Epic 7kw+ 12 fet controller...or there abouts :)

The mosfets arrived today.

CSD19536%20TI%20mosfets.jpg


My bench is grounded. There's a long section of 2"x2" angle aluminum across the front edge of the bench. It has a 16 awg wire going to earth ground. My arms rest on it and I touch it often. Rolling up to my bench and coming in contact with that aluminum angle is easy and natural and hard to avoid doing. This grounds out any static I may build up on me and keeps the bench top from building up any static. Never the less, I lay the mosfets on an anti-static bag that is in contact with the aluminum strip. This is done because mosfets can be damaged by static discharge.

Grounding.jpg


People have scoffed before and claimed there's no way a $30 Chinese component tester can tell you squat about a part and do it reliably and accurately. I disagree. I can do these exact tests with my expensive DMM and a bench power supply and get very close results and take 10X longer to do it. I can pull out my milliOhm meter and check Rds. I have the expensive test gear to test with! Why would I test the mosfets the hard way when I already know the results from the component tester are going to be pretty close to the same? I'm NOT...that's just stupid to bother spending an hour to set up my test gear and another hour testing mosfets. My $30 component tester from China can tell me the exact same thing in seconds! Scoff, complain, whine all you want. I already know the $30 parts tester is doing the right thing and where it is off and by how much. You are biased if you think this inexpensive tester can't do the job. IT's NOT true and you don't actually possess the facts! You are stating your opinions. I'm stating actual facts from my own tests. I actually tested to be sure. What did you do? Oh that's right, postured, whined and complained, but didn't actually do any test work to prove me right or wrong. You know who you are!

OK...enough...getting off the soap box! LOL!

I'm looking for 3 things on my component tester.
1. Does the mosfet work? Does the component tester even recognize it as a part and if so what kind? In this case...N-E-MOS or N-channel, enhanced mosfet.

2. Turn on voltage or Vt. This is the voltage needed at the gate to make the mosfet fully turn on. I want Vt to be the same across all mosfets. The spec sheet says Vgs(th) is typically 2.5v and max of 3.2v. Looking at the graph of actual turn on voltages, I see that it's really at about 3.2 to 3.3 volts. Looks like the $30 tester hit the mark! All mosfets tested at 3.2 or 3.3 volts except one at 3.4 volts. I think I can trust this tester so far.

3. Uf is the voltage drop across source to drain. I want Uf to be the same across all mosfets. Lower is better. Uf is based on an internal precision voltage source in the component tester. This voltage source is placed across Source to Drain and a series resistor. The voltage on the mosfet is then measured. Higher Uf indicates that Rds is higher. Lower Uf indicates lower Rds. There's another graph that plots the gate voltage and voltage across Source to Drain. The 25C (cold mosfet) line at 3.2-3.3 volts on the gate shows about 700mV across Source to Drain. I read 571mV on my component tester. I already know it's internal precision voltage source is a little low. This skews my results down by 60mV on the component tester. As a result, It does read Uf a little low. So then 571mV + 60mV = 631mV. The graph says 700mV, I say 631mV...39mV difference. I'm NOT worried! The $30 Chinese parts tester does what I could do in 10X more time with my expensive test gear in a couple of seconds!

Mosfet%20test.jpg


When done with the first test run, I had 3 piles of mosfets. I'm just looking that they all work and what Vt is. I want mosfets that all turn on at the same voltage. In this case, that was 3.2v, 3.3v and a single mosfet at 3.4v. This is pretty good! I've bought loads of AOT290's and had far worse results on this exact parts tester.

