Building the Best Controller

[Alan B]

Looks like $3.50 to $5.70 in 100 qty at Digikey, for example. The savings is only 6*$2.20.

One approach is to design for the TO247 and just bend the leads for TO220. Or put the pads in for both, but only if that doesn't compromise something else. The drivers should be selected to handle anything in the range.

The real benefit to TO220 is in a six FET design that is compact. That would require a separate board layout, and with several boards this design is not likely to be super compact anyway.

There are lots of TO220 FET boxes available out there. One goal here is to differentiate this one from the rest.

 

[Kingfish]

That sounds like a great arrangement; thanks for that my friend

Luke, the 4468 I think is fine for 100V people. Would you think the IXFK360N15T2 though is a bit overkill for 150V? Wouldn't that take a different board for all that currrent? Like bus-bars or something? In my dreams - I wish I could build a bike that would use it!

Alan_B, I am beginning to think we should just drop the TO220 then; lots of conversation to move to TO247 - I mean... it is the leading package.

Jazzed! KF

 

[liveforphysics]

Alan_B, I am beginning to think we should just drop the TO220 then; lots of conversation to move to TO247 - I mean... it is the leading package.

Jazzed! KF

Yes! Dropping TO-220 is FTMFW. If people want shit controllers designed around getting something that functions without spending a penny more, they should just buy any of the available offerings that all ready mastered this approach.

TO-247, TO-264 are like the bare minimums to make a controller worth messing with.

And for the folks looking for esoteric designs, like multi-hundred amp controllers the size of a deck of cards, the "new" IXYS package absolutely destroys everything. It's also an isolated package! So you don't have to mess with stupid silicon pads or kapton tape to insulate it form the sink, just bolt them on and go.

This is the 55V version, which would be an absolute gift from God for all the guys running 12s RC motor setups:
http://ixdev.ixys.com/DataSheet/DS100281(MMIX1T550N055T2).pdf

This is the 75v version, which is also a gift from heaven. Just choose a lower wind-count on your hubmotors for the frock-crew. They don't offer a 100v version.
http://ixdev.ixys.com/DataSheet/DS100269(MMIX1F520N075T2).pdf

 

[amberwolf]

Based on the little I know of MOSFETs, their characteristics, heat dissipation, etc., I'd also agree that TO220 is kinda pointless even in the base unit, simply for the fact that there *are* decent enough offerings that already use those. If this really is to become the best controller possible, then it should start with something better than the lowest-end (and still usable) FET case on the power stage.

 

[Kingfish]

Gawddamn what monstors! Can you imagine 500A/phase @ 75V! Peel the rubber off the rim!

So I guess I'm adding another stator to my AF motor and switching to building a motorcycle, right? :wink:

OK - so now I have to ask:
If we go with these higher-current packages then perhaps we need to rethink the number of FETs. I mean - does it make sense to have 12-FET design for Version 1? Certainly the Amp-limit conversation goes out the door; we pick one of these and design around it - boom: done.

Keeping an open mind: I'm not signing off; just entertaining all possilities.

Wicked! KF

 

[Alan B]

That sounds like a great arrangement; thanks for that my friend
...
Alan_B, I am beginning to think we should just drop the TO220 then; lots of conversation to move to TO247 - I mean... it is the leading package.

Jazzed! KF

I would leave the final TO-220 accomodation decision to the PC layout detail phase (put it as a low priority option), if it has no negative repercussions, do it. Otherwise let'em bend leads. Optimize for the favorite package. My $0.02.

One other thought on the PCB copper weight - do the pc layout with as much trace width as practical (and parallel traces on both sides if possible)(within the other design considerations), and route traces so wires can conveniently be soldered to them (make them straight and wide enough for that), in case the heavy copper is not practical or not enough. Go as far as practical and allow for improvement.

On those 1" square packages, how does one get them all to connect to the heatsink? Seems like they would not be easy to get exactly planar, sitting on those J leads. They might be best for separate heatsinks per device. Seems like a special project. I'm set up to solder those, how many folks are?

