liveforphysics
100 TW
strantor said:No way man I had the brushless design in the bag! lolliveforphysics said:Are you glad you're only starting with a brushed controller yet my friend?
8) 8) 8) 8) 8) 8)
MOSFET cooling in commercial controllers; how do they do it?
- Thread starter strantor
- Start date
8) 8) 8) 8) 8) 8)
Electroglide
100 mW
Here is another 2 cents worth of opinion. If I repeat anyone else's comment sorry for extra's. (EDIT: comments refer to BLDC but are applicable to 1/2 bridge or H-bridge)
Fets do share reasonably good if they share a common electrically connection point and share a common heat spreader. They are positive temperature coefficient devices...ie the Rdson goes up with temperature which means that parallel mosfets will reach an current equilibrium point that is dependent on the die temperature and the circuit impedance between the devices. That is why you see ballast resistors sometimes on parallel IGBT's since a lot of them are negative coefficient devices where Vce decreases with temperature. I have seen newer IGBTs that are now positive coefficient on the Vce sat.
The package of a mosfet defines its maximum long term continuous current if you are able to keep the heatspreader at a temperature that keeps the die below its maximum allow operating temperature. Tj = Pd * Rjc + Ta. (not considering life expectancy since I am sure someone will comment about that) TO220 is typically around 75Amps max though I have seen higher with the newer generation TO220 packages. In terms of short term current ratings, you can look at the datasheet at the transient thermal impedance curves for 1 time pulses, and look at what the heat capacity of the heatsink that your fets are attached to. Realistically for long term average current, I think that you are limited by the thermal resistance path from the die to where you are trying to remove the heat too. So if that is from the die to ambient, for steady state, it will be less than the 75A or whatever the package is rated to. For the short term current, it is kinda hard to characterize since it is really dependent on the design. It is the short term current that you probably want to figure out for motor acceleration purposes. That is like the 1 min rating that the motor controller manufacturers sometimes give.
The mosfet spec sheets are not always that easy to use. The headline specs like typical Rdson, voltage , and current are given at the ideal temperature of 25C and are for the device only, not the device within your controller. So you can't say I want a 300A controller so I will use one TO220 mosfet rated at 300A to do it for reasons that I gave above. My opinion for figuring out how many mosfets you need is only a swag since you don't always know what your thermal design will be since it is dependent on the controller package and thermal insulating tape used to isolate the mosfets.
1) Identify the maximum current that you want to push through the motor
2) Identify the average continuous current that you want to run the motor at
3) Identify the maximum voltage that you want to operate at (you want some margin to prevent avalanching)
4) Take swag at how fast that you want to switch the device on and off at (Ton, Toff mosfet time) Chinese controllers are in the order of 3 usec or so.
5) Identify the switching frequency that you want to PWM at.
6) Identify the thermal insulating tape that you will use(if using a non isolated package)
7) Take a swag at how hot you want your heatspreader to rise to, that you mosfets will attach to.
7) Choose some candidate mosfets that you think are good
Ok, so now that you have that you can swag the power losses
Switching losses for a swag can be estimated as Psw = Ids * Vds * (ton+toff) * Fsw /6. This assumes that the devices are being hardswitched ie it is the device interrupting or sourcing the current. I am sure there are better estimates to calculate the switch losses, but this is the one I use. (EDIT Ids will be the motor current (phase for BLDC or Armature for DC) Vds is the battery voltage)
Conduction losses assuming a BLDC configuration with synchronous rectification (ie one low side bank of mosfets always turned on during the sector and a pair of mosfets banks switching in a half bridge configuration) will include the half bridge that is switching and the bank that is not switching. We will ignore the deadtime and losses associated with it for the purpose of figuring out the number of mosfets.
Pcond total ~= PD(half bridge) + PD(sync), PD half bridge can be estimated as the sum of the upper and lower Pd which are duty cycle dependent. So Pd upper = D * Ids^2 * Rdson and Pd lower = (1-D) * Ids^2 * Rdson where D is the duty cycle and Ids is the motor current and Rdson is the mosfet drain source resistance at its maximum die temperature that you want to operate it at. I use 150C Tjmax as another swag, ignoring the life expectancy stuff that others get concerned about. So now you can estimate the power that you main heat spreader needs to get rid of to the ambient environment. We assume that the commutation frequency is high enough that you can assume that the power is not localized to one bank of mosfets during normal operation when calculating power leaving the heatspreader for the controller. In the worst case of motor stall, one sector could be energized for a long period of time, and that what I normally used for design purposes.