First%20mosfet%20sort.jpg


This is after the second test run. In the tube and on top of it are all the 3.3v mosfets. This time I'm looking at Uf. I now know all the turn on voltages or Vt. Do they also have the same voltage drop across Source to Drain? Again...doing this same test on AOT290's or Chinese mosfets, I got lots of variation. In this test all the 3.3v mosfets also had the same Uf of 571mV. The outlier (center most single part) at 3.4v Vt also has a Uf of 571mV. The mosfets with a Vt of 3.2 volts has 2 of them that had lower Uf than the rest. That's the left 2 single mosfets. One is at 569mV and the other at 570mV. The lower right pile is all the rest of the 3.2v Vt mosfets that have a Uf of 571mV. That means 28 of the 30 mosfets have the same Rds. Out of 30 mosfets 3 were different than the 2 groupings. Since the motor controller will supply more than 3.3v to turn on the mosfets, I'm not overly worried about the small Vt differences. However, the 12 mosfets that go in this controller will have the same Vt. I will use 12 mosfets with Identical Uf in the controller. Fortunately there are plenty that can work. Since there are 12 that have a Vt of 3.2v and Uf of 571mV it's a good grouping to use in the controller. I could also use 12 from the 3.3v grouping, but there are 15 of those. The three mosfets that don't match up will get used for something less critical...like a BMS or a mosfet switch.

Second%20mosfet%20sort.jpg


Conclusions:
1. Chinese mosfets are just awful. I've tested many on my parts tester and I see Vt and Uf all over the place for same model mosfets from the same source. Commonly the test results and the spec sheet are way different.

2. I don't know if AOT went through a bad spell about 2 years to 18 months ago and their quality control was horrible. I haven't bought any AOT290's since then. However I saw Vt and Uf vary a lot. I bought 100 AOT290's from digikey and Arrow over 3 orders and at different times over about a year. I forget the exact quantities now in each order, but something like 45, 30 and 25. I tested all of them on my component tester shown above. The Vt And Uf readings were pretty inconsistent across the 100 mosfets. I ended up with 8 piles. To get 18 for a single controller meant taking the single largest pile and then part of similar piles to populate a single controller. 12 more went into the 12 fet currently in the Currie. Those were picked first and they all matched each other, but not the spec sheet. A few of those 100 mosfets actually matched the spec sheet. Most didn't.

3. These TI mosfets are far more consistent. .2v difference in Vt and .002v difference in Uf and 27 of the 30 fell into 2 groups...that's pretty darn good! And more importantly they are all pretty darn close to the graphs in the spec sheet.
 
ElectricGod said:
I am going for 7kw+++ in a 12 fet running TO-220 mosfets. I'm looking to get pretty darn close to the leg limits of the TO-220 package.

Regarding the leg limit of TO220 package i believe it can be higher as the 75A mentioned before.
In the datasheet of the TI mosfets for instance it seems to be 120A and on AOT290 also higher than 75A.

That would explain why the Nucular 12F controller can do 200-250A phase current and 24F 400-500A :)
 
madin88 said:
ElectricGod said:
I am going for 7kw+++ in a 12 fet running TO-220 mosfets. I'm looking to get pretty darn close to the leg limits of the TO-220 package.

Regarding the leg limit of TO220 package i believe it can be higher as the 75A mentioned before.
In the datasheet of the TI mosfets for instance it seems to be 120A and on AOT290 also higher than 75A.

That would explain why the Nucular 12F controller can do 200-250A phase current and 24F 400-500A :)

Even 500 phase amps isn't really hitting the leg limit. That's about 63 amps per mosfet in a 24 fet...if I'm thinking about this right. That 500 amps is passing through 2 phases and 8 mosfets.

Die limits are typically what you see in a spec sheet. IE: The mosfet die can handle X amps which is usually much higher than the TO-220 leg limit of 75 amps. The point where the mosfet legs melt. I was looking for an article I read a while back about the cross section of the TO-220 leg, heat and how 75 amps was derived. As expected..can't find it. I'm pretty sure it was written by an Infineon engineer.

You can help the leg limits considerably by making them very short, adding solder to thicken the legs and having a good heat path away from them to a large hunk of copper. That's what I was attempting to do on the mosfets already in the controller. Since I'm swapping them out now with 12 known highly similar mosfets, I'll pre-tin the mosfet legs so they flow solder well once installed in the board. Of course, this is only effective to the edge of the TO-220 package, but it's better than not doing it. Do this, but better by getting solder flowed up the entire exposed leg on both sides.

