20kW Powerstage for "Infineon" 12 mosfet Controller

h0tr0d

h0tr0d

1 kW
Joined
Apr 28, 2012
Messages
460
What do you think about this powerstage?
powerstage.jpg
View attachment 1

I want a 20kW ebike controller (40S Lipo, 150A/phase).

Parts include in the design:
12 x IXFK230N20T (200V 230A) Mosfet's http://ixapps.ixys.com/DataSheet/DS100133A(IXFK-FX230N20T).pdf
30 x CGA9N3X7R2E105K (250V 1uF) Caps DF2.5% http://www.tdk.com/pdf/cga_series.pdf
2 x Allegro 200A current sensors http://media.digikey.com/pdf/Data Sheets/Allegro PDFs/ACS755xCB-200.pdf

3D PDF

My questions are:
  • Parts you would change
  • Component disposition
  • Doable?
 
You're going to need a good driver for those FETs. The driver on the Xiechang boards can only deliver around 100mA of gate drive, which isn't enough (by a fair margin) for a bunch of FETs with more than twice the total gate charge.
 
Those current sensors cannot handle 200A continous (5W dissipation in it). I use the same in my 5,5kW inverter with peak current of 200A per phase. 110A rms.

Driving it with sine algorithm would limit your phase amps to a max of 200/Sqrt(2). And, 200V200A from battery results in 280A phase rms amps - 400A peak. Trap algorithm would allow more power.

And then there's the derating of the mosfets you need to do. I use 100V mosfets in my inverter, that is for 72V operation.

Question of snubbers, step down regulator (for logic and mosfets), brains, communication, protection (thermal, esd, emi...). Build your powerstage and you're 20% along the road :p
 
Jeremy Harris said:
You're going to need a good driver for those FETs. The driver on the Xiechang boards can only deliver around 100mA of gate drive, which isn't enough (by a fair margin) for a bunch of FETs with more than twice the total gate charge.
Click to expand...

How many 4110s would you think 1 amp can push? I've been looking for any data on how the switching is affected by going from 10ohm to 20ohm gate resistors, specifically how much it slows down the gate charge. I bet the datasheet would have enough info to determine it, but it is still a bit beyond my learning.
 
JH:
100ma per fet, phase...?

this fet has less Qcc but worse Rdson...
http://ixdev.ixys.com/PartDetails.aspx?pid=4153&r=1

Would a 24 or 36 fet controller have enough "juice" for the wanted fets?

TS:

it's 168V 150A/phase max, ~25kW not 40kW. Xiechang I would imagine is trap algorithm... But honestly didn't understand your calcs :oops:
 
have you thought about water cooling ?

it would make it easy for you to pack the controller out of the way somewhere and then mount the radiator somewhere else ?
 
knighty:

100% agreed but not for this case. By my calcs (around 200W), I don't think it's needed. But for my c80100, I'll try to watercool it.
 
Is it possible to put more powerful gate drivers and connect them to the controller?
 
Jeremy Harris said:
You're going to need a good driver for those FETs. The driver on the Xiechang boards can only deliver around 100mA of gate drive, which isn't enough (by a fair margin) for a bunch of FETs with more than twice the total gate charge.
Click to expand...

You mean, even lowering the gate resistors?
 
Knighty:

If it proves necessary:
http://www.hobbyking.com/hobbyking/store/__13412__Water_Colling_Plate_for_90A_ESC_Super_Version_Surge_Cursher.html
or
http://www.hobbyking.com/hobbyking/store/__21192__Blue_Aluminum_Battery_Water_Cooling_Board_2pcs_.html

plus some PC water cooling pump, radiators, fans...
 
Ppl:

what about the caps, their specs and positioning regarding ESR and ESL?
 
Probably not exactly easy, but you can have a wire bypass on the current sensors do share some of the current.
You can also use one of those sensors that just wrap-around a wire.
 
