PC high-current PWM fan controller issues

[Zero999]

Mm, I guess there is some miscommunication what a totem pole is.

To my knowledge it is just a NPN/PNP pair.
As you can see here, page 17
http://www.ti.com/lit/ml/slua618/slua618.pdf

Level shifting with just a ground is pretty easy actually.
(it's a bit more a pain when working with a symmetrical power supply)

That's what I thought you meant but wanted you to post a schematic to confirm. The trouble is, that circuit has no voltage gain. The output simply follows the input minus 0.6V on positive cycles and plus 0.6V in negative cycles. Another transistor is required to provide voltage gain, to get 12V pulses from 5V pulses, hence why the circuit I posted previously has three transistors.

I guess what you're looking for here as well, is a bootstrap circuit.

The trouble with a bootstrap, is it will never allow full duty cycle. An off period is required to recharge the capacitor. Other than that it's a good idea and enables lower on resistance N-channel to switch the high side.

Here's another round of schematics and artwork.

Changes from last time:

Swapped 1st stage from BJT's and resistors to a TC4426 based driver
Added main filter cap C3 and bypass cap C4

---

The TC4426 is cheap enough. I don't want to use SMD if I can at all avoid it... This isn't a mass production project (I'm making 3-5 units, for me and a buddy or two). IMHO not worth it (not to mention not as easy to breadboard).

That looks good so far. Yes, the TC4426 is inverting and is just what you want.

The 1N4001 is only rated for 1A at mains frequencies. D4 and D2 should be replaced with some faster, higher current, diodes, such as the 1N5817.

I haven't checked the inductor calculations. I suggest doing it again, either manually or using a different calculator page, for a sanity check and round up the result to the nearest available value, never down. The current rating should be higher, than what the fans will draw, to avoid core saturation.

[b_force]

With a buck converter (or half bridge), you always need a bootstrap?

I don't know what the further idea is, switching freq, needed deadtimes etc. But there are enough halfbridge drivers that can do the job and deliver enough current.
There are also little high side drivers etc.
I think I would just go for half bridge configuration though.

But isn't all of this just a bit much for just a simple fan?

[Zero999]

With a buck converter (or half bridge), you always need a bootstrap?

I don't know what the further idea is, switching freq, needed deadtimes etc. But there are enough halfbridge drivers that can do the job and deliver enough current.
There are also little high side drivers etc.
I think I would just go for half bridge configuration though.

But isn't all of this just a bit much for just a simple fan?

A bootstrap is only needed, if the high side device is N-type, as the gate voltage needs to exceed the output voltage, by enough to turn on the MOSFET. In this case the high side device is P-type so it's not needed.

[b_force]

With a buck converter (or half bridge), you always need a bootstrap?

I don't know what the further idea is, switching freq, needed deadtimes etc. But there are enough halfbridge drivers that can do the job and deliver enough current.
There are also little high side drivers etc.
I think I would just go for half bridge configuration though.

But isn't all of this just a bit much for just a simple fan?

A bootstrap is only needed, if the high side device is N-type, as the gate voltage needs to exceed the output voltage, by enough to turn on the MOSFET. In this case the high side device is P-type so it's not needed.

That is correct.

The thing is that using two N-types is far more common nowadays.
Doesn't mean it's impossible of course. Just wouldn't be my path to make it work I think.

[ratdude747]

With a buck converter (or half bridge), you always need a bootstrap?

I don't know what the further idea is, switching freq, needed deadtimes etc. But there are enough halfbridge drivers that can do the job and deliver enough current.
There are also little high side drivers etc.
I think I would just go for half bridge configuration though.

But isn't all of this just a bit much for just a simple fan?

A bootstrap is only needed, if the high side device is N-type, as the gate voltage needs to exceed the output voltage, by enough to turn on the MOSFET. In this case the high side device is P-type so it's not needed.

That is correct.

The thing is that using two N-types is far more common nowadays.
Doesn't mean it's impossible of course. Just wouldn't be my path to make it work I think.

Off-topic: I can speak on this a bit.

In school I was taught to use NPN/N-channel to sink, and PNP/P-channel to source. Hence my design. I also read (on wikipedia) that NPN/N channel is often cheaper as it's more widely used (and thus made in bigger bulk), so I could understand why one would find a way to source with one if they had to.