Mosfet%20legs%202.jpg
 
I haven't seen CSD19536 before. It has great specs. Significantly better than the trusty old IRFB4110.

You could always use the tab for the drain connection instead of the leg to at least eliminate that as a weak spot. The high side drains will be fed by the V+ rail, which could be a big aluminum or copper bar that doubles as a heat spreader. Making the legs as short as possible obviously will help too. One possible problem with that is thermal expansion and fatigue fractures. Things will expand a little every time they get hot and if there is enough mechanical stress from heating, you can get fatigue fractures after a bunch of heating/cooling cycles. I've seen it happen a bunch in high powered switching power supplies.
 
fechter said:
I haven't seen CSD19536 before. It has great specs. Significantly better than the trusty old IRFB4110.

You could always use the tab for the drain connection instead of the leg to at least eliminate that as a weak spot. The high side drains will be fed by the V+ rail, which could be a big aluminum or copper bar that doubles as a heat spreader. Making the legs as short as possible obviously will help too. One possible problem with that is thermal expansion and fatigue fractures. Things will expand a little every time they get hot and if there is enough mechanical stress from heating, you can get fatigue fractures after a bunch of heating/cooling cycles. I've seen it happen a bunch in high powered switching power supplies.

I use 4110's in stuff like BMS's becasue they are cheap and just sit there on all the time. However they are not as good at these TI mosfets...where performance will really matter.

Kelly does that. They use a ceramic backer that has the traces on it to solder all the mosfets onto. They can get at the tabs electrically and not hamper heat flow away from the mosfets.

The problem is the design of the controller. The PV 12 fet is not exactly set up for something like adding a ceramic backer. Soldering onto the actual tabs with a piece of copper would make that "leg" even longer. I'm not sure it would be better. Also, what about the Source leg? I can't heat sink it at all. At best I can give it lots of solder pad to dump any heat into.

I mentioned in the Mobipus thread that potting was a bad idea. Thermal expansion will occur, parts do fail and potting makes repairs impossible. With lots of solder on a connection, the likelihood of failure is reduced. Like you, I too have seen broken solder connections hundreds of times on boards due to thermal expansion and contraction. In the Mobipus thread, they looked at me like I had 3 heads. Potting was the best thing since the invention of electricity in their minds. OK...ignore the guy that has been repairing electronics since the 70's. What could I possibly know.

Since I have the controller apart, I'll be changing the phase and power wires slightly. I realized that my wire gauge adapter idea was dumb. What I really should do is tin the end of the 8 awg wire. Then thin it down enough to fit in the 10 awg hole. Add some heat shrink around the end so no loose strands can be exposed and then solder the thinned down end into the board. This will get the wires lower in the board and eliminate the 8awg to 10 awg "adapters". Right now with those "adapters" on them means the wires press onto the roof of the shell. I have no idea if this will become a problem eventually or not, but I can completely eliminate them and any losses they may introduce. This will go away.

8%20AWG%20wires%201.jpg
 
ElectricGod said:
OK...ignore the guy that has been repairing electronics since the 70's. What could I possibly know.

You can learn a lot from other peoples' design mistakes. When you see the same kind of failure over and over and over, you eventually get it.

I agree you shouldn't totally redesign the PV layout, but if one was going to design from scratch, there are some other possibilities.

I've had the wire-too-big-for-the-hole problem before. I just snipped off enough strands so the rest could fit and soldered it. Solder holds the strands together.

Below is an old Crystalyte controller. The FETs bolt directly to a piece of aluminum heat spreader. The little spreaders have insulation under them and bolt to a much larger spreader that holds all 12. The idea is to minimize the thermal resistance from the die to the heat sink. The little spreaders have a lot more surface area for the insulators so will have a better thermal response than using insulators directly under the FET tabs.