 

[liveforphysics]

Gawddamn what monstors! Can you imagine 500A/phase @ 75V! Peel the rubber off the rim!

So I guess I'm adding another stator to my AF motor and switching to building a motorcycle, right? :wink:

OK - so now I have to ask:
If we go with these higher-current packages then perhaps we need to rethink the number of FETs. I mean - does it make sense to have 12-FET design for Version 1? Certainly the Amp-limit conversation goes out the door; we pick one of these and design around it - boom: done.

Keeping an open mind: I'm not signing off; just entertaining all possilities.

Wicked! KF

The whole talk of using TO-220's was what repulsed me from this thread months ago... Seriously, it's like 6 of the correct FETs is under $100... It's f'ing stupid to play with TO-220s for anything you're not trying to save every last penny on each unit because you're going to build 100,000 of them.

A 6-fet design is always going to be more optimized than any design that parallels fets (12-18-24 etc). No more matching gate voltage plateaus, inductive path sharing issues, first gate trigger event ringing the other gates, etc etc.

If you want esoteric designs, you simply choose the FET that suits the phase needs, and use it. You don't try to bundle together groups of inadequate fets to try to combine into suiting the needs of the design.

For lower power applictions, if you use a 500amp capable FET, you run it at whatever your needs happen to be, like 50-100amps or whatever, and just enjoy a controller that never gets warm, and will never have to be a concern for you when chugging up hills etc. It's something you can just cross off your list of parts to worry about on the bike. And, with this extremely clever package, you can make a super easy to build and easy to heatsink FET layout (like seriously, 2 copper bars, some screws, and an old CPU heatsink), and end up with a smaller end package than you would have had from using TO-220's to begin with.

 

[liveforphysics]

That sounds like a great arrangement; thanks for that my friend
...
Alan_B, I am beginning to think we should just drop the TO220 then; lots of conversation to move to TO247 - I mean... it is the leading package.

Jazzed! KF

I would leave the final TO-220 accomodation decision to the PC layout detail phase (put it as a low priority option), if it has no negative repercussions, do it. Otherwise let'em bend leads. Optimize for the favorite package. My $0.02.

One other thought on the PCB copper weight - do the pc layout with as much trace width as practical (and parallel traces on both sides if possible)(within the other design considerations), and route traces so wires can conveniently be soldered to them (make them straight and wide enough for that), in case the heavy copper is not practical or not enough. Go as far as practical and allow for improvement.

On those 1" square packages, how does one get them all to connect to the heatsink? Seems like they would not be easy to get exactly planar, sitting on those J leads. They might be best for separate heatsinks per device. Seems like a special project. I'm set up to solder those, how many folks are?

Let the TO-220 suckers bend legs if they want. It would be retarded to make the board with the narrow traces to fit TO-220's rather than the width the TO-247 package enables with it's leg spacing. You have seen the mounds of solder and copper we have to pile on the back of stupid TO-220 spaced board traces right? And then they still pop traces...


With this new IXYS package, there is actually no reason to have any high current PCB. You simply have 3 pairs layed out on the sink, with a big - copper bar on one side, a big + bar on the other, and the inside legs get soldered together around a big bullet connector that becomes the phase-leg wire connection point. Then snuggle in as much cap as your application needs across the +- rails. PCB's are really poorly suited towards things with current levels much over 100amps anyways.

 

[rhitee05]

Sorry I've been out of circulation for a few days. All quiet for a while, then it gets busy!

Boost, re: TO-247 current limits:
If you read the datasheet you linked carefully, there's a little note at the bottom of the page which reads:
"Calculated continuous current based on maximum allowable junction temperature. Bond wire current limit is 195A. Note that current limitations arising from heating of the device leads may occur with some lead mounting arrangements (See AN-1140)."

If you look up that app note and read, near the end you'll see the following:
"Table 1: Ultimate current ratings for packages with different wire bonding configurations. With no attention to lead thermal management, the recommended current for all of the packages above is 75A."

