Low voltage, high amps, can it scale > 100kW?

[Punx0r]

You'd need an awful lot of FETs for 25,000A (to make 100kW @ 4V). It would make poor use of their voltage capabilities, resulting in very low power density for the controller.

 

[Lebowski]

I would think, with increasing currents, at a certain point wiring inductance becomes the bottle neck. Based on the physical size of the controller components the wiring inductance cannot
be reduced lower than such and such. Then there is no way to reduce the energy storage in the parasitic inductors (other than reducing the current), and its this parasitic energy that
kills FETs (or whatever you have for a switching element).

About the parasitic energy.... every PWM cycle wiring inductance is charge with energy (when there is current flowing) to the account of 0.5*L*I^2 . During the other part of the PWM cycle
(when the current in this part of the wiring inductance is stopped) all the energy is dumped in the FETs. So wiring inductance is like a bucket 'used' to transfer energy into the FETs.

 

[Danny Mayes]

You'd need an awful lot of FETs for 25,000A (to make 100kW @ 4V). It would make poor use of their voltage capabilities, resulting in very low power density for the controller.

Yes sure. I was more thinking light EV applications say around 10Kw. Like you say power density would suffer if you were not using the full voltage range of the fet. I am out of my depth here but is there not lower voltage fets? Like 6 volt? I guess I am off topic now though as I am not talking >100Kw...

 

[Punx0r]

I'm not sure. I suspect there isn't a straight trade-off between voltage and current capability, i.e. a 10V MOSFET won't handle the 10x the current of a 100V version using the same silicon die. I could be wrong.

Lebowski makes a very interesting point regarding inductance.

I think Liveforphysics has previously suggested a similar of single-cell voltage for an ebike-sized application (to make use of Nissan Leaf battery modules, IIRC), so it's probably not such a bad idea for the right application

 

[Danny Mayes]

Thanks Punx0r.

Because i know nothing about IGBTs the first place i look is wikipedia. there is a picture of an IGBT module with a rated current of 1,200 A and a maximum voltage of 3,300 V. Is that not nearly 4mw?? Is it possible to have something similar do 1 million amperes at 4 volts? and if not what about say a quarter of that? Maybe then i would be back on topic

There would be benefits also in the motor would require less turns. Maybe only one turn. And maybe that one turn could be made out of copper tubing that you pump water through. The pipe could be seamless from the IGBT or whatever is switching to the motor. You are cooling right at the source and could carry more amps...

 

[Arlo1]

I think the bottle neck will be fet leg limits the rest of the system can be designed to work at what ever power levels you need. So you will run a set of TO220 fets at their leg limits or 247/264 at their leg limits for ~10 sec burst and a little less for continuous. But that means the power density will go up as you run higher voltage. Until a manufacture runs bigger legs...

With current package designs you can have something with 150v rated stuff and change out the fets and caps for 600v rated igbts in the same package size and run ~350-400v at the same amperage. No manufacture is going to pay the size penalty to run lower voltage and take up more space to try to increase the amperage to make up for the lost power from reducing the voltage.

 

[Punx0r]

The spec on IGBTs is something else, but I believe their forward voltage drop is their main drawback. At high voltages this is a small proportion of the total, but at low voltage/high current I think it will be a killer...

 

[IanFiTheDwarf]

In stead of winding the motor in a star why not drive each phase seperately that will half the required voltage straight away. Ok it will require three single phase bridge drivers rather than a single three phase driver, effectively twice the number of fet's but that's not necessarily a bad thing.

Go even farther and rather than winding each tooth in series have a driver for each.

On my axial motor design I have 18 teeth/coils or whatever you want to call them and I am thinking of having18 drivers mounted directly to the coils. No end turns, no connecting cables and all driven from a quarter the supply voltage.

Now do I run two huge buzzbars to power all the drivers or do I group them and fuse them seperately so I can use smaller cheaper cable and fuses and possibly allow the motor to still run if a driver or coil fails.

I have been trying to think if this arrangement will need any change to the controller ic and I think it is just a matter of connecting everything up the right way for driving but I'm not sure about the regenerative breaking. Maybe someone who knows more than me can comment here.

 

[Lebowski]

In stead of winding the motor in a star why not drive each phase seperately that will half the required voltage straight away. Ok it will require three single phase bridge drivers rather than a single three phase driver, effectively twice the number of fet's but that's not necessarily a bad thing.