Ceramic insulators might be even better. Diamond would be the best, but not very cost effective. Pyrolytic graphite may be almost as good.

IMG_0795.JPG
 
fechter said:
ElectricGod said:
OK...ignore the guy that has been repairing electronics since the 70's. What could I possibly know.

You can learn a lot from other peoples' design mistakes. When you see the same kind of failure over and over and over, you eventually get it.

I agree you shouldn't totally redesign the PV layout, but if one was going to design from scratch, there are some other possibilities.

I've had the wire-too-big-for-the-hole problem before. I just snipped off enough strands so the rest could fit and soldered it. Solder holds the strands together.

Below is an old Crystalyte controller. The FETs bolt directly to a piece of aluminum heat spreader. The little spreaders have insulation under them and bolt to a much larger spreader that holds all 12. The idea is to minimize the thermal resistance from the die to the heat sink. The little spreaders have a lot more surface area for the insulators so will have a better thermal response than using insulators directly under the FET tabs.

Ceramic insulators might be even better. Diamond would be the best, but not very cost effective. Pyrolytic graphite may be almost as good.

IMG_0795.JPG

I thought this was unique. This image is from my icharger 4010DUO. It uses clear glass insulators. They are not white ceramic. I have never seen that before. It ought to be better than kapton or some other non-conductive option. Probably as good as or better than using mica. Too bad the graphite stuff conducts current. It is a good heat conductor. I have the kapton insulator in the controller now. Might have to look at ceramic or glass insulators.

iCharger%204010%20duo%20heat%20sink%20bottom%201.jpg


Reducing wire size...what you said...that's exactly what I'll be doing. Something I discovered is that if you tin the wire and then shake out the hot solder, that once the tinned end has cooled that you can squeeze the strands together pretty tightly and they will stay compressed together. I doubt I'll get them squeezed together enough to make 8 awg worth of wire fit in a 10 awg hole, but I bet I'll get close and need to thin it down only just a little bit. I have a 6 sided berrol crimper. It ought to do the trick.

I like those extra large insulators. OF course no one does that anymore. I guess if I really wanted to be serious about it, I'd replace the factory heat spreader with a larger section of 1/4" thick aluminum that filled the entire wall space. It would not be hard to make this bigger and add a bit more thermal mass at the same time.

Controller%20in%20shell%202.jpg


I'm also thinking about "fixing" this too. The bottoms of the CPU coolers are not particularly flat and it takes 3 layers of thermal tape to fill the gaps. I'm losing a fair bit of thermal conductivity here. I bought some .5mm thick copper sheet. I don't think it would be very hard to solder a square of copper onto here to make a flat surface. I bet if I put the CPU coolers in my reflow fryng pan with a copper square under each one that I can get these less than optimal surfaces very flat and fully covered in solder between them and the square. I can put flux on here and get solder to generally flow with my Hakko on MAX. I'll do the same for the copper squares. I can then mount the aluminum adapter plate to hold the squares flat to the bottoms of the CPU coolers. Stick the whole thing in my reflow frying pan until the solder melts together. Metal to metal will make a much better heat path as will the flat copper squares.

CPU%20heat%20sink%20-%20bottom.png
 
I found these on ebay...ceramic insulators. I'll buy a bunch of them. There's no rush in getting this controller done and this ought to be better than the kapton currently in there.

https://www.ebay.com/itm/10pcs-TO-220-Alumina-Ceramic-Transistor-Triac-Thyristor-Insulator-Protection/263689683162?hash=item3d652140da:m:mTdFANsbpHqNco4boa3NGIA:rk:3:pf:0

Just thought of this...the mosfets might be too close together to use those ceramic insulators. They are 14mm wide and TO-220 is 10mm wide. I guess I'll find out soon enough! Ceramic is hard stuff and brittle, but I might find myself at the diamond wheel reducing the width of the insulators.