Gate drive is fairly complex, and not just a question of stuffing enough current into the gates, unfortunately. The gate drive resistors on the Xiechang controllers, for example, have a single 91 ohm resistor to drive between 1 and 3 FETs, depending on the version. They have a transistor that shunts the gates off directly though, so the turn off gate drive current is a fair bit higher than the turn on gate drive current.

However, the Xiechang controllers (and the Wuxi controllers (Greentime, Hua Tong, KU series etc) that use a near-identical gate drive circuit) have a pretty slow FET switching speed, the best they do is around 1uS or so, which results in quite high FET switching losses (in practice, the switching losses are often around 50% of the total FET power dissipation in these controllers).

If you want to use a Xiechang as the core for a much more powerful controller, then I'd suggest using something like an NCP5181 driver, with gate resistors of around 10 ohm, but bearing in mind that board layout and the proximity of things like the commutation capacitors (and their ESR) can have a big impact on the fidelity of the gate drive signal. It's almost as much art as science, as board layout, track lengths etc make an appreciable difference to the way the FETs turn on and off. The Xiechang controllers (and some of the others) get around many of the potential problems by just using a low current gate driver and accepting the losses from the slow switching speed, a reasonable trade off for a low power controller.
 
Jeremy Harris said:
Gate drive is fairly complex, and not just a question of stuffing enough current into the gates, unfortunately. The gate drive resistors on the Xiechang controllers, for example, have a single 91 ohm resistor to drive between 1 and 3 FETs, depending on the version. They have a transistor that shunts the gates off directly though, so the turn off gate drive current is a fair bit higher than the turn on gate drive current.

However, the Xiechang controllers (and the Wuxi controllers (Greentime, Hua Tong, KU series etc) that use a near-identical gate drive circuit) have a pretty slow FET switching speed, the best they do is around 1uS or so, which results in quite high FET switching losses (in practice, the switching losses are often around 50% of the total FET power dissipation in these controllers).

If you want to use a Xiechang as the core for a much more powerful controller, then I'd suggest using something like an NCP5181 driver, with gate resistors of around 10 ohm, but bearing in mind that board layout and the proximity of things like the commutation capacitors (and their ESR) can have a big impact on the fidelity of the gate drive signal. It's almost as much art as science, as board layout, track lengths etc make an appreciable difference to the way the FETs turn on and off. The Xiechang controllers (and some of the others) get around many of the potential problems by just using a low current gate driver and accepting the losses from the slow switching speed, a reasonable trade off for a low power controller.
Click to expand...

Using your calcs from here http://endless-sphere.com/forums/viewtopic.php?f=30&t=20919#p304664
Switching 2 fets = 2*378nC and a 10ohm gate resistor = 600 ns
 
I'd have said 10 ohms should be about right with a good gate drive chip that can deliver the peak current. You probably can't just swap the 91 ohm resistors on a Xiechang for 10 ones, though, as that would almost certainly cook the tiny gate drive transistors on the board. Also, the gate drivers on the Xiechang seem to be slugged down a bit by design to slow down the FET switching speeds.

If you made up some driver boards, with something like NCP5181s, and hooked these up to the logic lines from the Xiechang ucontroller, then that might work OK.
 
This will not work. Or, even if it does, will never be a reliable and efficient system.
You can't simply put more fets and hit the road.

To pull this off, many things must be done differently. Beginning with drivers, and ending with control algorithms.
 
circuit said:
This will not work. Or, even if it does, will never be a reliable and efficient system.
You can't simply put more fets and hit the road.

To pull this off, many things must be done differently. Beginning with drivers, and ending with control algorithms.
Click to expand...

Well, The Xiechang controllers are available as 36 FET versions already, with the same controller chip and firmware as the smallest controllers, so that suggests there's no need to worry about any algorithms being changed for this sort of power level.

