Please educate me on controller design.

Something's been bugging me.

Controllers for BLDC motors are commonly referred to be the number of FETs which they contain. I commonly see controllers advertised as "6 FET", "9 FET" or "12 FET", with 18 and 24 FET models comprising the extremely high-end of things.

Now, from a schematic point of view, we have three motor windings in a Y configuration, and thus three phase legs that we need to deal with. And at any given moment, each phase leg can be either positive or negative. So that requires two FETs per phase (one each to + and -), times three phases.

So that makes perfect sense for a 6-FET controller. We have the minimum number of switching devices required for a complete circuit. And it also works for 12, 18 and 24 FET designs, as these are even multiples of six, and must thus represent parallel banks of 2, 3 or 4 FETs in each position.

So how in bloody hell does a 9 FET controller work? Are the FETs doubled on only one side, possibly reflecting the use of a mix of devices with different ratings? Are people making use of some alternate definition of the word "nine" with which I am not familiar?

I'm sure that once it's explained to me I will probably slap my forehead and say "Duh, that makes perfect sense," but for the life of me I can't seem to find either a schematic or an explanation which frees me from this mental block.

In a 9 FET, there are 2 on one side (highside IIRC) and one on the lowside - all identical components though.

Huh.

I finally managed to find a couple of images of the underside of a 9-FET controller (thank you, Google images) and it would certainly appear that you're right.

And now I'm even more confused. As they say, a chain is only as strong as its weakest link.

The basic idea of paralleling FETs is sound. They load-share well and so two FETs in parallel have double the current capacity and half the on resistance of one. Thus, 12 FET controllers make perfect sense- at every point in the circuit where phase current must be switched, two FETs always share the burden in parallel.

But what's the problem that a 9 FET controller is the solution to?



If the idea is to allow the unit to pass more current, I'm not seeing how this helps. The one FET on the low side of each phase is still going to be the limiting factor- it'll be passing as much current and dissipating as much heat as any one FET in a 6 FET controller. The two FETs in parallel on the high side will be running easier and cooler, but that doesn't change the dynamics of the overall circuit much at all insofar as where the operating maximums wind up.



I guess one could argue that the lower RDSon of the two parallel FETs decreases the total on resistance of the circuit as a whole, but this is a pretty trivial matter. And when I sat down to the math on that, things got even more confusing.

I'm no expert on all of the various controllers out there, but I gather that the Infineon controllers which Lyen sells seem to be at the higher-end of the quality spectrum. Looking at his 6 FET controller, I see that he uses IRFB4110s, which have an RDSon of 3.7mΩ typical. Two of those in series makes 7.4mΩ (we'll assume "typical" figures to be true for this example).

But then I look at his 9-FET controller, and see that it uses IRFB4310s, spec'd at 4.8mΩ. So a series/parallel combination of these (two high-side, one low-side) has a total on resistance of 7.2 mΩ, which is, for all practical purposes, the same.

Now, this is where I get murky on the internal operating principles of these controllers, but comparing the two FET types, their dynamic and recovery properties seem to be quite similar, so regardless of whatever sort of PWM-based wizardry is going on, I would expect them to be making about the same amount of heat.



I'm sure there must be SOME logical explanation for why this architecture exists, but for the life of me I can't figure out what it is. And I hate not understanding things.

They simply have two high-side switches to get the conduction losses down. I havent examined these controllers closely - but I think they only pwm the lower fets, also reducing switching losses. Ofc this makes for unevenly distributed heating between the fets. But given a finite heatsink capabillity, a 9fet definitely have better current handling than a 6-fet.

Could it have something to do with regen or dynamic braking? Perhaps some dedicated FETs for that purpose...or doubled up on side for that purpose?

As I understand it, they parallel the leg that gets the freewheeling diode currents, so that there's less heating in each one, theoretically making it more efficient than a controller with equal number of FETs per leg.

I'm not sure how effective it is, as I don't have a 6FET and 9FET that are otherwise identical to test the theory out, using the same bike under the same conditions.

One big issue with anything other than a 6FET is current-sharing, due to tiny differences in resistance between each FET--unless they were all matched to each other, thsoe differences if great enough can negate much of the point of having paralleled FETs, when running near their limits, as the lowest resistance unit will get most of the current and could heat to the point of failure before the other FETs get a chance to help share.

Teh Stork said:

They simply have two high-side switches to get the conduction losses down. I havent examined these controllers closely - but I think they only pwm the lower fets, also reducing switching losses.

