact same BEMF for each set of wires as you did before. It's the Nm/Amp that will drop in this situation for each winding.

However, from what I understand, Doc split his windings into two parts in series, not in parallel as is described above. This would yield two windings with half the BEMF each, but the same Nm/Amp for each winding as they each have half the turns.

Pat