The drivers certainly need to be changed, but these accept a logic level input and I can't see any reason why the 6 outputs from the 116 ucontroller shouldn't drive something like an NCP5181 just as readily as the discrete drivers that are already there. Some other bits of hardware would need to be changed, like the current sense circuits (there are two on the Xiechang, one analogue, one a transient detector that triggers an interrupt and shutdown), but none of it is insurmountable for someone absolutely determined to go down this route.

Whether it makes sense is a separate question and a personal one. In my view I'm not sure it does, as there are better ways of getting the front end functionality than chopping up an existing cheap controller (using something like Lebowski's chip or maybe one of the other FOC control chips that are around, for example).
 
Great input so far, thanks.

Jeremy, three questions:

- a 36 fets XieChang controller will also only supply ~100mA??? (if so :p)

- Can you supply info, datasheet, pinouts, etc of the xiechang stuff? Can't find them around here...

- As anyone tried your "NCP5181" solution? Seems to be an intelligent solution, don't know if it's really necessary...
 
h0tr0d said:
Great input so far, thanks.

Jeremy, three questions:

- a 36 fets XieChang controller will also only supply ~100mA??? (if so :p)

- Can you supply info, datasheet, pinouts, etc of the xiechang stuff? Can't find them around here...

- As anyone tried your "NCP5181" solution? Seems to be an intelligent solution, don't know if it's really necessary...
Click to expand...

The 36 FET Xiechangs have more than one driver for each bank of FETs, to get around the drive current problem.

I'll have a dig around and see if I can find the schematics. They are on here somewhere...............

I don't think anyone has looked at using a better driver, like the NCP5181 with these controllers, probably because there may be easier ways of making a big controller. As well as adding the driver chips and bigger FETs, you're also going to have to work out how to handle the ucontroller current limiting systems. There are two of these, one linearly controls the battery current, the other acts as a peak current detection system, which act as an over current fault protection shut down. Both of these circuits use the voltage sensed across the shunt to measure current, so using a larger shunt with a lower resistance, plus adjusting the programming may get this to work.
 
Dont forget cap placement as a design priority.
 
I was trying to follow the xiechang plus bigger fets because I thought it was the easiest.

What I need is some programmable brain, good fets, phase current limiter, pedal sensors, torque and cadence, and a brake cutoff...

LFP
Did you see the 3D pdf on the first post?
 
h0tr0d said:
I was trying to follow the xiechang plus bigger fets because I thought it was the easiest.

What I need is some programmable brain, good fets, phase current limiter, pedal sensors, torque and cadence, and a brake cutoff...

LFP
Did you see the 3D pdf on the first post?
Click to expand...

Lebowski's chip is a pretty good basis for a great controller. What's more it has torque control built in, which is a pretty good thing for a high power controller. It's not much more work to build a controller around this chip compared with adapting a cheap ebike controller.
 
I agree it's a pretty good basis. But I don't have your manufacturing skills/capabilities. http://www.endless-sphere.com/forums/viewtopic.php?f=2&t=45291&hilit=another+lebowski&start=15#p667767

For my set of limited electronic skills, a 24 or 36 fets controller (with decent gate driver current), my "powerstage" and other simple stuff for phase current management would be nice.

Even a 6 fet controller with 2 extra gate driver soldered on top of the existing gates drivers could work (parallel connection), I think.

circuit said:
This will not work. Or, even if it does, will never be a reliable and efficient system.
You can't simply put more fets and hit the road.

To pull this off, many things must be done differently. Beginning with drivers, and ending with control algorithms.
Click to expand...
http://endless-sphere.com/forums/viewtopic.php?f=2&t=32853
Already been done already with success, despite of really bad design. Efficient? some doubts. Working? hell yeah! :wink:
 

Attachments

  • current control throttle ron.jpg
    current control throttle ron.jpg
    104.8 KB · Views: 1,280
A C80100, probably watercooled.
 
You must log in or register to reply here.
Back
Top