[retired] [WIP] 20mm single sided & 17mm double-sided ?-amp linear driver - surprisingly good!

I’m pretty thrilled myself (if you couldn’t tell ;)).

The FET definitely needs heatsinking once you go beyond a certain point. I just don’t build enough to know where to draw the line, so I’d say that anything which dissipates over 2W in the FET should involve heatsinking. Maybe a person could “get away” with much more, but why bother? At least throw in some kind of TIM cube like the ones from IOS.

With a set current of 10A your entire resistor bank should be dissipating 0.5W total. I don’t expect a need for heatsinking there. Depending on what you do and how you do it (ie bypassing the 7136 to turn the FET fully on for 10A while using a low set current, such as 3A) you’ll either destroy resistors or need to heatsink them. For most users that won’t be much of an issue I think - just don’t do a bypass and there’s no problem up to around 20A.

Right now the 17mm version has a trace for the bypass: note the trace that goes from Pin5 of the MCU down to the 7136 and then unbroken over to the Gate pin on the FET. Using this trace requires a customized version of the firmware which puts Pin5 into “high impedance” mode during normal operation and then pulls it “high” for bypass operation. I expect to either:

  • eliminate this trace (that sucks for the people that wanted to use it)
  • implement a solder jumper (these are tight quarters for a solder jumper… but I could eliminate some of the LED+ pad to make space)
  • implement a custom ATtiny13A footprint with solder mask completely covering the pad for Pin5. This would allow most folks to solder the IC in place with no connection to Pin5, allowing the use of normal firmwares normal firmwares with a prescaler or divider of 1:8. Users who wanted to use a bypass firmware could just scrape the pad before reflowing, no big deal.

I’m not really happy with any of these ideas at the moment…

I’m not sure which of the things I mentioned above is the problem. After reviewing the datasheets some more last night I’m leaning towards turn-on-delay-time as the culprit.

I’ve been using an older STAR_momentary firmware without dual-PWM support. That made the code less complex for me to work on when implementing a rough, testing-only, bypass mode (which I appear to have broken recently). I think it will be interesting to see how the driver responds to Dual-PWM. Dual-PWM should provide ~1.2kHz Phase Correct and ~2.4kHz Fast PWM. Frankly it may not do anything useful, probably it would just compress all the modes even higher into the PWM range. Not really a problem, just don’t use Fast PWM if it’s not useful.

I’ve been wanting to play with this in SRK/M6 size, but I pushed that to the back of my mind since the driver wasn’t working. Now that it operates, I took a couple of hours and laid something out.

  • 47mm
  • MCU on the bottom. I was inspired by Mike C’s - Mod: My SupFire M6 “BMF” edition (new beamshots in OP) and the way Mike C made the MCU available for flashing.
  • LDO optional. An LDO will work fine for all applications, but it costs slightly more than a diode. You can just solder on C1 and solder the diode between C1 and the empty C2 pads.
  • 1.3mm LED+ vias will take 18AWG wire from the careful modder, IIRC.
  • Extra pad on top for Pin 6 (extra PWM pin, temp sensor, or status LED)
  • Extra pad on top for Pin 3 (probably for a temp sensor or status LED)
  • OTC in case you are somehow making a clicky light…
  • I’m aware the MCU GND needs to be moved. This will not be a problem.

The optional LDO is present for e-switch MT-G2, MK-R, XHP50, XHP70, etc. With the current bottom it’s intended for 8x18350 (2s4p) or similar. I think it’s apparent that the bottom copper could be easily changed to be like Mike C’s layout.

Darn this looks really good, now I’m considering rebuilding my BTU Mtg2 around these drivers. They’d be a much neater solution, and I’d imagine less thermal issues since sinking 3 large components with handy pads on the bottom is much easier/more efficient than doing the same to 48 triple stacked 7135s.

Can you put the final nail in the coffin for me by confirming that actual voltage dropout is lower or at least equal to an equivalent 7135 based board? :stuck_out_tongue:

Cheers, and keep up the great work!
Linus

Thanks!

I only did a very small test, but Vdo seems to climb with set current for some reason. The Vdo from a bank of 7135s seemed to quickly eclipse that of the 7136-based driver as I increased the drive current (by adding 7135’s and increasing the sense resistance, respectively). I’m not sure why it worked that way. I think I wrote a very short post earlier in this thread about it.