Ptotal ~= ids^2 * Rdson * 2 + Ids*Vds*(Ton+Toff)*Fsw/6 for an Hbridge Synchronous rect single quadrant operation.
To figure out how many device you need you can now look at the worst case operating condition of 90% duty cycle on the bank of mosfets that are switching.
Pd max for bank will be ~= (Ids/n)^2*Rdson*0.9*n + Ids*Vds*(Ton+Toff)*Fsw/6. in this case is the number of parallel mosfets. For the single device in the bank, its Pd ~= (Ids/n)^2*Rdson + (Ids/n)*Vds*(Ton+Toff)*Fsw/6
Now you can figure out how many devices you need.
1) use Rdson for Tj of 150C from the datasheet of your favorite mosfet
2) choose the maximum allowed heatspreader temperature that the fets are mounted to.
Calculate maximum PD for the device. Pdss ~= (Tj-Ths)/(Rjc + Rt + Rhs) where Tj is 150C, Ths is the maximum temperature you want your heat sink to get to, Rjc is the thermal resistance listed on the mosfet datasheet, Rt is the thermal resistance of your insulating tape, Rhs is the thermal resistance to the back of your heat sink if you are bolting it to something else that has a lot of metal. So you either can rearrange the above equation to solve for n or you can calculate the pd for various number of parallel mosfets and choose a number that has power dissipation less than that calculated based on the thermal resistance. You will want to choose later how to get rid of the heat from the heat spreader base on your steadystate continuous power dissipation requirement. That could be a finned heatsink attached or a larger mass of metal.
The short term current requirement will be driven by how much heat capacity the main heat sink(spreader) will have, which means how much metal there is in terms of volume. The heat capacity is like a big thermal capacitor. Just like a capacitor and resistor, the temperature will rise in a similar fashion to the voltage on a capacitor being charged through a resistor. This is what gives you the ability to drive a higher current through your controller for a short time since the maximum power that you can dissipate in a mosfet is determined by the power being dissipated and the temperature differential between your mosfet die and the main heatspreader if the thermal resistance remains the same. If that main heat spreader takes some time to rise in temperature, then you can operate longer at higher currents. That is where those 1 minute ratings come in for the controller manufacturers. They have a certain amount of metal in their heat spreaders that allow them to run higher currents for a short time.
I think that I have laid out a fairly conservative method for calculating the number of mosfets you need. I have used it in the past and has work reasonably well to give you a starting point. Once you have real hardware you can figure out the real ratings by monitoring the various temperatures.
Fets do share reasonably good if they share a common electrically connection point and share a common heat spreader. They are positive temperature coefficient devices...ie the Rdson goes up with temperature which means that parallel mosfets will reach an current equilibrium point that is dependent on the die temperature and the circuit impedance between the devices. That is why you see ballast resistors sometimes on parallel IGBT's since a lot of them are negative coefficient devices where Vce decreases with temperature. I have seen newer IGBTs that are now positive coefficient on the Vce sat.
The package of a mosfet defines its maximum long term continuous current if you are able to keep the heatspreader at a temperature that keeps the die below its maximum allow operating temperature. Tj = Pd * Rjc + Ta. (not considering life expectancy since I am sure someone will comment about that) TO220 is typically around 75Amps max though I have seen higher with the newer generation TO220 packages. In terms of short term current ratings, you can look at the datasheet at the transient thermal impedance curves for 1 time pulses, and look at what the heat capacity of the heatsink that your fets are attached to. Realistically for long term average current, I think that you are limited by the thermal resistance path from the die to where you are trying to remove the heat too. So if that is from the die to ambient, for steady state, it will be less than the 75A or whatever the package is rated to. For the short term current, it is kinda hard to characterize since it is really dependent on the design. It is the short term current that you probably want to figure out for motor acceleration purposes. That is like the 1 min rating that the motor controller manufacturers sometimes give.
The mosfet spec sheets are not always that easy to use. The headline specs like typical Rdson, voltage , and current are given at the ideal temperature of 25C and are for the device only, not the device within your controller. So you can't say I want a 300A controller so I will use one TO220 mosfet rated at 300A to do it for reasons that I gave above. My opinion for figuring out how many mosfets you need is only a swag since you don't always know what your thermal design will be since it is dependent on the controller package and thermal insulating tape used to isolate the mosfets.