Kapton%20insulator%20closeup_zpsgkz8ggim.jpg
 
ElectricGod said:
Even 500 phase amps isn't really hitting the leg limit. That's about 63 amps per mosfet in a 24 fet...if I'm thinking about this right. That 500 amps is passing through 2 phases and 8 mosfets.

There are 8 FET's per phase, but they split into 4 high side and 4 low side, so at 500A we are talking about 125A per FET.
At least it will be like this if we talk about trapezoidal 6-step commutation where only two of the three phases are activ.
With FOC all three phases do some work so the current per FET could be less.
 
madin88 said:
ElectricGod said:
Even 500 phase amps isn't really hitting the leg limit. That's about 63 amps per mosfet in a 24 fet...if I'm thinking about this right. That 500 amps is passing through 2 phases and 8 mosfets.

There are 8 FET's per phase, but they split into 4 high side and 4 low side, so at 500A we are talking about 125A per FET.
At least it will be like this if we talk about trapezoidal 6-step commutation where only two of the three phases are activ.
With FOC all three phases do some work so the current per FET could be less.

I see my mistake...I was thinking about H-bridges and had those mosfets in parallel instead of series in my mind.
You are right...the leg limit is 75 amps and yet they see 125 amps at 500 phase amps in a 24 fet.

I think maybe the fact that a motor controller is a class D amplifier plays into this. The mosfets turn on and off rapidly while creating a current wave form that is sinusoidal. Maybe the PWM pulses are short enough to keep the legs from burning out and that makes this possible? At a given moment in time for a few milliseconds there could be 125 amps on the leg, but then the leg "rests" for a couple of milliseconds before it sees another 125 amps. The over all effect would be over 4 or 5mS a leg sees an average of 75 amps max. Does that make sense or am I delusional?

I'll post pics later. The new mosfets are installed and they have much more solder on the legs than before. I tinned the mosfet legs in advance of installing them and that made adding solder to them after they were soldered in very easy. I'll assemble the controller part way so that I can spin it up, but then I'll be taking off the heat spreader again to use ceramic insulators. In the old mosfets, I found a single one had died. Usually they die in pairs...a high side and low side that were on when the problem occurred.

I bought some nasty stuff for removing glue and stickers. It worked really well at removing old flux too. The bottom of the board is very clean now. I was concerned that a solder blob might be hidden between large traces and create a short. An old tooth brush dipped in this gum remover stuff did a brilliant job of taking off the solder flux. It evaporates off quickly and should be used with ventilation. I'm sure it's BAD stuff. The fumes are pretty fierce. Years ago I could get spray cans with some kind of solvent in it that was for removing flux from boards. The fumes remind me of that stuff.
 
ElectricGod said:
I think maybe the fact that a motor controller is a class D amplifier plays into this. The mosfets turn on and off rapidly while creating a current wave form that is sinusoidal. Maybe the PWM pulses are short enough to keep the legs from burning out and that makes this possible? At a given moment in time for a few milliseconds there could be 125 amps on the leg, but then the leg "rests" for a couple of milliseconds before it sees another 125 amps. The over all effect would be over 4 or 5mS a leg sees an average of 75 amps max. Does that make sense or am I delusional?

Yes, that makes sense.

The leg limit is based only on heating. The current averaged over about 1 second would determine the limit. The peak current could be up to the silicon limit for shorter durations.

The ceramic insulators are interesting. No issues with dielectric strength or temperature rating. But the Kapton might be just as good or even better since it is a lot thinner. I'm sure there's a way to do the math on it. Thickness x thermal conductivity.
 
fechter said:
ElectricGod said:
I think maybe the fact that a motor controller is a class D amplifier plays into this. The mosfets turn on and off rapidly while creating a current wave form that is sinusoidal. Maybe the PWM pulses are short enough to keep the legs from burning out and that makes this possible? At a given moment in time for a few milliseconds there could be 125 amps on the leg, but then the leg "rests" for a couple of milliseconds before it sees another 125 amps. The over all effect would be over 4 or 5mS a leg sees an average of 75 amps max. Does that make sense or am I delusional?