In my day job, I'm an engineer at an automotive component plant (body shell division) who works with a lot of PLCs, welders, robots, and automation. NPN/PNP comes up a lot as some devices have sinking inputs/outputs (referred to as NPN devices), and others have sourcing inputs/outputs (referred to as PNP devices). This is both for controller I/O and sensors. The thing is, this is a regional issue too. We're a Japanese owned company and most Japanese equipment is NPN (Mitsubishi PLCs, Kawasaki/Motoman robots, and many Keyence/Omron sensors), so our company standard is NPN. However, for safety devices (required by OSHA/CSA), the standard is PNP (as is most American equipment), which in the case of one my projects where I'm updating the safety systems on some old robot cells, is a bit of a pain (micro relays for conversion all over the place, ugh).

[b_force]

There is a lot to say to use two (close to) identical devices.

Besides the differences in behavior of a p-channel and n-channel, finding complementary MOSFETs (to get specs as close as possible) is difficult.
I work with high power Class-D amplifier (500-1000W) on a regular basis and these are all n-channels MOSFETs.
With these currents, something simple like a difference in even Rds(on) can ruin your day.
Not speaking about difference in capacitance and all kinds of other parameters, which will result in different switching ringing noise and so on.

Having more similar parts on your BOM is always cheaper in the end.

[ratdude747]

Inductors: https://www.digikey.com/product-detail/en/abracon-llc/AIUR-06-4R7K/AIUR-06-4R7K-ND/2343590

I designed it around 5A per channel, although 6A is the absolute max I'd expect each channel to run. These are rated at 6.2A. Sadly, the calculator I used was for boost converters, so they're probably no good. I'm not getting consistant results from buck calculators; it seems they're not designed for the currents I'm running. I might have to do some old fashioned math myself on this one...

[ratdude747]

Worst case scenario: high load, ~2V drop across mosfet (as per datasheet and ohm's law) low fan speed requested (50% duty cycle). Used this calculator: http://www.daycounter.com/Calculators/Switching-Converter-Calculator.phtml

230uf and 160uH is what I'm getting. I have way more capacitence and nowhere near enough inductance. Whoops. That said, at 50% I bet it's only running 3A, not 6.


Perhaps this: https://www.digikey.com/product-detail/en/signal-transformer/HCTI-220-5.8/595-1732-ND/7362983

[Zero999]

I agree with your conclusions: you need more inductance and the inductors need to be rated for a much higher peak current, than the output current. The current drawn, when the load is the maximum inductor current, in this case.

A voltage drop across the MOSFET sounds too high. It would give a power dissipation of 12W at 6A, so you'll need a fairly large heat sink. Find a MOSFET with a lower on resistance.

I think you need to develop a better understanding of buck converters, even if you don't fully understand the calculations, otherwise you're just blindly trusting some calculator program. Refer to the links below:
http://www.learnabout-electronics.org/PSU/psu31.php
http://www.radio-electronics.com/info/power-management/switching-mode-power-supply/step-down-buck-regulator-converter-basics.php
https://en.wikipedia.org/wiki/Buck_converter
http://www.ti.com/lit/an/slva477b/slva477b.pdf
http://www.ti.com/lit/an/snva038b/snva038b.pdf

[ratdude747]

Other information:

I will be running a heat sink. See attached picture. But its' only rated for 7W at ambient... and that's for both devices.


Supposedly my Mosfet was rated for 10A (or more) at ambient temperatures... but I didn't realize what sorts of voltage drop I'd be talking.

Datasheet for my mosfet: https://www.fairchildsemi.com/datasheets/FQ/FQP7P06.pdf

According to that, with 10V gate voltage, I'd get 0.35 Ohms at 6A... so 2.1V drop, and therefore 12.6W. Darn. (and to think I used to do this math all the time in college... )

That said, for not a low more there are better options. This one is overkill:

https://www.digikey.com/product-detail/en/fairchild-on-semiconductor/FQP27P06/FQP27P06-ND/965349

But from what I'm seeing it's the cheapest unit with a low enough resistance not to burn a ton of power. 1.44W to be exact. Comparing it's 1400pf max capacitence to TC4226 datasheet, I'd get a rise time of 30ns and a fall time of 35ns. Given that a 25khz has a period of 40000ns, that should be fine, right?