Go even farther and rather than winding each tooth in series have a driver for each.

On my axial motor design I have 18 teeth/coils or whatever you want to call them and I am thinking of having18 drivers mounted directly to the coils. No end turns, no connecting cables and all driven from a quarter the supply voltage.

Now do I run two huge buzzbars to power all the drivers or do I group them and fuse them seperately so I can use smaller cheaper cable and fuses and possibly allow the motor to still run if a driver or coil fails.

I have been trying to think if this arrangement will need any change to the controller ic and I think it is just a matter of connecting everything up the right way for driving but I'm not sure about the regenerative breaking. Maybe someone who knows more than me can comment here.

Disadvantage is that every phase will need 2 output stages, as each side of the phase winding needs its own output stage (so 4 FET minimum)

 

[flathill]

Homopolar motors are an attractive alternative to conventional dc and induction machines for hybrid and all-electric vehicles, both civilian and military. Known for their high current capability, this inherent characteristic allows high power levels to be achieved from a low voltage power supply for improved safety and reliability over higher voltage systems. Low voltage controllers lend themselves to the use of high efficiency MOSFET switching devices and the unique design simplicity of homopolars provides great potential for reducing the cost of EV drive trains.

Work on this project began in May 1995, and a prototype motor design has been completed. This design uses a four-pass armature and operates at a peak current of 5,000 A from a 48 V battery pack. Full power efficiency is currently at 87% with the majority of losses (about 10%) coming from the brushes.

Brush testing is in progress to address efficiency improvements, document wear rates, and confirm operation of several new low-cost and compact actuator designs. Test data obtained so far shows excellent operation of a constant force spring-actuated brush and brush holder design. Brush wear rates at normal driving conditions appear to be far less than those obtained from worst case speed and current loading conditions cited in previous studies. This is very encouraging considering the typical duty cycle of the bus.

To regulate power flow to the motor, UT-CEM is developing a high-current pulse width modulator (PWM) controller based on a highly parallel array of MOSFET switching devices. For low voltage systems, MOSFETs are an ideal choice given their low on-state resistance, high switching frequency, low switching losses, and ease of parallel operation. A 300 A prototype has been assembled, and testing is under way to confirm operation with a low impedance load.

-----
After they built the 300a proto they built the full size 5000A 48v controller with off the shelf mosfets back in the 90s and drove the bus around. There is was an obscure SAE paper on it. There is no limit on how many mosfets you can parallel

So there is the answer to your question. 240kw from a 48v source has already been done 20 years ago

 

[Thud]

Here is a question from a controller ignorant:

I hear talk of the 5-phase motor...but I am more curious to hear about 5 phase control.

Will Lebowskis chip drive 5 phases?
what about any other standard motor drive chips?

I can imagine a 20fet TO247 simple controller....but I can imagine unicorns too.

Please, Educate me bois.

 

[Lebowski]

Algorithm wise a 5-phase is no problem. But finding a microcontroller that
has a 5 channel motor control PWM block, that's another issue... If your motor
has a multiple of 3 phases (for instance 6 like Johns motors) then you can
use multiple of my controllers (don't know whether other controllers will work)

 

[IanFiTheDwarf]

Disadvantage is that every phase will need 2 output stages, as each side of the phase winding needs its own output stage (so 4 FET minimum)

72 FET's is a lot but still cheaper than the IGBT module required to produce the same power, maybe reliability would suffer and I'm not sure about the affects on the driver circuit being within millimetres of the motor windings and magnets.

 

[Punx0r]

Wasn't there recently a car-sized motor (Australian?) someone had developed and used a driver per coil?

 

[Arlo1]

Disadvantage is that every phase will need 2 output stages, as each side of the phase winding needs its own output stage (so 4 FET minimum)

72 FET's is a lot but still cheaper than the IGBT module required to produce the same power, maybe reliability would suffer and I'm not sure about the affects on the driver circuit being within millimetres of the motor windings and magnets.

Hogwash. You can get very expensive fets and very cheep IGBT modules. Or the other way around. Also who says you need to use a module? You can run high voltage IGBTS in the same package size as any fet. I'n fact this is what Tesla already does. You would have a IGBT in your hand thinking it was a mosfet only to find it can run at 4-5x the voltage making it 4-5x the power density.