1) Identify the maximum current that you want to push through the motor
2) Identify the average continuous current that you want to run the motor at
3) Identify the maximum voltage that you want to operate at (you want some margin to prevent avalanching)
4) Take swag at how fast that you want to switch the device on and off at (Ton, Toff mosfet time) Chinese controllers are in the order of 3 usec or so.
5) Identify the switching frequency that you want to PWM at.
6) Identify the thermal insulating tape that you will use(if using a non isolated package)
7) Take a swag at how hot you want your heatspreader to rise to, that you mosfets will attach to.
7) Choose some candidate mosfets that you think are good
Ok, so now that you have that you can swag the power losses
Switching losses for a swag can be estimated as Psw = Ids * Vds * (ton+toff) * Fsw /6. This assumes that the devices are being hardswitched ie it is the device interrupting or sourcing the current. I am sure there are better estimates to calculate the switch losses, but this is the one I use. (EDIT Ids will be the motor current (phase for BLDC or Armature for DC) Vds is the battery voltage)
Conduction losses assuming a BLDC configuration with synchronous rectification (ie one low side bank of mosfets always turned on during the sector and a pair of mosfets banks switching in a half bridge configuration) will include the half bridge that is switching and the bank that is not switching. We will ignore the deadtime and losses associated with it for the purpose of figuring out the number of mosfets.
Pcond total ~= PD(half bridge) + PD(sync), PD half bridge can be estimated as the sum of the upper and lower Pd which are duty cycle dependent. So Pd upper = D * Ids^2 * Rdson and Pd lower = (1-D) * Ids^2 * Rdson where D is the duty cycle and Ids is the motor current and Rdson is the mosfet drain source resistance at its maximum die temperature that you want to operate it at. I use 150C Tjmax as another swag, ignoring the life expectancy stuff that others get concerned about. So now you can estimate the power that you main heat spreader needs to get rid of to the ambient environment. We assume that the commutation frequency is high enough that you can assume that the power is not localized to one bank of mosfets during normal operation when calculating power leaving the heatspreader for the controller. In the worst case of motor stall, one sector could be energized for a long period of time, and that what I normally used for design purposes.
Ptotal ~= ids^2 * Rdson * 2 + Ids*Vds*(Ton+Toff)*Fsw/6 for an Hbridge Synchronous rect single quadrant operation.
To figure out how many device you need you can now look at the worst case operating condition of 90% duty cycle on the bank of mosfets that are switching.
Pd max for bank will be ~= (Ids/n)^2*Rdson*0.9*n + Ids*Vds*(Ton+Toff)*Fsw/6.
in this case is the number of parallel mosfets. For the single device in the bank, its Pd ~= (Ids/n)^2*Rdson + (Ids/n)*Vds*(Ton+Toff)*Fsw/6Now you can figure out how many devices you need.
1) use Rdson for Tj of 150C from the datasheet of your favorite mosfet
2) choose the maximum allowed heatspreader temperature that the fets are mounted to.
Calculate maximum PD for the device. Pdss ~= (Tj-Ths)/(Rjc + Rt + Rhs) where Tj is 150C, Ths is the maximum temperature you want your heat sink to get to, Rjc is the thermal resistance listed on the mosfet datasheet, Rt is the thermal resistance of your insulating tape, Rhs is the thermal resistance to the back of your heat sink if you are bolting it to something else that has a lot of metal. So you either can rearrange the above equation to solve for n or you can calculate the pd for various number of parallel mosfets and choose a number that has power dissipation less than that calculated based on the thermal resistance. You will want to choose later how to get rid of the heat from the heat spreader base on your steadystate continuous power dissipation requirement. That could be a finned heatsink attached or a larger mass of metal.
The short term current requirement will be driven by how much heat capacity the main heat sink(spreader) will have, which means how much metal there is in terms of volume. The heat capacity is like a big thermal capacitor. Just like a capacitor and resistor, the temperature will rise in a similar fashion to the voltage on a capacitor being charged through a resistor. This is what gives you the ability to drive a higher current through your controller for a short time since the maximum power that you can dissipate in a mosfet is determined by the power being dissipated and the temperature differential between your mosfet die and the main heatspreader if the thermal resistance remains the same. If that main heat spreader takes some time to rise in temperature, then you can operate longer at higher currents. That is where those 1 minute ratings come in for the controller manufacturers. They have a certain amount of metal in their heat spreaders that allow them to run higher currents for a short time.
