Newbie Q. on sync. buck converter for solar

[elt]

I'm having a hard time finding practical info on buck converters.

I've draw what seems to me to be a basic synchronous buck converter but I am confused on a few points.

The high and low side FETs for conversion are on the right. Since I want this to be a charge regulator as well, I've put a FET on the left to short the panel when the batteries are charged.

The micro has complementary PWM outputs with dead band timing generation to make sure that both FETs are synchronized; this seems to be a requirement but I haven't found anything that tells me how long the dead band should be? I think I can do multiples of about 20 nanoseconds. I'm guessing that I need a deadband that insures that the turn-off delay and turn-off time finishes before the turn-on delay finishes plus, perhaps, a little bit more for safety?

If I read correctly, I can use either a P-channel or N-channel FET on the high side. Right now I have it drawn as a N-FET but I've noticed that shorting the panel shorts the battery too. I can throw a diode up top but that seems like wasting power.

I've never used a P-FET in my life but it seems to be the opposite of an N-FET and I'd need to change the pull-down on the gate to a pull-up?

I'd like to handle an input of 4 amps and 100v; 100v so that it can handle the Voc of four 12 volt panels. I'm hard pressed to find a logic level P-FET that can meet those specs. But does it have to? ... Isn't the max voltage across the high side Voc - Vbattery? It might be easier looking for a 80 volt part. I can do the math of loss across a diode's Vf vs. a P-FETs Rdson and realize that the P-FET is only a marginal winner over an N-FET and diode. (That assumes that I really can use and N-channel FET on the high side?)

Hm... I think the issue goes away with a diversion load instead of shorting the panel but I'd like the option of not using a diversion load.

Occasionally, I've seen a schottky diode inserted across the low-side FET without explanation. Is that a snubber or is it part of the rectification? ... it just seems a little strange to me since I put the FET on the low side just to get rid of diode in exactly that position!

Comments, help and pointers are greatly appreciated!

 - Ed.

[s4w2099]

Try not to use a P channel fet if you are concerned with efficiency. Usually P channel mosfets have very high Rds(on) and will drastically reduce efficiency.

To properly use an N channel mosfet with a floating source pin, you need a high side fet driver. Something like the ir21834. You need proper driver for the high side because you wont be able to fully turn off that fet once its on because the source pin voltage is simply floating, and you know that to turn off a N channel  mosfet you need to be at 0 volts (or below) in the gate with respect to its source pin. This IC will take care of all that for you (including dead time)

Dead time in this circuit would be used to prevent both mosfets to conduct at once. You only need enough time to let the mosfets fully open the circuit before the other mosfet starts to close.

This is very dependent on the mosfets characteristics. Check out the datasheet and look for fall time and set your deadtime on the driver IC to something a little bit higher than that value. This should be in the order of 100ns on so so mosfets. Much larger on IGBTs because of the tail off current. Also check to see the reverse recovery time of your mosfet's body diode. And pick that to be as low as possible.

Many times during a switching cycle the body diode will be forward biased. Your dead time should also be enough to let that diode recover and loose its charge or it will just conduct current to ground while your high side fet is turning on. And this will also lower your efficiency, or in very bad case, destroy the body diode.

There are less chances of that happening if you put that red diode across your mosfet's drain and source. but make sure its fast, fast.

[Flux]

Once you get away from play level voltages and currents you have to sort out your fet driving. Driving from logic level may be possible but I can't do it ( neither do I want to but that is for another reason).

You need proper high side and low side mosfet drivers. I am not sure that going synchronous is really worth the trouble at 12v and at 24 I certainly wouldn't bother.

If you use dedicated synchronous driver chips they will be intelligent enough to sort out the dead band. Doing it your way you will need to use more deadband than theoretically necessary and you will need the fast diode in parallel that you describe to control thing in the region where you have got things wrong. You will be partly synchronous and the gain compared with a Schottky diode at 12v will not be great.

P channel fets are best avoided, they virtually don't exist in higher power circuits.

Once you start using dedicated drivers there is no need for them. High side driving is a pain but up to 60v there are plenty of synchronous chips available. When you get near 100v there may still be some chips available but you are getting close to the point where you have to start stringing your own high high and low side drivers together.

Having sorted the driving you need to tame your layout to keep inductive loops as small as possible, you are not going to get power circuits to work on a rat's nest with high impedance drives from logic. Plenty of theory on the internet but precious little useful advice.