I’m not sure about the “handy pads” you mentioned. The large pad is electrically connected to LED-, so it must be isolated from GND. If you just meant the large flat area on the top of the FET then carry on: I agree. One issue is that these FET’s are so slim that the QX7136 and ATtiny13A are both taller IIRC. Not the end of the world, there are various ways to deal with that.

Ah whoops. :frowning: I wasn’t paying attention and assumed the large pad underneath the fet was for ground/heatsink. Had thoughts of mounting those fets remotely and soldered directly to a copper insert or something in the driver cavity.
That makes it a little less ideal than I thought, I know from dealing with the 7135s that heat sinks making contact with just the top of the chip are next to useless in keeping the chip from going critical.

But again this is a massively bigger package so I’m keen to have a play around with this driver and see how it deals with the situation thermally.
Cheers

It may be useful to the wiring diagram M6.
IC U2 is something like: The ADM803 supervisory circuit monitors the power supply voltage
My M6 mod. v1 by RMM, works very well.

Technically the FET manufacturers expect you to heatsink these things through large pours on the PCB, but I think it’s reasonably common for that to not happen. I’m not too concerned about it, even at high power, as long as there’s heatsink contact somewhere on the FET package. If you look at the construction diagram (page 4) of an LFPAK56 MOSFET, it’s pretty serious in there. Big copper slabs on the inside. With the ones we are using being rated at 50-150W of dissipation (depending on the model), I think some fudging is possible.

What specifically are you referring to when you say “with the 7135s that heat sinks making contact with just the top of the chip are next to useless in keeping the chip from going critical. ” ?

Good schematic making skills aslpg!

[quote=wight]

Warning: I’m plugging in my Waffle Iron! :stuck_out_tongue:

-

Well if you expect a 7135 to dissipate quite a lot and it isn’t soldered directly to a copper pcb area then it won’t be long until the temperature get’s way too hot at the chip level and funky things start happening (i.e going critical :slight_smile: ). I came across this particularly badly in my BTU triple mt-g2 build where the 7135 chips at the top of the double stacks were getting particularly toasty and resulted in very unstable output/flicker not long after start up. These were the chips only connected to the heatsink/ground trace through a thin solder bridge between the two ground tabs.
In that case I tried having the top of the 7135 package in contact with a heatsink surface but that didn’t make much of a difference at all. I actually noticed that the grounding tab would be almost too hot to touch before the case of the chip was even warm.
I was just saying, that to shift a lot of heat it doesn’t make much sense to me to use the top surface of the device to do that. But maybe it’s sufficient here?

As an example in my mt-g2 light driven at 6A (with ~7v vF) from a relatively sag free 8.2-8.4v input, I believe thats over 7watts of heat dissipated per 17mm sized board, and in this case it’s all through a single component and into an electrically isolated copper trace that can’t be in direct contact with the flashlight host and to the outside of the light. Are these mosfets happy running at very high temperatures, like 100degC and up with no ill effects? I have no idea on this stuff.

I do know that with the 7135s as the temperature rises above a certain point output seems to drop off quite rapidly below the rated 350ma/380ma. Some kind of active throttling or just performance deterioration I don’t really know. If it get’s really bad then other interesting stuff like flickering and stuttering sets in. If you connect a 7135 up without any heatsink contact on the ground tab and run it for a bit you should see what I mean.

Maybe it’s not the tiny fet inside the chip that’s suffering under the heat but some of the other sense components that are also in there.
A problem solved by separating the sense parts of the circuit and the hot mosfet as with this driver design perhaps?

-

I can definitely see from studying your mosfet package that having the internals laid out like that is directly linked to the performance of the fet, both in terms of thermal and electrical resistance.
For my application in particular, but maybe for linear flashlight drivers in general a power mosfet with the large tab as the ground/source like the 7135 would seem like a far better option. Ultimately the body of the light is the primary heatsink and if the mosfet is electrically and thermally isolated from that it’s not ideal. Does something like this even exist though?

This fet may be able to dissipate silly amounts of heat like 50-150W but surely that’s not a realistic figure in this case and heavily dependent on an ideal heatsink path?