Click to expand...

Ah!

This actually starts to make sense if I reverse what you wrote- that they only PWM whichever set of FETs in paralleled. That's where the real heat is going to get generated, so by spreading this load, and running the single FET in continuous conduction while the opposing FET pair is modulated, the overall capacity of the system would be raised.

I had never even considered that it's only necessary to modulate one side of the phase, but now that I see this to be the case, it's becoming obvious.

amberwolf said:

One big issue with anything other than a 6FET is current-sharing, due to tiny differences in resistance between each FET--unless they were all matched to each other, thsoe differences if great enough can negate much of the point of having paralleled FETs,

Click to expand...

Well, one of the interesting dynamic properties of a MOSFET is that RDSon increases as temperature increases.

So to start with, you might have a current imbalance due to one FET conducting more than the other. But as soon as that FET started to heat up a bit, it would begin shedding load onto the other. The system as a whole should tend to self-equalize provided that the FETs are not seriously mismatched.

If there was time for this to happen, then yes, it would. but in the case of hotrodded controlelrs or running near their limits, they may blow before they get that chance. I've had single FETs blow on controllers being pushed really hard, and I expect this is why it happens.

Also, since all the FETs are heating anyway, then RDSon goes up for all of them, but not equally. It is likley (havnet tested this) that RDSon goes up faster for ones that are already higher RDSon to start with, so this will cause further imbalance, if true.

amberwolf said:

If there was time for this to happen, then yes, it would. but in the case of hotrodded controlelrs or running near their limits, they may blow before they get that chance. I've had single FETs blow on controllers being pushed really hard, and I expect this is why it happens.

Click to expand...

Could be, if you're really pushing them.

But the magnitude of the imbalances we're talking about is pretty small.

Take a 4110 as an example. We'll start them at 25 degrees C, and assume that we have one whose RDSon conforms to the "typical" spec of 3.7 milliohms while the other exhibits the worst-case value of 4.5 milliohms. It's easy to solve for current in parallel resistors. If we assume a total current of 50 amps, the first FET will be carrying 27.4 amps and dissipating 2.8 watts, while the other will be carrying 22.6 amps and dissipating 2.3 watts. It's a relatively small imbalance, and one which will quickly self-correct. Even though the two devices may be sharing a heatsink, they still have an internal junction-to-case and case-to-sink thermal resistance, which will permit them to operate at different temperatures internally.

The manufacturers would seem also to agree. Quoting from Fairchild Semi's "MOSFET Basics" (document # AN-9010):

4. Forward voltage drop with positive temperature coefficient - easy to use in parallel.

When the temperature increases, the forward voltage drop also increases. This causes the current to flow equally through each device when they are in parallel. Hence, the MOSFET is easier to use in parallel than the BJT, which has a forward voltage drop with negative temperature coefficient.

Click to expand...

I'm not saying that it's impossible to blow one FET in a pair- empirical evidence proves this. But the fundamental concept of putting two FETs in parallel to increase current-carrying capacity is a sound one. Bear in mind, also, that when two FETs are placed in parallel, the effective resistance of the pair follows the same rules as for normal resistors in parallel- two FETs have a lower total resistance then one FET (precisely half, assuming they are all evenly matched.) And because the total power dissipation is less, the total heat generated will be less, and thus the effect of RDS rising with temperature will also be less.



Of course, all of this presupposes that in a 9-FET controller, PWM is in fact being applied to the paired FETs and not to the single FET, which is the opposite of what Stork wrote. The vast bulk of heat generation here comes not from static RDS, but from the period of time that the FETs spend in the linear region during the on-off and off-on switching due to gate capacitance. Thus, the PWM'ed FETs should run a lot warmer than the non-PWMed ones, all else being equal.

I wish I had a 9-FET controller handy to test this theory on...

Joe Perez said:

I'm not saying that it's impossible to blow one FET in a pair- empirical evidence proves this. But the fundamental concept of putting two FETs in parallel to increase current-carrying capacity is a sound one.

Click to expand...

Other things to consider (when only one of a paralleled pair of FETs blows) are that the blown FET...
- Had one or more manufacturing defects and would have never lasted long anyway.
- Was mounted with too much or too little thermal paste.
- Had too much (big problem with TO-220-cased FETs) or too little tightening of the mounting screw.
- Was mounted in a location that resulted in a higher operating temperature (i.e., between FETs and not on the end, near something else hot, etc.).
- Was closer to another PCB trace that carried a signal that might have generated gate noise (and excessive power levels) if the gate resistance wasn't high enough.
- Had static damage during handling or installation.