Flux

[RandomJoe]

Serious question, as I don't know the answer!

Why are you shorting the panel?

I understand the need for that with a wind generator, so you can stall it out to keep it from over-revving, but is there a need to do this with solar panels?  I know the cruddy charge controller that came with my HF panels just open-circuits them, as does my SunKeeper-6 (a PWM 3-stage charge controller).  But I've heard from a few different places about shorting them when not using the power so it's obviously something that at least used to be done.

If you don't have to short the panel, I'd think that would be the ideal way to go - just open-circuit the input and be done.  

[Flux]

Most controllers open circuit the panel, some short circuit it. I don't think it matters much which way you go.

Open circuiting it with a mosfet needs a high side driver, shorting it wouldn't but I see no need for either in this case. The buck converter should phase back far enough to keep you below charging volts without need for additional mosfets.

Flux

[DamonHD]

Flux,

I know this is teaching my grand-mater to suck ova, but of course you can do the controller all low-side N-channel like my cheap Morningstar SHS-6 http://www.morningstarcorp.com/products/SHS/ but the -ve side of everything cannot be grounded, it doesn't meet at least US codes AFAIK, and thus isn't even sold in the US and Canada.

Rgds

Damon

[elt]

> The buck converter should phase back far enough to

> keep you below charging volts without need for additional mosfets.

Thanks everyone!

I really appreciate pointing me in the right directions. I know that this project is at toy power levels but I hope this will be a learning exercise for putting a big buck converter on the wind mill... so I really want to get things right!

I've googled synchronous drivers and have an idea, more or less, what they are doing. Seeing that, I've put two FETs into my toy-quality simulator and, using a much higher voltage to drive the high side FET, I get it to turn on and off. And seeing it turn off, sure enough, I see that I don't need to short the panel to keep it from charging the battery bank. (I often write that I'm not an EE but I'll bet that no one reading this post ever thought that I was...)

I see something else playing with the low side FET: I have to be precise with the off time... if I leave the low side FET on then the battery will discharge though it. I'll need to be very careful with the control that I don't melt the low side FET.

I did find a 100v driver with 5v inputs: LTC4444-5.

Most drivers I've found want more than 5v. But I also found this cool voltage regulator and gate driver combo, TD220. If I get one of the one-input kind of synchronous drivers then it will make a nice high voltage gate drive signal right in the v-reg package. Somewhere I saw a one-input driver chip with an automatic inhibit if the duty cycle gets too short... it looks like a three chip solution for the wind mill control circuit but I think I'll get a LTC4444-5 for this solar panel circuit ... as always, I appreciate comments, especially folks telling me if I'm totally off base!

Thanks again,

- Ed.

[elt]

Thanks everyone!

I've put a hi-side driver in and juggled the power supply section to handle real FETs.

Higher resolution here.

I am concerned about that happens if a solar array is connected but not a battery. I think that enough power will enter the left-side ADC input to power the microprocessor. I've seen this before, the uP can pretty much be powered from any of its input pins. Perhaps reading the value on the right-side ADC will give an indication of whether it should run its program or not.

The simulator doesn't show but a few tenths volt on battery side if the battery is disconnected so I'm assuming that the solar array won't blow my voltage regulators?

I sampled some chips, with the driver I got I'll have to handle the dead band myself so I left the diode on the output side. Am I going to need that one? I have a big 16nS recovery time diode in my box so I spec'ed that in the circuit. Is that fast enough?

I have to look into the lowest acceptable value of C2. The data sheet says that a low ESR cal is needed to hold down spikes on the input of the high side FET. I think that a formula that scottsai gave me here will tell what's needed but I can't have one too big since higher C2 extends the delay time needed to read the voltage of the unloaded solar panel array. On the other hand, if I run Scot's formula and put in 100KHz switching (planned) I'll get a pretty big uF number. I need to figure something out here...

I don't know what other protection or robustness is normally added to the output side of the driver. In an app note for this particular chip and other circuits, I saw some backward pointing diodes with similar specs to 1n914s so I put some of them in.

Is there something I've forgotten or am ignorant of?

There were concerns about layout. The datasheets show top and bottom high current loops but, honestly, I'm not sure what to do with that info. I've tried to get stuff in the power section close to what it connects to. I don't really know what else to do.

Comments, corrections and feedback is always appreciated.

Thanks again!

 - Ed.

[ghurd]

Hey Ed.

"Clearly, I don't know what I'm doing... "

You are doing pretty well at faking it!