I’d be really interested in seeing a test showing how this driver responds when it has to dissipate ~3-5Watts for an extended amount of time. Seems that would be a useful power dissipation range to look at for driving an mt-g2 at over 5 amps.

I got stung by unexpected thermal issues with those 7135s so I’m probably a bit hyper sensitive to the issue at the moment :wink:

Cheers
Linus

Issues with the 7135’s when they are treated like that are pretty well known….

The MOSFETs are happy at extremely high temps. For example, the NXP PSMN3R0-30YLD should be derated to 20% of it’s full power dissipation rating when the tab is at 140°C (so that gives us 18.2W of dissipation after derating from 91W). Your PCB should be kept under 120°C though I think, or risk delamination?

The more direct your thermal path, the better. N-channel has been king for a while and all n-channel have this heatsinking arrangement as far as I know. I’m sure heatsinking to GND would be more convenient for our application, but that’s just how it is. It’s not as convenient, but we don’t get a choice about it. If you don’t want to heatsink the top of the device you don’t have to, you just can’t connect LED- to GND so you’ll need electrical isolation. I recommended potting earlier in the thread and I’m recommending it again now. It’s also quite possible to use a PCB like I posted in Post #32 and just electrically isolate it from the heatsink. That’ll work effectively to allow the heat to move into the heatsink.

Don’t look to me to do a lot of testing. Sorry.

That sounds promising and hopefully output will be stable up to a higher temperature compared with a 7135 board.
I have some of these boards and parts on order to test this kind of thing, was just wondering what experience you had with these fets at high dissipation levels.

Potting is something I try to avoid on my own builds just for ease of tweaking and maintenance, but if that’s the only way to get the heat out then so be it :slight_smile:
Cheers

I’m sure you could switch to TO220 with an insulator. That’s a bolt together solution. I’ve been ignoring TO220 because it’s huge compared to PowerSO8 / LFPAK56. Development probably has not kept pace, but I’m not really certain how much that matters with a linear driver like this.

TO220 - https://www.google.com/search?q=to220&es_sm=93&source=lnms&tbm=isch
TO220 insulator - to220 insulator - Google Search

Yep that looks like the kind of thing I was dreaming about, that form factor with source connected through the heatsink pad and I’d be in heaven. How many millions do I have to order to get a custom mosfet spun? :bigsmile: :wink:

I actually have some silpads lying around in to-220 format salvaged from some old 7812 regulators, so I could experiment with those.
Problem is, when it comes to picking a mosfet I have no clue. Is there like a range of specs that you look for which are most critical in a driver like this?

Cheers
Linus

I think a MOSFET of the type you desire is like Love - some things can’t be bought. :~

As for MOSFET selection from what we can buy: Start with the specs you do understand, those will narrow it down a lot. See what you come up with by focusing on these things:

  • I’d stick to 20-30v rated stuff
  • high current rating
  • high dissipation rating (of course)
  • a fairly low gate threshold voltage since it will be powered by no more than whatever the MCU supplies to the QX7136. This will be spec’d as Vgs(th) / gate-source threshold voltage, etc. Note that the spec’d number doesn’t necessarily apply to our case since it’s for something like 1mA or similar. Look for something which has graphs indicating good performance down to low Vgs.
  • gate charge - don’t let it be too high. Look at the other selections we’ve been discussing, but I think you should be fine with <50nC or <100nC or something.
  • “times” may matter, this has yet to be nailed down (see this discussion thread). turn-on delay time, rise time, fall time, turn-off delay time, etc.

Remember, both NXP PSMN3R0-30YLD and Vishay 70N02 failed to work with a standard firmware, but slowing the PWM down to 1khz (or whatever I said above) seemed to get them both working just fine.

Pinning down your voltages, gate charge, and current rating should quickly narrow the field.

Awesome thanks so much, that’s exactly the kind of explanation I was looking for! :slight_smile:

Don’t know if this helps you, I was just lurcking around datasheets of led-CC modules since I plan on modding my just purchased New117/TK35 alike driver.

Since the QX7136 datasheet doesn’t offer much info, I thought I’d share that find.