Joe Perez said:

The vast bulk of heat generation here comes not from static RDS, but from the period of time that the FETs spend in the linear region during the on-off and off-on switching due to gate capacitance.

Click to expand...

Switching losses are generally fairly modest, probably never more than around 25% to 30% of the total FET losses in a typical 15 kHz PWM frequency ebike controller. The pattern of failure we typically see seems to be from FET junction heating due to I²R losses. The primary cause seems to be high phase current with a low PWM duty cycle - this seems to be the time, more often than not, when FETs fail in controllers. Under these conditions current sharing between devices doesn't have time to become very effective - a high Rdson FET will heat its junction to failure temperature before current has been effectively transferred to the others in the parallel set. Vth also varies with junction temperature, so it's possible that not all the FETs in a parallel bank will turn on at exactly the same instant when operating at high phase current with a short PWM pulse, adding to possible current imbalance under these conditions.

Heatsink effectiveness has a marginal effect, as the bulk of the problem seems to be the high thermal resistance between the junction and the heatsink (the combined thermal resistance of the junction to case and case to heatsink) together with the very small thermal mass of the FET junction. When FETs fail dramatically in this way the local junction heating is often enough to blow a hole in the encapsulation, as the junction heats up far too quickly for heat to have time to be conducted away to the heatsink.

The problem is compounded as many of the controllers available use poor FETs to start with, FETs with an Rdson of 6 to 10 mohms aren't uncommon. There are also very large numbers of Chinese made counterfeit FETs around, particularly FETs that attract a premium price, like IRFB4110s. It's unlikely that these counterfeit FETs have characteristics that are the same as the genuine parts, so may well exhibit higher than expected Rdson, for example, giving higher than expected losses.

This may help or create confusion! Below is the result portion of one of m spreadsheet models for a 12 FET IRFB4110 based controller. Note that for low PWM periods the flyback diode losses dominate, also how things change as you go to higher PWM periods. Note that more heat is produced at low PWM periods. The junction temperature is not calculated, but is an input parameter.

Those switching losses look much higher than I'd have expected, plus I suspect that the phase current of 200 A is a fair bit higher than the real maximum phase current a 12 FET controller would see, particularly if driving a hub motor, with it's very high LR.

Has you're model been validated? The reason I ask is that I'm not at all sure that real-world switching losses are anywhere near 59 watts per FET, based on the typical operating temperatures we see.

Switching Loss model from Fairchild: AN-6005 Synchronous buck MOSFET loss calculations. I have not validated my code implementation with test.

Here is the run witht 75A phase current. Does it fit your intuition better at a lower phase current?

bigmoose said:

Switching Loss model from Fairchild: AN-6005 Synchronous buck MOSFET loss calculations. I have not validated my code implementation with test.

Here is the run witht 75A phase current. Does it fit your intuition better at a lower phase current?

Click to expand...

Thanks for the source, much appreciated.

It's not so much intuition, as knowing the sort of temperatures that these things run at that bothers me about those high switching losses, together with experience of changing out some stock 75NF75 FETs (13 mohm Rdson) with IRFB3077 FETs (3.3 mohm Rd on) and getting a very significant change in controller temperature. This was with the same controller settings on the same 6 FET controller on the same bike. The controller went from getting uncomfortably hot after some heavy riding to getting just barely warm. That told me at the time that the bulk of the FET losses must have been coming from the I²R loss in the silicon and package, rather than from switching or diode loss, as the switching and diode loss would be pretty much the same for both of these FETs. I believe that others have experienced the same thing, with controllers running a lot cooler with lower Rdson FETs, too.

It makes me wonder whether the Fairchild model for buck converter losses is properly applicable to controllers. I agree that in principle it should be, but practical experience seems to indicate that it isn't.

Good points worthy of consideration! That is why we need more folks modeling and analyzing!

Jeremy Harris said:

It makes me wonder whether the Fairchild model for buck converter losses is properly applicable to controllers. I agree that in principle it should be, but practical experience seems to indicate that it isn't.

Click to expand...

Yes, a motor is basically a three phase buck converter - so it should give good results.