I am indeed rusty on buck coonverters... sadly it wasn't covered much in college (other than what they are and why they're better (and sometimes worse) than linear supplies. I did skim this though: http://www.daycounter.com/LabBook/BuckConverter/Buck-Converter-Equations.phtml

Plugging in new numbers for the new mosfet, 48uf and 180uH. Is going bigger on either one OK, or is that going to cause an issue? AFAIK 25khz square waves don't have a harmonic that would match the resonant frequency... which is a good thing for stablity? (although being PWM I'd have to look at my old fourier tables to see what I could get at common PWM fan duty cycles).

[ratdude747]

Found another good reference:

http://www.ti.com/lit/an/slva477b/slva477b.pdf

According to that calculation, since worst case is 50% PWM (top part of the equation is a parabola, vertex is indeed at Vout = 0.5*Vin), I'm getting 66.7uH at 30% current ripple. at 20% (what I originally used for the calculator above) I'm getting 100uH.

Also, from that, I'm getting a minimum capitence of 200uf (20% of 6V for output swing (ripple?)... although I'd want to go higher, so 1000uf per rail should be more than adequate. To make room for the inductors, I'll swap to an 8mm 1000uf 16V (Nichicon HE, my favorite cap for PSU work; I do a lot of power supply recapping on the side).

These numbers seem more reasonable?

[ratdude747]

^ based on the above, I'm thinking this inductor would work:

https://www.digikey.com/product-detail/en/signal-transformer/HCTI-100-7.0/595-1728-ND/7362979

rated for 7A, which should be good, right?

edit- I'm thinking these for the diodes: https://www.digikey.com/product-detail/en/smc-diode-solutions/SB3200TA/1655-1523-1-ND/6022968

and capcitors, some Panasonic FR's should fit the bill: https://www.digikey.com/product-detail/en/panasonic-electronic-components/EEU-FR1C102LB/P15332CT-ND/3072212

20% current ripple at worst case 6A is 1.2A. These have a high-frequency rating of 1.56A, and while I'm not running at the spec'd 100Khz, the 120hz rating of 1.17A tells me that at 25kHz 1.2A shouldn't be an issue. All of this is worst case (3 2.0A fans per channel, not a likely configuration).

Working on a revised PCB design now...

[ratdude747]

Fresh schematics and PCB.

I had to add half an inch to the board size to accommodate the torroids (which after a review of the datasheet, led me to choose 120uH instead of 100uH). Also, I decided to remove some of the solder mask from the back side in current-dense areas to give me an edge. Still, maybe I should shoot for 2oz copper on the next order?

[Zero999]

That's much more sensible: more beefy components and a larger inductor. Please click on the links I posted and at least scan through them. I know you haven't because you went on to post one of them. 

One thing I think you might have overlooked is, the capacitance seen by the MOSFET driver, is higher than the static gate capacitance, so the switching time is slower, than you've estimated. This is due to the Miller effect. Look up MOSFET gate charge.

[ratdude747]

That's much more sensible: more beefy components and a larger inductor. Please click on the links I posted and at least scan through them. I know you haven't because you went on to post one of them. 

One thing I think you might have overlooked is, the capacitance seen by the MOSFET driver, is higher than the static gate capacitance, so the switching time is slower, than you've estimated. This is due to the Miller effect. Look up MOSFET gate charge.

Yeah, looking back I saw that and probably should have edited my post.

---

Looked at: http://www.ti.com/lit/an/snva038b/snva038b.pdf . I got 68uH using some of their assumptions (r = 0.3, etc.), which is in the ball park of what I already had.

All of you other links (minus the one genius here found independantly  ) didn't have equations for L in them, so obviously I can't post about such. 100uH (120 nominal) seems to be a safe number.

----

As for the miller effect, from what I read, that's the reverse transfer capacitence, which for my MOSFET, at 1MHz has a max of 155pf. When combined with the Drain-source static resistance, I'm getting 109ns for T, or 218ns of of charge time (IIRC 2T is generally what's used for "full" charge?). But I'm sure I did something bone headed there. Still, I don't think this will be a huge issue at 25kHz.

MOSFET Datasheet:

https://www.fairchildsemi.com/datasheets/FQ/FQP27P06.pdf

[ratdude747]

One more PCB refresh, moved diodes D2 and D4 (and by extension, C1) due to a heatsink clearance miscalculation.