I think that I have laid out a fairly conservative method for calculating the number of mosfets you need. I have used it in the past and has work reasonably well to give you a starting point. Once you have real hardware you can figure out the real ratings by monitoring the various temperatures.
Gordo
100 kW
bigmoose said:
bigmoose;
I have had several conversations with fellows thinking of liquid cooling using everything from water to mercury. I have made the suggestion of using CCL4, Carbon TetraCloride as it was used in aircraft magnetos, many years ago. It is a non-conducting liquid. Is there something wrong with this idea? Even with pure water and inhibitors, I don't see how you can have a solution which does not become contaminated from the tubing and conduct electricity? The idea is to use metal tubing to conduct the phase current and have the cooling pass through it. There would be an insulating junction just before the windings so the liquid could go in one phase and out another 2. Or drill the solid side of the axle and have the coolant come out there.
strantor
100 W
Hi, thanks for that lengthy reply. I'm not yet schooled enough to address most of it; I am reading however, at a torturous rate. maybe in a couple of weeks I will be able to carry on a decent conversation on the topic. one thing I would like ask, regarding this:
and later on in section III (d) (vi)
I believe this paper was written in 1981, so things may have changed. Also, this paper seems very "theoretical", and I don't know whether to take the things is says literally, or if things in real life might be different. So, What say you? is it better to have separate heat sinks or common heat sink?
this seems to be disagreement with moose's last document he linked to (the way I interpret it). If you have a look in that doc, section III (b) (i):Electroglide said:
and later on in section III (d) (vi)
Seems to me, by using a common heat sink, you are "locking them in" to a common temperature, which undermines their intrinsic tendency to share the load.
I believe this paper was written in 1981, so things may have changed. Also, this paper seems very "theoretical", and I don't know whether to take the things is says literally, or if things in real life might be different. So, What say you? is it better to have separate heat sinks or common heat sink?
bigmoose
1 MW
texaspyro did a nice write up on how he matched gate threshold voltages for his capacitive discharge spot welder. It would be good to understand what texas did and why he did it. He knew what he was doing, and did what he did for a good reason!
strantor
100 W
I looked for it the first time you recommended it, but could not find it. Read somewhere that some posts were lost in a server transfer. Is it still around? I asked texaspyro for the link too...bigmoose said:
bigmoose
1 MW
liveforphysics
100 TW
Electraglide- I used to think it was that simple, and it was until I started trying to do multi-hundred amp switching. Then things just entirely stopped behaving as I expected. Layout began to matter more than anything else.
strantor
100 W
Sorry about that; I was looking for a build thread, something like "my capacitive discharge welder" by texaspyro.I completely skipped over those posts. Here is what he sent me:bigmoose said:
texaspyro said:
liveforphysics said:
My father was a scientist level evaluation engineer at Motorola semiconductor and although most of what he talked about was over my head, that's one concept I caught on to. There's the textbook equations that you start with on first design, but then you have to measure the results and develop a whole new set of design parameters to optimize the circuit functions.
http://www.3m.com/product/information/Fluorinert-Electronic-Liquid.htmlGordo said:
strantor
100 W
Why does every controller I've seen have a ton of little discrete FETs in it? Fussing around with matching the parameters of 40+ little FETs and designing the perfect PCB so they all share seems almost like a kludge compared to using something like this: http://ixapps.ixys.com/DataSheet/VMO650-01F.pdf. In my line of work I get to dissect Industrial Variable Frequency Drives for AC induction motors from time to time, big and small. They all (all the ones I have ever seen) use these modules, except they use IGBT modules, not MOSFET modules, and they are the full bridge modules. If a company went around trying to sell a VFD full of discrete MOSFETs I imagine they would be laughed out of the game; so why is it different here in the EV world? Is just the cost factor? The module I linked to costs a little over 200$. or is there something specific about the application that makes the discretes better? From a design standpoint, I think it would be much more reliable, as you wouldn't even need a PCB (for the power stage). Bus bars can bolt directly to it, capacitors can bolt to the bus bar, as well as diodes (http://www.mouser.com/ProductDetail/Ixys/MEO550-02DA/?qs=sGAEpiMZZMvcRsgoMFfeP7o6u7e9pzAG). Easy, simple, lowest inductance circuit possible, WAY more efficient cooling (module body is electrically isolated). My little circuit calculator says it would be about .5% less efficient than with a bunch of paralleled FETs but that's not that much. so, why not?
bigmoose
1 MW
Strantor the module you linked to is $200, you will need 6 of them... $1200 parts price for Fets alone, now consider that Retail is typically at least 4X parts price, you now have a $4800 controller and you haven't included a processor, caps or the metalwork. Now price out the same Rds on using IRFb4110's and see how the price matches up.