But, theory and practice does not allways go hand in hand. Some of these cheap china controllers does not seem to actively "allways keep one fet on" as in synchronous rectification. They rely that the diode supplies the current - and does not care if this means higher losses, it seems...

bigmoose, thanks for that analysis. It confirmed my suspicion that switching losses vastly eclipsed FET conduction losses (even at the relatively low PWM frequency that we're operating), though I was surprised to see that diode conduction loss was as high as it was.

I hadn't previously read Fairchild AN-6005, but it's quite illuminating as well.

So I guess the underlying theory is sound. If we call the two FETs on any given phase Alice and Bob, and only Bob is being PWMed while Alice remains on for the entire phase duration, then the amount of heat generated in Bob will be roughly double that of Alice. So assuming equitable current distribution between Bob and his Siamese twin brother boB, two BoBs and one Alice make for a reasonably well-balanced design.

Great. Now I'm wondering why the heck we bother building 12 FET controllers.

(Note to the EEs: design a 12 FET controller in which the allotment of which FET group gets PWMed is swapped on every other cycle. Actually, that's probably a patentable idea. Forget I said anything.)



Jeremy noted earlier that "Under these conditions current sharing between devices doesn't have time to become very effective - a high Rdson FET will heat its junction to failure temperature before current has been effectively transferred to the others in the parallel set. Vth also varies with junction temperature, so it's possible that not all the FETs in a parallel bank will turn on at exactly the same instant when operating at high phase current with a short PWM pulse, adding to possible current imbalance under these conditions."

This is true. But I'm having a hard time imagining a scenario in which any reasonably well-matched pair of FETs could be so imbalanced that the amount of current passing through them differs by more than 30% or so. So maybe 2 FETs don't precisely double current-carrying capacity. They should at least improve it by 1.7x or more.

Jeremy Harris said:

It makes me wonder whether the Fairchild model for buck converter losses is properly applicable to controllers. I agree that in principle it should be, but practical experience seems to indicate that it isn't.

Click to expand...

Teh Stork said:

Yes, a motor is basically a three phase buck converter - so it should give good results. ...

Click to expand...

A motor controller, on an e-bike, is a thee phase buck converter in a tornado, during an earthquake, perhaps.
I am poking around in controllers a bit, lurking regularly and Googling excessively.
It is taking some time to get up to speed. I've not encountered even one 12AX7 or 25L6.
It's an entertaining challenge to figure out how things work while wondering what the hell were they thinking.

Thank you, Ye Gurus, for the many great posts!

tedcs said:

It is taking some time to get up to speed. I've not encountered even one 12AX7 or 25L6.

Click to expand...

Hahaha. Believe it or not, I think I still have some old 12AX7s rolling around in a box somewhere along with a bunch of other small-signal tubes. I have utterly no idea why I keep hanging onto these.

Joe, I will take every 12AX7 you can ship me for $2 and shipping!! After I get them, I'll tell you what I do with them!

... don't throw them out. check here: http://www.audiotubes.com/12ax7.htm

I think the only ones I have left here are working in my oscilloscope.... (tubes; dunno if any are those specific models)

bigmoose said:

Joe, I will take every 12AX7 you can ship me for $2 and shipping!! After I get them, I'll tell you what I do with them!

Click to expand...

Well, I went and checked the box. I've got five RCA 12AX7s, and about a dozen other mixed small-signal tubes, plus a couple of miniature CRTs and camera tubes.

Understanding that these are not NOS units (they were pulled from working broadcast equipment and probably have a decade of runtime on them) I'm damned curious to know what use you'd have for them.

I have a buddy that does custom designed guitar tube amplifiers. His tube of choice is the 12AX7 for the pre amplifier. He has a penchant for NOS 12AX7's from the "good old days..." but as you say they are getting very scarce. He has us all looking for any 12AX7's from the 60's and 70's hoping for one or two good ones out of a lot of 5 or 6.

That was the basis for my comments. Even though I am going partially deaf, when I go to his shop/lab and he changes tubes out and plays for me, oftentimes I can tell the difference between the 12AX7s of the past and the Soviet replicas of today.

bigmoose said:

I have a buddy that does custom designed guitar tube amplifiers. [...] oftentimes I can tell the difference between the 12AX7s of the past and the Soviet replicas of today.

Click to expand...

I've been told that a vacuum tube is a FET with a light on, plus signal processing. Based on the substantial demand, they are useful.
My supposition is that, with sufficiently thorough analysis, the transfer function of a vacuum tube could be duplicated exactly with other types of circuitry... that which is possible is not necessarily practical or desirable, however.