 

[IanFiTheDwarf]

Disadvantage is that every phase will need 2 output stages, as each side of the phase winding needs its own output stage (so 4 FET minimum)

72 FET's is a lot but still cheaper than the IGBT module required to produce the same power, maybe reliability would suffer and I'm not sure about the affects on the driver circuit being within millimetres of the motor windings and magnets.

Hogwash. You can get very expensive fets and very cheep IGBT modules. Or the other way around. Also who says you need to use a module? You can run high voltage IGBTS in the same package size as any fet. I'n fact this is what Tesla already does. You would have a IGBT in your hand thinking it was a mosfet only to find it can run at 4-5x the voltage making it 4-5x the power density.

Sorry, what I should have said was the IGBT module (actually an IGBT IPM) I was looking at. the reason I was looking at the IPM was because the drivers are on the module and I'm no expert in power electronics. there is a lot of info and has been a lot of good work done on here on driving FET's including your own so I thing I could follow what others have done on FET's but I would be lost trying to drive discrete IGBT's.

 

[Thud]

Lebowski, thanks for the reply.

Algorithm wise a 5-phase is no problem. But finding a microcontroller that
has a 5 channel motor control PWM block, that's another issue...Please detail that statement? IIUC that is the whole point of 5 phase control...to realize the angular advantage that topology would afford.

If your motor has a multiple of 3 phases (for instance 6 like Johns motors) then you can use multiple of my controllers (don't know whether other controllers will work)

I discount the stacked/series' configurations that are in Johns examples....it is fairly simple to wind a stator with split 3-phase & gang multiple controllers...but does it address the possible advantage of 5 phase rectification?

& I wish I had the time & skillz to assemble one of your controllers....waiting patiently for a complete list of parts, board schematics, test point measurement's for calibration & default programming before I jump into the deep end.

thanks for the input...I am inspired.

 

[zombiess]

There is no limit on how many mosfets you can parallel

There is a limit to how many device may be paralleled, that limit is equal current sharing.

Modules, even though expensive can quickly become a more cost effective option vs many parallel devices. If the operating mode has a high continuous duty cycle then a large module might be a be a better choice because the large format modules have a lower Tjc than smaller TO-220 and TO-247/TO-264 devices. Modules can also far exceed the current leg limit that smaller devices have which is about 35A continuous from what I've seen in commercial devices. A high burst output can really skew ones view on what power handling devices are actually capable of.

 

[Arlo1]

In side a modual is a set of mosfets or igbts. So the only advantage for a modual is its had someone design/build/test it that knows what they are doing. But they are not the most power dense solution when looking at a whole three phase inverter. As well I've put over 200amps through a TO247 leg for over 10 seconds. The continus rating of 120-160 amps DC with proper cooling is nmo lie!

 

[squeegee]

Wouldn't the IXFZ520N075T2 be a good candidate to Lo V/Hi A ?
Couldn't find much info other than a 2010 advanced tech sheet and that 22 are available at mouser (for a hefty $).
Except for high Qs and Cs, specs look nice. Could be cooled on both side (with some ingenious yoga mounting) and 1.3mOhms...nice
I guess running that at 72V is a little to close, so parallel 3 for an 18FET module should get 1200A (?) at 60V(?) = 72KVA


edit: just realized at spec it would >100KW

 

[Arlo1]

Your peaks will need to be a fair bit below 1265 and the RMS phase amps should be .70.7% of the peaks so DC amps will be a max around 700 in best case scenario.

 

[liveforphysics]

There are packages that don't pose leg limits, and still can be cooled.
http://www.ixys.com/ProductPortfolio/PowerDevices.aspx
http://www.eeweb.com/company-news/ixys/avalanche-rated-fast-power-mosfet

 

[Arlo1]

Wow data sheet says less voltage drop as increasing current that will make parallel components tricksy.

 

[flathill]

There is no limit on how many mosfets you can parallel

There is a limit to how many device may be paralleled, that limit is equal current sharing.