That is why our stuff has a plethora of $2 FET's!
That is why our stuff has a plethora of $2 FET's!
strantor
100 W
Ah, sorry, I forget where I have & where I haven't updated people. I'm only doing a brushed controller, not a brushless one. If I were doing a brushless controller they have one for that as well, with 6 switches inside (for a slightly bigger price tag)bigmoose said:
strantor
100 W
for a BLDC, something like this (http://search.digikey.com/scripts/DkSearch/dksus.dll?site=us&lang=en&keywords=fm600tu-2a&WT.mc_id=Semiconductor+Modules&WT.medium=cpc&WT.campaign=Semiconductor+Modules&WT.srch=1&WT.content=text&WT.source=google) can be had for 400$. That's the price of a whole controller, but from what I have seen on this forum, the high power BLDCs on the market made out of discretes are overrated and like to blow up. I think this would make for a much more robust design that could live up to its rated performance all day long (I think, I don't know, I've never used one)
strantor
100 W
I've never used a 4110 or one of these modules so I can't speak from experience, but from the looks of it I would think that 120A flowing through that little MOSFET leg would get it pretty hot pretty quick. I'm assuming (from the looks of the module, e.g. busbar material coming straight out of the body of the device & plenty of thermal mass) that the module would more likely be able to reliably deliver it's rated performance without a significant amount of derating. It's my guess that for the 4110's to match the performance of the module, you would need more than 2 or 3 or 4 in parallel. I doubt I'll ever find anybody in an internet chat room who has designed a system around one of these module devices to confirm/deny that though.
strantor
100 W
here's an application note on the topic (http://www.pwrx.com/pwrx/app/Hyb%20Mod%20Alt%20Parl%20Dis%20Dev.pdf)
Highlights:
Highlights:
liveforphysics
100 TW
strantor said:
4110's fail at around 45amps continuous when paired to an excellent heat sink. For burst loading, they can handle >200amps for very brief periods, but it's just a race to various failure modes.
strantor
100 W
Luke do you have any experience with these modules? I've seen you link to them a couple of times back in the day.liveforphysics said:
megacycle
100 kW
Anyone know if mosfet solid state relays could be used, they have a wide range and are cheap $10's and widely available
usually come in single dc switch or probable no good 1ph or 3 ph ac blocks.
They're usually used for switching loads on/off instead off contactors, i know they are derated for repetitive switching, but never heard them used for motor speed controllers
Can they be switched in khz apps
Tell me if i'm missing something :?
Otherwise be in agreement with strantor, i'm no electronics eng just a tech working in heavy current a.c. electrical.
Discreet components in relatively high energy apps have issues, love the idea of sticking test probes in those places too
boom. I would think larger and less power components on a seperate power board with connectors to control board.
usually come in single dc switch or probable no good 1ph or 3 ph ac blocks.
They're usually used for switching loads on/off instead off contactors, i know they are derated for repetitive switching, but never heard them used for motor speed controllers
Can they be switched in khz apps Tell me if i'm missing something :?
Otherwise be in agreement with strantor, i'm no electronics eng just a tech working in heavy current a.c. electrical.
Discreet components in relatively high energy apps have issues, love the idea of sticking test probes in those places too
boom. I would think larger and less power components on a seperate power board with connectors to control board.
strantor
100 W
SSRs are usually beefy triacs. If you tried to use them with DC they would pulse once, then latch, and you would have to disconnect the battery. Unless there is a SSR for switching DC I don't know about, which there probably is, I couldn't comment.
strantor
100 W
Oh cool, I just read the threads (big fet drive challenge, 12kw motor, resource page for MEV/HEV applications, etc) where you guys are already using these modules.Arlo1 said:
Has anbody got a working controller yet? I've seen lots of pictures of the modules.
Looks like I will be joining the big fet drive challenge (or maybe just my own weak sauce side challenge); I just ordered mine. http://ixdev.ixys.com/DataSheet/VMO1200-01F.pdf
100V 1200A single switch
megacycle
100 kW
Thanks strantor, there are cheap mosfet ssr blocks 150A/50V but yeh duty cycles at these levels would kill thm expect.strantor said:
That beafy IXY
, nice want some too where are they bought from Similar threads