Where "all of the packages above" includes TO-220, TO-247, and TO-262. If you read through the rest of the discussion in this app note, you'll find that the standard thru-hole mounting provides very little in the way of heat dissipation for the package leads, so a continuous limit of about 75A is appropriate for both TO-220 and TO-247 packages in most cases. Since this is a thermal limit, peak loads could be closer to the 195A limit you quoted for brief periods so long as the average limit is respected.

-----
ZapPat:

The 150Vmax spec is only possible if we keep our current requirements low and/or succeed in making a functionnal switching-type pre-regulator circuit (as Eric said he might succeed at a while back - have you managed to test that, Eric?).

I haven't had the time to do much more since I talked about that earlier. I'm confident that the design is valid, though. I would still like to put them through some paces with SPICE, but I think it should be mostly a matter of design tweaks.

 

[liveforphysics]

For current levels in the <200amp range (like 99.9999% of ebikes), there is no need to have to struggle with control heat and over-current and fans on controllers and waiting for them to cool down, and blowing them climbing hills etc etc.

This is a problem CREATED by the china bean-counters that don't want to spend over $0.15 on a FET, so they tool up the board to use TO-220's.

You do not have to create this problem into your own designs as well.

IMO, the whole reason to do your own controller is to not bring with it the problems inherent to the all ready available controllers.


But anyways... I've said my piece. Go for whatever you guys want. Save that $20-40 on parts cost to inherit the status-quo controller problems.

 

[liveforphysics]

so a continuous limit of about 75A is appropriate for both TO-220 and TO-247 packages in most cases.

I've got the IRFB4468 TO-247 package and IRF4110 TO-220 package both in my hand right now. I'm looking at roughly 4x the leg cross section on the TO-247 package here my friend. I don't have a mic handy or I would measure it exactly for you guys. I can definitely confirm that you could bundle all the TO-220's legs together and not equal one leg on the TO-247.

 

[liveforphysics]

If I had to take a stab at the absolute minimum thickness difference, I would say +30% thickness, and +50% width in the most narrow part of the leg. +30% thickness and +250% width in the upper part of the leg (the area I always make sure my FETs sink down to for mounting anyways).

 

[liveforphysics]

Just looked at the datasheet for the IRFB4468

Shows an interesting range in leg thickness. Evidently they leave a generous range in potential mfg/coating deviations.

0.38mm thick minimum to 0.89mm maximum
1.65mm wide mimimum to 2.39mm maximum


IRFB4110

0.36mm thick minimum to 0.61mm maximum
0.38mm wide minimum to 1.01mm maximum

 

[Kingfish]

...
FET

• Number of FETs: 12. Locking.
• FET Package: TO-247 is favored; personally I’d like to have both packages available on the PCB for greatest flexibility. Let’s see how this plays out.
• PCB Copper Plate: Ultimately the layout will decide, though presume > 1 ounce.

Open Issues
FET High Current: Where to draw the line? Should we define a matrix as ≤ 75V, 100V, 150V vs. ≤ 50A, 100A, 150A? I am happy to work up a chart for a BOM if someone will provide the data
...

ADDENDUM:
FET

• Number of FETs: 12. UNLOCKING.

That is... if we're all in agreement :wink:

Let's sort out a recommended matrix for various voltages.

~KF

 

[oldswamm]

Gee lfp, didn't mean to get you upset. KF mentioned the possibility of making the pc board accept to220s, and I said I didn't think it would be much problem.

You are one of the smartest and most knowledgeable people on ES in my opinion. If your price for participating is that we never mention to220 again I'm all for it. I've known for decades that they were shit. Why we used to use TO3 if we needed to dissipate heat. Only only reason the to3 went obsolete was they were large expensive and hard to use. (And no, I'm not suggesting trying to find to3 mosfets )
If someone wants to use those little transistors, they can bend the leads, (and not tell you about it).
That 75A lead limit does sound bogus. And I noticed that they allow themselves a HUGE variation in lead cross section. Might make transistor purchase interesting. Even the smallest should be good for 100A you would think though.
I suspect that you're going to want to use it with external buss bars mostly anyway. Can't you take the drives from a board designed for on board FETs, and use them for bus bar mounted fets. Could it be modified to drive the big, block FETs? Are you using them yet? How are you driving them?