Yes, that makes sense.

The leg limit is based only on heating. The current averaged over about 1 second would determine the limit. The peak current could be up to the silicon limit for shorter durations.

The ceramic insulators are interesting. No issues with dielectric strength or temperature rating. But the Kapton might be just as good or even better since it is a lot thinner. I'm sure there's a way to do the math on it. Thickness x thermal conductivity.

What I've read is that ceramic is better than kapton as a thermal conductor. Kapton is a better electrical insulator, but conducts heat less. Kapton wins for resiliency to damage. Graphite while a superb heat conductor is not an electrical insulator. Ceramic is a bit brittle so make sure whatever is mounted to it is flat. I bought .65mm thick ceramic insulators instead of 1mm thick. I wonder if I'll regret that choice. With kapton or ceramic you want to use thermal paste to fill the tiny voids.
 
I'm sure the ceramic is a better heat conductor than Kapton, but the thickness makes a huge difference. It's just like resistance in a wire. Shorter wire has less resistance. At some point a short skinny wire will have the same resistance as a longer, fatter wire. Do you know how thick the Kapton is?

If somebody has tested both side-by-side with everything else equal, then I'll totally buy it.

Just dreaming... if there was a way to make a heat pipe that was electrically insulated, you could have one end directly on the FET tab. Seems possible (but not cheap). I've seen xenon lamps that use a ceramic insulator that is somehow brazed to copper end pieces and it runs at very high pressure.
 
fechter said:
I'm sure the ceramic is a better heat conductor than Kapton, but the thickness makes a huge difference. It's just like resistance in a wire. Shorter wire has less resistance. At some point a short skinny wire will have the same resistance as a longer, fatter wire. Do you know how thick the Kapton is?

If somebody has tested both side-by-side with everything else equal, then I'll totally buy it.

Just dreaming... if there was a way to make a heat pipe that was electrically insulated, you could have one end directly on the FET tab. Seems possible (but not cheap). I've seen xenon lamps that use a ceramic insulator that is somehow brazed to copper end pieces and it runs at very high pressure.

The kapton insulator is .2mm or so. The ceramic is easily 3X thicker. You may be right...well there will be a temp sensor mounted to one of the center most mosfets. I'll know soon enough if this is a good or bad idea. Actually, I have a second temp sensor...might as well mount that to the heat spreader and see what kind of temp differentials I see.

I read about thermal transfer in kapton and in ceramic. The ceramic was better, but it had to be a lot thicker too and was still better. The dual temp sensors should be telling. The mosfet should be just a few degrees warmer than the heat spreader.

Hmmm...copper heat spreader...who cares about heat pipes! Dream on...I'm right there with you.

I found these on ebay. The existing heat spreader is 10mm thick. I have no idea how tall I can go inside the shell and still keep the added internal aluminum one.

10mm x 35mm
https://www.ebay.com/itm/1pcs-99-Copper-T2-Cu-Metal-Flat-Bar-Plate-10mm-x-35mm-x-200mm-EC-i-GY/181456975934?hash=item2a3fadb83e:g:FOAAAOSwEK9Ttjy8:rk:31:pf:0

10mm x 60mm
https://www.ebay.com/itm/1pcs-99-Copper-T2-Cu-Metal-Flat-Bar-Plate-10mm-x-60mm-x-200mm-EC-J-GY/181456976405?hash=item2a3fadba15:g:FOAAAOSwEK9Ttjy8:rk:12:pf:0

I bet glass or ceramic will have similar heat conductivity. All I need is a thin layer of gorilla glass which should be resilient. There's lots of after market smart phone glass stick-on coatings that are supposed to be tougher than the phones own glass. Sounds like it could work. =)

This ought to work. I can cut it with a diamond wheel, not score, cut all the way through to the width and length I need. I have nooo idea how to drill it for the 12 screws. I'm thinking, stick it down to a piece of aluminum for all the cutting and then use heat later to make the glue come loose.

https://www.ebay.com/itm/Tablet-Tempered-Glass-Screen-Protector-For-Amazon-Kindle-fire-7-HD-8-HD-10/273584098211?hash=item3fb2e20ba3:m:mkLpd0mub8yqXer62WbQRdg:rk:2:pf:0
 
Measured the factory heat spreader...8mm x 20mm.