G-

[elt]

Hi ghurd,

I try to learn something new every day. If some of that shows between the top of the page and here, then it's a tribute to the folks like you and others that are so good about helping people. Thank you. I suspect, though, that I'm not at the bottom of this page yet!

Thanks again,

- Ed.

[s4w2099]

I would say this version is much better than the previous idea. I would like to suggest you to always try to use a separate ground plane for the micro joined at a single spot to the power ground. This will make your circuit much more stable by guaranteeing the same ground level at all points in the circuit, also provides some shielding.

This is how mine looks from the top and bottom respectively:

This one is still a work in progress and it might change a bit during the next few days. Hope that helps you on your quest. Click on the pictures to get a bigger version of them.

Right now it has nothing special to it feature wise. It can limit the output current to avoid self destruction and high temperature shutdown. I plan to add a serial port to download a log stored in EEPROM of current and voltage, but for now this is a good test platform to optimize MPPT algorithm. Also I think it would need an external memory for some real useful logging as there is very little memory in this device. Plenty of space left there for new stuff. I should have it in hand by next week hopefully.

[elt]

That's interesting but do I need to go that far with this project? To me, planes are usually a PITA unless you have plated vias. Even in this simple layout, I have a top and bottom connection under an electrolytic capacitor so I'll have to stick a little wire through and solder to the top before I insert the cap. I try to minimize that kind of stuff. If I were sending this to a fab house, then that would be a different matter.

I did, however, made separate digital and analog grounds with a single point connecting them. The bright colors show the top and bottom side ground routing; do you see a problem?

Thank you,

- Ed.

[s4w2099]

Ah yes, if thats the case I would not bother, its too much of a hazzle but it would certainly increase the overall stability. Your mosfet arrangement seems to have much less inductance than mine but I wanted to ask you how are you exactly going to heat sink those?

I thought about doing something like that in the past but for any decent size heat sink it would be a problem. I am aiming for 20-30A output capacity for this version and that would need a more or less decent heat sink even though my mosfets are relatively efficient stp40n20.

Please let us know when you have it working, I am very interested to see how much you gain! nice project.

[elt]

Do I need heatsinks on these FETs?

I'm not confident with "heat." Is the "Thermal resistance junction-ambient Max" spec the one I use? That's 62C/W for this device. I calculate an average dissipation of about 1/4 watt. So if that's how you do it, I'm figuring about a 15C temp rise and I calculated my Rdson based on 90C ... so I was thinking that I wouldn't be needing heatsinks. If I do actually need heatsinks then I can pull the FETs apart a little bit.

But you must be handling a some issues I've been unsure of -

How well does your copper handle the current? At 30 amps the power planes in my diversion controller are "uncomfortable" to touch while the heatsinks on the FETs are only "warm" to touch. Even though the FETs and load should handle 50 amps, I limit in software to 30 amps because I don't want to see copper peeling off the board. The board is one ounce copper and, since then, I've melted solder along the planes after reading the plating helps current carrying. But since I'd like to see 60 amps in the windmill version of this circuit, it's a question I have. And I've never seen 2 or 4 ounce copper from the surplus houses where I normally buy...

Likewise, I'm curious about your connectors. Looks like you'll be using two conductors per connection, that's a nice idea, but what are using for even 15 amps per wire?

Thanks a lot!

- Ed.

[ghurd]

You have heard "Take this with a grain of salt"?

Take this with a 50 pound bag of ice-melt salt.

I have used small tinned wire soldered to the PCB traces to increase the ampacity.

Like solid telephone wire (#22~24), or a couple pieces of it twisted then tinned.

That was always for plain DC.  Not sure of the implications at X-Hz.

Sometimes I solder a strand or 2 of tiny wire on the device legs before it goes into the PCB, to help with the ampacity and heat.

No clue if that has much effect, but it makes me feel better.

I believe in too much heat sink.  Even if it is a copper fender washer, or 2 copper fender washers with a regular sized copper washer between them.

I have seen factory TO-220 devices which looked like a AL washer was pop-rivited on.

It usually falls under the "it can't hurt" heading, but again I am not sure of the implications at X-Hz.

Now go rinse the salt out.

G-

[s4w2099]

Yes, I do exactly as ghurd says. I usually have  my boards made by a fab shop but you can specify to leave off soldermask to those tracks that you would like to reinforce.