I'm all for lower voltage and higher current. I don't think we SHOULD be using 150V on an ebike. It's partly because of a dearth of high current controllers that the commercial motors use relatively high voltage and low current, (and the cost of copper). But, THIS controller should ALSO be backward compatible with the commercial motors that most of us will want to use it with, if it's going to attract enough contributors to make this project succeed.

KF
I'm fine with 6 FETs, but would vote for keeping it at 12 if it didn't cost much, esp in terms of drive complexity, efficiency or speed.
I'm for anything that will make the design more versatile without adding much to the cost or limiting it's functionality.
It seems to me that the more versatile it is, the more contributors you will attract, and the more likely the whole project is to grow and prosper.
The modular design pretty much does that, if we can get past V1. After we have a working controller, boards can be designed specifically to drive external fets, or someone can even design a board using to92s (if they think of a reason), or whatever.
If we keep it at 12 FETs, 150V @ 150A+, or 100/200 should be no problem should it. I wouldn't mind building up traces for a maxed out build.

Bob

 

[rhitee05]

I'm not going to dispute that the legs are in fact thicker on a TO-247. But, read the app note - I quoted it directly, so if you disagree take it up with IRF.

I think it's mostly a moot point anyway. If you need to be pushing more than 75A per package continuous you need to be using a bigger FET than even TO-247. You should be using one of the large insulated-substrate packages.

 

[ZapPat]

That's an interesting document (AN-1140)- it probably describes best our true current handling bottleneck. Thanks for posting the link, Eric.

I read this app note as saying that our real problem is not so much the lead size itself, be it TO-220 or TO-247, so much as the nature and quality of the contact area's between the FET and the PCB. Here's an additionnal quote from the app note that sums up the conclusion they reached:

The primary constraint upon the amount of current is the temperature of the source contact on a printed circuit board.

There's also another problematic too, which is the PCB's current handling capacity. Expecting much over 140A out of even a 3oz PCB will be pushing it I think. Bus bars would solve this issue, and if some sort of mechanical source pin clamping setup (instead of soldering) could be made then the source connection point main current bottleneck described above could be eliminated too.

 

[Jeremy Harris]

I'm not going to dispute that the legs are in fact thicker on a TO-247. But, read the app note - I quoted it directly, so if you disagree take it up with IRF.

I think it's mostly a moot point anyway. If you need to be pushing more than 75A per package continuous you need to be using a bigger FET than even TO-247. You should be using one of the large insulated-substrate packages.

It ain't a lead thickness issue, it's a bond wire issue, specifically the thermal constraints of chucking loads of amps through very thin wires. AFAIK, the 75A limit is pretty much constant across all the three lead package FETs, be they TO220, TO247 or whatever.

You're spot on in picking up that app note key point, the thermal limitations of these packages is really the killer problem. I matters not one jot what the silicon is rated for, in fact it doesn't really matter what the leads are rated at, the key issue is just as it's always been, getting heat out of a tiny, crappy package fast enough. Even with a perfect heatink (one with a 0 deg C per watt rating) these small FET packages will cook way below their rated 'limit'.

The reason I don't chip in here on this thread is because it seems stuck on using a crap approach to the really big challenge with any controller - providing a very robust way of switching fairly high currents, reliably. Anyone can design in loads of fancy features at the front end, that's just a bit of code hacking. The really serious issue is switching currents ranging from a few tens of amps up to many hundreds of amps cleanly, efficiently and above all reliably. It seems not to be a priority with this project though, which is a shame. I'll go back to my 6 FET Xie chang driving some big Ixys FETs -as far as I can see all the features most want are already in the Xiechang firmware/hardware.

Jeremy

 

[Kingfish]

Mechancial clamping makes me think of CPU-latching. What about using high-quality silver solder instead?

Biting on the hook, what packages other than TO-247 should we consider then?

~KF

 

[liveforphysics]

I matters not one jot what the silicon is rated for, in fact it doesn't really matter what the leads are rated at, the key issue is just as it's always been, getting heat out of a tiny, crappy package fast enough. Even with a perfect heatink (one with a 0 deg C per watt rating) these small FET packages will cook way below their rated 'limit'.