Good thing I didn't get that 10mm thick copper yet! There is zero room for more thickness inside the shell. I have 31mm of height inside.

This is really what I need. 4mm of waste...not too horrible.

https://www.ebay.com/itm/1pcs-99-Copper-T2-Cu-Metal-Flat-Bar-Plate-8mm-x-35mm-x-200mm-EC-G-GY/181456974700?hash=item2a3fadb36c:g:FOAAAOSwEK9Ttjy8:rk:18:pf:0
 
ElectricGod said:
I think maybe the fact that a motor controller is a class D amplifier plays into this. The mosfets turn on and off rapidly while creating a current wave form that is sinusoidal. Maybe the PWM pulses are short enough to keep the legs from burning out and that makes this possible? At a given moment in time for a few milliseconds there could be 125 amps on the leg, but then the leg "rests" for a couple of milliseconds before it sees another 125 amps. The over all effect would be over 4 or 5mS a leg sees an average of 75 amps max. Does that make sense or am I delusional?

It would be like this at partial throttle or if PWM is less than 100%.
At full throttle or 100% PWM this current will be true continuous or RMS, and peak current will be always higher due to the current ripple which depends on the inductance of the motor.

I don't think such high currents are a problem as long as there is no poor heat path (good controllers anyway have thermal roll back), and i cannot see anything in the datasheets of TI or AOT290 FET's which would say that 75A would be the maximum continuous current. Or did i miss anything?

btw: on latest Adaptto controllers i noticed they were using mentioned ceramic pads instead of those grey ones (silicon?) they had before. The thickness did look like something between 0,5 and 1mm.
That makes me think you don't do anything wrong if you use them, but i would go with the thinnest ones i could find.
 
Here's the technical data for Kapton tape:
http://www.dupont.com/products-and-...kapton-polyimide-film/products/kapton-mt.html

They make a special "MT" version
DuPont™ Kapton® MT polyimide film is a homogeneous film possessing 3x the thermal conductivity and cut through strength of standard Kapton® HN.

OK, so the super duper MT thermal version has a thermal conductivity of 0.46 W/m.k
The ceramic insulators were spec'd at 20-25 W/m.k, which is a lot higher.

For example, 2mil Kapton tape would be about 0.05mm thick. The ceramic insulators are 0.65mm thick.
So the ceramic is 13 times thicker but has at least 43 times more thermal conductivity for a given thickness.
So ceramic wins by a factor of about 3.3.

Just for reference, pure copper has a thermal conductivity of around 380 W/m.k
Pyrolytic graphite sheet in the preferred direction: 1,500 W/m.k.
Diamond: 2,200 W/m.k
 
fechter said:
Here's the technical data for Kapton tape:
http://www.dupont.com/products-and-...kapton-polyimide-film/products/kapton-mt.html

They make a special "MT" version
DuPont™ Kapton® MT polyimide film is a homogeneous film possessing 3x the thermal conductivity and cut through strength of standard Kapton® HN.

OK, so the super duper MT thermal version has a thermal conductivity of 0.46 W/m.k
The ceramic insulators were spec'd at 20-25 W/m.k, which is a lot higher.

For example, 2mil Kapton tape would be about 0.05mm thick. The ceramic insulators are 0.65mm thick.
So the ceramic is 13 times thicker but has at least 43 times more thermal conductivity for a given thickness.
So ceramic wins by a factor of about 3.3.