The somehow related QX9920(those with LEDA on top, same manufature) datasheet is somewhat more helpfull with stats, wich would be nice, but I can’t read Chineese. But it’s possible to squeeze out some stats by figuring it’s relations, since the numbers and units are latin digits.
It states a T_off time of 621ns in the sheet, that would -equal to a frequency of 1610.31khz -wich would just about equal to the frequency range you’d find out works good.
Might be a coincidence but I don’t think so.

nice work you do here, thanks a lot. :slight_smile:

EDIT: Crap, just turned on my brain. Was out of caffeine… T_off on QX9920 relates to the offtime memory cap. Nevermind still great work you do here.

T_off for the QX9920 (which is a step-down / buck controller) is supposed to refer to the fixed off-cycle time it uses. In other words it turns OFF for a fixed period of time, then turns ON for a variable period of time (approximately until the set current is reached), then repeats.

Whether the freq is related I’ll have to think about. :slight_smile:

On a driver like the A17DD-S08 would it be feasible to make a copper “heat sink” that attaches to the gate and wraps around to sit on top of the FET? Glued on with Arctic Alumina Thermal Adhesive as an isolation? This would be “live” I know, but as long as it wasn’t touching anything else it would be acting to dissipate heat from the MOSFET, right?

In some of the lights I use this driver in there is plenty of room in the pill for up to 3/4” tall sink. I can see it being possible to make quite a large heat sink in copper to connect to the gate and the LED negative lead could actually connect at the top of the sink as it would be live.

Would this do any good for pulling heat away from the MOSFET and keeping it nice and stable? Some of the triple and quad builds I’ve been doing are seeing in excess of 14A.

You mean attaches to the drain? The large heatsink pad on the bottom is the drain, the gate would be a pin equivalent of the pwm input on a 7135.

It’s that age old argument of having thermal mass/internal heatsinks vs an efficient thermal path to the outside of the light isn’t it.
I went for more of the thermal mass approach on my BTU build (even though I don’t actually have to worry about isolating the 7135s electrically) and neglected maximizing the thermal path to the outside of the light. Since that is the only area that can actually get rid of the heat generate from everything inside I was simply filling up the internal heatsink and when it was full the temps skyrocketed. The fact that this mosfet is always going to need an electrical and as a result also somewhat of a thermal isolation between it’s heatsink pad and the rest of the light is my primary concern.
It may well be an unfounded concern though. I have a bunch of parts on order to put this stuff to the test! :slight_smile:

I realize my application is somewhat extreme but if you’re dumping up to 15watts of heat into any internal heatsink that doesn’t have a great connection to the dissipating surface you’re probably not going to have a very long running, stable driver.

Thinking about it again now, I’m not really using the body of the flashlight as a ground conductor, I could probably isolate the “pill” from ground without too much hassle. That would certainly make heatsinking the fets more straight forward.

-

Wight if I have found a mosfet in TO-220AB format that hits or exceeds all the other criteria but has a “Qg” total gate charge of 212nC @ 10v, is it not even worth trying?
I’ll keep looking for something that hits all the requirements but I guess I’m not too clear how gate charge affects the situation, does it take longer to “fill” by the MCU pwm signal before the fet is fully “On” or am I looking at that completely backwards?

On a related note, seeing as there may be quite a few people planning to run this driver in a 2s configuration. Wouldn’t a zener diode rated higher than 4.3v (but obviously lower than 6v) help in any of this stuff. Certainly looking at Rds On for these fets it seems a higher voltage on the gate is beneficial, wouldn’t that also help the rest of the circuit run at a higher frequency or allow for less specific fet requirements perhaps? Just thinking out loud here…

At 212nC I think you're going to have a hard time switching it at fast frequencies direct from the attiny like we usually do.

I built a light bar where I was driving 2x external TO-220 FETs mounted on the external LED heatsink at 36V and quite a few amps. On that setup I was running a ~5.5V zener diode because I figured I needed all of the voltage overhead I could get to switch those FETs and keep them turned on fully when needed. The zener diode was obviously very inefficient stepping down from 36V, but an extra 0.5 watt of loss wasn't a concern in that application, and it was a prototype setup anyways. For a momentary setup, a zener diode is obviously not the best solution.

Cheers Richard, I’ll keep looking for a better mosfet.
Good to get confirmation that higher voltages aren’t a bad idea in this type of thing, I’ll make sure to order in some 5.5v zeners just in case.