I use copper braid from radioshack or add a thin layer of tin to add a few mils of thickness to it. This is not great at high frequencies because of the skin effect, in which case you need to make tracks as wide as possible.

I am with ghurd about the heatsinks too, its better to have too much than too little. It is true that if we have too much heat in the mosfets that would mean that the efficiency pretty much sucks. There are a couple of factors that must be taken into consideration when designing power electronics circuits in this regard. One of the losses that you will have in the switchers is the conduction losses when the device is operating in the saturation region. The other one are the losses associated with the switching frequency. The higher frequency you switch the more losses you will have because you would be spending much more time in the linear region.

http://www.altera.com/support/devices/power/thermal/pow-thermal.html

Thats a link about heatsinking. I am not an expert on it but so far it works for me very well. Do the math and check that for most TO-220 devices dissipating even as little as 1 watt might be a very bad idea without a heatsink.

Also note that there is a derating factor for mosfets. A mosfet operating at 25C can be allowed to dissipate more power than a mosfet that is operating at 100C. And yes, to calculate weather or not you will be able to get away without a heatsink you have to use Tja. If you are sure that all you are going to dissipate is 1/4watt then you might be able to get away without a heatsink. But be very careful when first starting it. I would play it safe just because I dont know how to calculate losses related to switching frequency.

[elt]

I really like ghurd's idea of pop riveting fender washers to the FET's ... that's just the sort of post-apocalyptic approach that I like to take to things!

> I usually have my boards made by a fab shop [...]

Who do you use? I see that Advanced Circuits has sub-$100 pricing for a prototype... I'm very happy sending to China when I want ten or twenty boards but haven't found anything other than home-brew to satisfies my cheapskateness for one or two.

 - Ed.

[s4w2099]

You can try PCBfabexpress.com. They have nice deals. And barebones are very very cheap too. I use Advanced Circuits because I get the student discount, so I only pay 33 bucks per prototype. They also have the cheapest prices Ive found for large quantities and the quality is great.

[elt]

I've never done large quantities but for small quantities I send my stuff to Gold Phoenix... they seem similar to PCfabexpress, which I didn't know about, thank you. Since I most often make small boards, GT appears to have better pricing with its free  with step and repeat but I can see that for larger boards, PCfabexpress advertised pricing would be much cheaper.

Thanks again,

 -Ed.

[s4w2099]

And here is the finished Product, Top and Bottom:

It works great specially when batteries are dead (10V). It does 85% efficiency with a bad diode. I am waiting for the real diode (backordered). It should go up to the 90s efficiency.

I tested it without the output diode (not charging batteries) just with a fixed output voltage feeding a 12V blower motor from a 18Vdc source with the diode removed and it did 92% efficiency.

In one of the pics you can see I left off the soldermask to reinforce those tracks.

I am very pleased with it.

[elt]

That's a great looking board!

Can I ask a few questions?

What does stacking the toroidal cores do?

What about slots for the current sensor pins? One thing that did attract me to Advanced Circuits was their claim to do slots and plated slots; that would allow you to make the pads is in the Allegro data sheet. Thing is that I don't know to specify them in the CAM files and didn't figure I'd ask AC until/if I had a circuit to send them.

 - Ed.

[s4w2099]

Stacking those toroidal cores increase the amount if energy that they can handle without going into saturation. I did it like that because using 2 small toroids stacked vertically would take less space than using a larger diameter toroid.

It is true that they make slots but they cost extra. I just use a big regular plated hole and fill it up with solder. The hole diameter is about the same size as the Allegro's terminal is, so, the copper tracks touch the terminal directly. I made it so the sensor would fit in the holes a little bit tight.

"For additional power handling capability, LI^2, multiple stacking of cores will yield an equivalent multiple power handling for a given core size. For example, double stacking of the 55908 core will result in a doubling of its power handling capability to about 1000 mH-amperes^2." -- Magnetics Inc. Catalog

[elt]

I'm very interested in how your board works out and look forward to more info on your testing.

Okay, I think I saw that slots are free at Advanced Circuits with "standard pricing" ...

Eventually I'll go to more power than what I have now but will likely experiment with multiple bucks running out of phase (I forgot the term  for that) to get the total power up to where it would help cool my mill. BTW: I did move my high side driver to the top side of the board and added a ground plane but I haven't done much more than mount the components because I'm fixated on the idea of using the FET's Rdson as the current sensing shunt being discussed in the boost converter thread. I'll do some buck testing and post after I'm done with the booster version.

Thanks again,

- Ed.