The reason I don't chip in here on this thread is because it seems stuck on using a crap approach to the really big challenge with any controller - providing a very robust way of switching fairly high currents, reliably. Anyone can design in loads of fancy features at the front end, that's just a bit of code hacking. The really serious issue is switching currents ranging from a few tens of amps up to many hundreds of amps cleanly, efficiently and above all reliably. It seems not to be a priority with this project though, which is a shame. I'll go back to my 6 FET Xie chang driving some big Ixys FETs -as far as I can see all the features most want are already in the Xiechang firmware/hardware.

Jeremy

I'm with you 100% on this Jeremy.
This is why I hardly take more than a glance at the current specs on the sheet.
This is why I'm in love with the special packages more than anything else. And why the new IXYS series makes me drool with it's RthJC 0.09 °C/W
RthCS 0.15 °C/W in ceramic isolated packages that you can screw right to the sink.

Or the FETs that you and I invested in... with the 0.04 combined Rth in an isolated package, and effectively no package limits. Not really practical for ebike controller needs though when the off-the-shelf price on the FETs alone exceeds most folks total battery budget. lol

 

[Jeremy Harris]

Luke's already pointed to a pretty good package with those Ixys FETs. The very best bet seems to be the industrial FET packages, not necessarily the really big ones, but the smaller Isoplus 227 packages. The other big advantage of going down this route is that it makes the main board design and construction independent of the power stage. If the power stage is going to be some large format FETs bolted to a decent heat sink, then the design breaks down into a universal 'brain' part, plus some FET drivers mounted near the FETs. For those that really want to stick with the TO220/247 format the driver board could include FET connections, but for those that want to exploit the true capability of the controller the driver boards(s) could be fitted to any one of maybe two or three large format FET packages.

Jeremy

 

[rhitee05]

The concept here all along has been for a modular design, so the same MCU and power supply could drive a variety of different FET stages. Jeremy, perhaps you point out a useful added feature. Besides the footprints for on-board FETs, it shouldn't be that difficult to add footprints for connectors so that the board could drive off-board FETs like the big modules. You might still want a local high-current buffer at the FET package, but one of the key improvements this design will include is a reasonably strong driver which the Xie Changs sorely lack.

Super-size FETs aside, I think this project offers several advantages over the Xie Chang line:
- Use of switching main power supplies. This seems like small change, but saving a few watts here and there will help keep things cooler.
- Open software design, add your own custom features, modify it, or custom design from scratch.
- Modular, documented board-to-board interface to allow add-on modules.
- Proper gate drive design. Well laid-out, good current capability, none of this 2-3 us switching time business. Again, less heat.
- Now I'll add one more: ability to drive off-board FETs.

I don't think the goal of the project is to compete with the Xie Chang directly in any way. It would never be competitive on cost, and the added features are of no interest to probably 90% of the market. I think the goal is to provide a better platform for the other 10% who have more interest in tweaking and customizing for themselves. Thus the emphasis on modular design and expandability.

 

[Kingfish]

...
- Now I'll add one more: ability to drive off-board FETs.
...

I could see this one developing into a feature; adding to the scope, and I'll leave it open as an option for V1 if it is a small adjustment to the pcb layout.

~KF

 

[Kingfish]

TO-247:

• IRFP4468PbF: 100V, 195A/290A
• IXFK360N15T2: 150V, 360A (also in TO-264)

TO-264:

• IXFK360N15T2: 150V, 360A

Compare/Contrast TO-247 & TO-264

• Pad Spacing = same
• Slight differences in lead/hole size
• TO-264 is larger in Width and Length.

Conclusions:
Design for physical size of TO-264 and the TO-247 should drop right in.

Square package

• MMIX1T550N055T2: 55V, 550A
• MMIX1F520N075T2, 75V, 500A

Conclusions:
Difference board layout.

Do we have candidates for 75V in TO-247 & TO-264, or should we just not worry?
~KF

 

[rhitee05]

IRFP4368 is a good choice in TO-247 for 75V.