Just for reference, pure copper has a thermal conductivity of around 380 W/m.k
Pyrolytic graphite sheet in the preferred direction: 1,500 W/m.k.
Diamond: 2,200 W/m.k

Thank you for posting that...looks like ceramic is the right choice even over MT kapton.

by chance did you find the thermal conductivity of aluminum? I'll be going from an aluminum heat spreader to a copper one and it will be larger. That ought to be a significant improvement.

Since I'll be using .65mm ceramic insulators, I'll need to use 7mm copper. 8mm copper assumes that the spacing behind the mosfets doesn't change and there is not even .5mm gap between the board and the far wall of the shell. What you just said about the thickness of the kapton insulator made me realize this error in my planning.
 
madin88 said:
I don't think such high currents are a problem as long as there is no poor heat path (good controllers anyway have thermal roll back), and i cannot see anything in the datasheets of TI or AOT290 FET's which would say that 75A would be the maximum continuous current. Or did i miss anything?

btw: on latest Adaptto controllers i noticed they were using mentioned ceramic pads instead of those grey ones (silicon?) they had before. The thickness did look like something between 0,5 and 1mm.
That makes me think you don't do anything wrong if you use them, but i would go with the thinnest ones i could find.

The insulators are .65mm. I could have gotten 1mm, but thought thinner was better. Nice to know I'm not alone in doing this...not that I think much of adaptto.

Heat path...yup...doing everything I can think of to make that as optimal as possible.
1. inner angle aluminum
2. outer angle aluminum
3. CPU coolers with modded bases.
4. larger copper heat spreader
5. ceramic insulators
6. reinforced TO-220 legs
7. reinforced traces on the board

Did I miss anything?
Is there anything else I can do?

I'm wide open here to changes and improvements.
 
copper purchased...

This controller will never see the light of day at this rate!

Price is not the object here...let's be honest...this might be the most expensive 12 fet there is.

It's getting the bloody raw edge of POWER out of 12 mosfets.

Thanks guys for your ideas and conversation...I'm having fun!
 
ElectricGod said:
Did I miss anything?
Is there anything else I can do?

You could use dry ice or other crazy methods to cool it far below ambient temperature :D
No, seriously i think in terms of cooling it is more than enough, but as mentioned i think you cannot push those top notch FET's to the limit if you leave the power stage at stock. It needs better electrolyte and additional ceramic caps.
 
Here is a great article about thermal conductivity.

https://www.electronics-cooling.com/2016/08/design-considerations-when-using-heat-pipes/#

As the article notes, heat pipes are a couple magnitudes better than solids for heat conduction.

It would seem that getting a couple heat pipes and integrating them into the 8mm copper plate that you will mount to
the fets may be ideal. As has been observed, the shortest possible heat path is optimum.
 
Dear god! You are very serious about this ! As normal I'm lost a little with the detail of your build but I fully expect you to have an amazing controller! I'm keen to know how how well this performs on your build and what it can successfully drive (motor wise).

I think you should set a goal .......

CAN AN ELETRIC STAND UP SCOOTER HIT 100mph. At this rate I think so! You just need someone willing to try it out!

You should set up shop and build custom eletric scooters! The EV Elon Musk
 
madin88 said:
ElectricGod said:
Did I miss anything?
Is there anything else I can do?

You could use dry ice or other crazy methods to cool it far below ambient temperature :D
No, seriously i think in terms of cooling it is more than enough, but as mentioned i think you cannot push those top notch FET's to the limit if you leave the power stage at stock. It needs better electrolyte and additional ceramic caps.

What do you recommend for the ceramics? I think you mean these specifically right? They are 104's or .1uF right now. There's some SMT caps in each gate control circuit too.

Cap%20close-up%201.jpg

Cap%20close-up%202.jpg




I'll be adding an external large cap bank...that's the best I can do for power filtering. The laid down ones can go a little bigger, but not much.

Cap%20close-up%203.jpg
 
Back
Top