[Zero999]

That's much more sensible: more beefy components and a larger inductor. Please click on the links I posted and at least scan through them. I know you haven't because you went on to post one of them. 

One thing I think you might have overlooked is, the capacitance seen by the MOSFET driver, is higher than the static gate capacitance, so the switching time is slower, than you've estimated. This is due to the Miller effect. Look up MOSFET gate charge.

Yeah, looking back I saw that and probably should have edited my post.

---

Looked at: http://www.ti.com/lit/an/snva038b/snva038b.pdf . I got 68uH using some of their assumptions (r = 0.3, etc.), which is in the ball park of what I already had.

All of you other links (minus the one genius here found independantly  ) didn't have equations for L in them, so obviously I can't post about such. 100uH (120 nominal) seems to be a safe number.

----

As for the miller effect, from what I read, that's the reverse transfer capacitence, which for my MOSFET, at 1MHz has a max of 155pf. When combined with the Drain-source static resistance, I'm getting 109ns for T, or 218ns of of charge time (IIRC 2T is generally what's used for "full" charge?). But I'm sure I did something bone headed there. Still, I don't think this will be a huge issue at 25kHz.

MOSFET Datasheet:

https://www.fairchildsemi.com/datasheets/FQ/FQP27P06.pdf

The characteristic, you're after is the gate charge, which takes the Miller effect into account.
https://www.microsemi.com/document-portal/doc_view/14692-mosfet-tutorial

I haven't done the calculations, but I think it will be fine.

One more PCB refresh, moved diodes D2 and D4 (and by extension, C1) due to a heatsink clearance miscalculation.

Should be fine but I haven't checked everything. The fat traces and planes have a low inductance and resistance and are good practise for high currents.

[ratdude747]

^PCB wise, I did checks using an actual heatsink and a PCB printout. The same way I've checked new custom footprints for proper pitch and the like.

The Heatsink "fingers" go edge-to-edge, but the IC, resistors, capacitor c4, and diodes D1 and D3 will fit underneath. I don't like putting things under a heat sink but the PCB size was getting a tad excessive.  The issue with D2 and D4 was they were in the flat/bend area where things won't fit (hence the empty channels flanking Q1 and Q2); I'm sure D2 will also fit underneath (as it is in the current PCB layout)

[ratdude747]

That's much more sensible: more beefy components and a larger inductor. Please click on the links I posted and at least scan through them. I know you haven't because you went on to post one of them. 

One thing I think you might have overlooked is, the capacitance seen by the MOSFET driver, is higher than the static gate capacitance, so the switching time is slower, than you've estimated. This is due to the Miller effect. Look up MOSFET gate charge.

Yeah, looking back I saw that and probably should have edited my post.

---

Looked at: http://www.ti.com/lit/an/snva038b/snva038b.pdf . I got 68uH using some of their assumptions (r = 0.3, etc.), which is in the ball park of what I already had.

All of you other links (minus the one genius here found independantly  ) didn't have equations for L in them, so obviously I can't post about such. 100uH (120 nominal) seems to be a safe number.

----

As for the miller effect, from what I read, that's the reverse transfer capacitence, which for my MOSFET, at 1MHz has a max of 155pf. When combined with the Drain-source static resistance, I'm getting 109ns for T, or 218ns of of charge time (IIRC 2T is generally what's used for "full" charge?). But I'm sure I did something bone headed there. Still, I don't think this will be a huge issue at 25kHz.

MOSFET Datasheet:

https://www.fairchildsemi.com/datasheets/FQ/FQP27P06.pdf

The characteristic, you're after is the gate charge, which takes the Miller effect into account.
https://www.microsemi.com/document-portal/doc_view/14692-mosfet-tutorial

I haven't done the calculations, but I think it will be fine.

I'll look into it later. Both driver IC and mosfet are designed for the SMPS application so I'm confident it "should" work... then again, I have come across exceptions.

One more PCB refresh, moved diodes D2 and D4 (and by extension, C1) due to a heatsink clearance miscalculation.

Should be fine but I haven't checked everything. The fat traces and planes have a low inductance and resistance and are good practise for high currents.

That was the goal... plenty of track width for high current traces (and where I couldn't make the back side wide enough, I exposed solder mask to do the old solder flow trick), decent track width for in-between stuff (driver to mosfet connections), and 0.5mm tracks for signal wires (PWM and Tach signals). No need to go to 0.25mm traces on this board, so I won't.