Modules, even though expensive can quickly become a more cost effective option vs many parallel devices. If the operating mode has a high continuous duty cycle then a large module might be a be a better choice because the large format modules have a lower Tjc than smaller TO-220 and TO-247/TO-264 devices. Modules can also far exceed the current leg limit that smaller devices have which is about 35A continuous from what I've seen in commercial devices. A high burst output can really skew ones view on what power handling devices are actually capable of.

Wrong. They are many different methods of forced current sharing. You need to match mosfets becuase of of unequal sharing of losses during turn on and turn off. This is because one may be on before all the other in parallel turn off. One solution is drive each transistor with a different delay. Then you no long need to even closely match the mosfets, which are never truly matched anyway as they heat/age. The delay can be tuned in real time with feedback.

 

[zombiess]

Wrong. They are many different methods of forced current sharing. You need to match mosfets becuase of of unequal sharing of losses during turn on and turn off. This is because one may be on before all the other in parallel turn off. One solution is drive each transistor with a different delay. Then you no long need to even closely match the mosfets, which are never truly matched anyway as they heat/age. The delay can be tuned in real time with feedback.

I'm not wrong, it's called diminishing returns and it's real. A delay isn't a very good solution for trying to equalize the current sharing either as there is still the problem of having devices in the ohmic region when others might be off or fully on. Another interesting thing the happens in part of the ohmic region is the temperature coefficient is negative not positive as it is when in saturation. It's rare to have a failure due to this from what I have read, but I have personally observed this phenomenon. I do need to test it further as I did not perform it under a sizable load.

I do believe I have come up with at least 2 low cost methods to increase the number of devices which can be paralleled. Needs more testing which means I need to finish up some projects I'm working on.

The main reason MOSFETs parallel so well is not that they have a positive temperature coefficient once on as most people think. It's actually because the manufacturing process produces a high yield of devices with fairly close tolerances. The gate oxide layer thickness between devices is the #1 determining factor how well it will parallel with like devices. The PTC is #2 and close tolerance of RDSon is #3. All of these need to be taken into consideration when trying to scale way out.

The legs on small package MOSFETs with advertised high current are usually the ultimate current limit factor. They get really hot, sometimes hot enough to melt solder and cook the PCB (been there done that). There is also a pretty high loss in the legs when pushing them hard. I've measured the resistance of several MOSFET legs and it's pretty easy to lose 10W per MOSFET at the legs if you push them hard. Big bus bars help pull heat out of the legs based on personal observations.

Cooling multiple small package devices also becomes a challenge as they often do not have favorable Tjc impedance. If the operation of a device is continuous this concern is very high on the list. A reasonable rating for conventional small package devices is < 40A.

The package Liveforphysics posted is interesting.

If it were easy to add a bunch of dice into a module, manufacturers would be coming out with ever higher current handling devices.

Who wants to talk about the massive switching / diode losses many parallel devices have? It can become quite substantial. Sure SiC devices perform with 30-60% less switching losses, but then the penalty for higher cost must be paid. I haven't seen any low RDSon High current SiC devices yet, but I haven't looked hard.

Speaking of switching losses there is also the issue of meeting radiated emissions standards and switching 1000's of amps in 100s of nanoseconds would sure make that interesting.

This is an engineering game. Trade offs must be made. Bus voltage, amps, number of windings in a motor, device package, number of devices, etc.

I just had another thought... how much DC Link cap would be required in order to meet the ripple current demand. I'll probably have more concerns if I think about this for a while. If someone is serious about this they should start crunching numbers and thinking of the problems and ways to overcome them. Just had another, at some point gate drive path could run into EM issues.

As well I've put over 200amps through a TO247 leg for over 10 seconds. The continus rating of 120-160 amps DC with proper cooling is nmo lie!

Stop, put down the Koolaid they have you drinking. You will not be putting 120A continuous (not 10 seconds) through the legs of a TO-247 package continuously without a very extreme cooling method. There is also the matter of the wire bonds and thermal stresses which get applied to the device over it's life span. High stress and temps will shorten the life of the part. Wire bonds can also become an issue.

Datasheets look great, but they are far from reality when it comes to continuous current handling capability. I myself have done ~600A per TO-247 device for a short period.

If you have the time, I would like to see a current test done on different packages. I'd like to see how long they can last with DC at a given DC current for 1 hour.

I'm not against trying to find the balance between high current and high voltage, I want to find solutions to the problems. Poking holes in ideas is how I usually get to my own goals.