[ratdude747]

Barring any other issues, I'll probably order another round of 5 boards after work today.

[texaspyro]

PWM the fans at 1 Hz!  Use a minimum pulse width of say 100 ms.   When staring up you can give a couple of seconds of 100% duty cycle to make sure they start spinning.   This has worked on every two wire fan I have tried.   Of course the tach output wont work very well with many fans, but it does with a lot of them.

My PWM fan controller PID code n Lady Heather can control the temperature of a GPSDO to millidegree levels...  micro-degree long term (actual absolute accuracy dependent upon the temp sensor)

[ratdude747]

PWM the fans at 1 Hz!  Use a minimum pulse width of say 100 ms.   When staring up you can give a couple of seconds of 100% duty cycle to make sure they start spinning.   This has worked on every two wire fan I have tried.   Of course the tach output wont work very well with many fans, but it does with a lot of them.

My PWM fan controller PID code n Lady Heather can control the temperature of a GPSDO to millidegree levels...  micro-degree long term (actual absolute accuracy dependent upon the temp sensor)

That's not the point of this project.

The intent was to use the PWM control already present on many PC/Server motherboards to control high-current non-PWM fans. In my case, a rackable systems NAS server (which I'll use for other things too) that had 4 high current fans on proprietary hot swap connectors going to a dummy boards that took a standard "molex" plug's 12V and put it on 6 standard fan ports. The Tach was wired up (but not used by the dummy board) but the fans are not PWM. I have a buddy with a similar situation (old dual-motherboard monster case, far newer motherboards to drop in), hence the idea.

To do what you're suggesting would require the addition of a microcontroller with few exisiting components removed.

As for fan spinup, Motherboards already handle this. Since it's a pulled up signal (on the fan/PWM amp side), no signal=full speed. Likewise, PC motheboards from my experience run the fans at full blast during startup as normal procedure either intentionally (to spin the fans up) or because the low level controls haven't started yet.

Using the onboard PWM is desired as it both is an already controlled (sometimes with tachometer feedback) signal source that can also be controlled from software on some platforms. In addition, for non-PWM systems there are already 3rd party PWM controllers for computer fans out there, so one could combine this with that (or do both and use one channel for each) making this a partially universal solution.

I see no need to go through a bunch of trouble to convert/interpret a PWM signal to a lower frequency when the hardware to directly use and drive with the higher frequency exists and isn't cost prohibitive.

Why reinvent the wheel?

[ratdude747]

Ordered more PCBs... 10 of them. The cost difference is marginal and even after adding 10 of another of my projects just for fun DHL sipping still costs more... Yay cheap overseas PCB makers.

[ratdude747]

New parts and PCBs came in, still not working.

Some intial issues:

• Somehow the bypass caps didn't make the order, so I'm running without
• Inductors were too big (lead size)... so I had to trim them and solder on thinner leads (from diode trimmings)

Past that, same as before, Tach signal works great, but no speed modulation. Grounding the input kills the fan.

I wonder if the lack of direct feedback is causing drift due to low load... if so, then this concept is DOA and I've wasted my time and money. I might as well run extension cables to the existing fans and learn to deal with the 6000RPM whine (or once I get the house it's for bought, put it in a closet or something).

Unless there's a chance the lack of a bypass cap is causing the driver to not pick up the PWM signal, I'm out of ideas.

[Zero999]

New parts and PCBs came in, still not working.

Some intial issues:

• Somehow the bypass caps didn't make the order, so I'm running without
• Inductors were too big (lead size)... so I had to trim them and solder on thinner leads (from diode trimmings)

Past that, same as before, Tach signal works great, but no speed modulation. Grounding the input kills the fan.

I wonder if the lack of direct feedback is causing drift due to low load... if so, then this concept is DOA and I've wasted my time and money. I might as well run extension cables to the existing fans and learn to deal with the 6000RPM whine (or once I get the house it's for bought, put it in a closet or something).

Unless there's a chance the lack of a bypass cap is causing the driver to not pick up the PWM signal, I'm out of ideas.

Bummer.

Is the PWM actually changing the output voltage? Have you tested it with a PWM signal source from a signal generator or 555 timer circuit?