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

I’m just not well enough versed in MOSFETs to determine for certain what properties of the DTU30N02 make it work so well here.

Been testing a few random power mosfets with the 7136 today and while it’s been interesting to see how each behave in the ramp up period I didn’t come across a really good candidate at 4.7khz.

Or indeed a much better idea of what specs are contributing most towards it working better or worse with the 7136 signal.

One particular standout comparison comes with two ON Semi power mosfets, the NTD60N02RG and the T70N03G
Both of these fets came on the old East092 DD drivers, the 70N first and then the 60N on the later models.

These two fets going by the datasheet appear to be very very similar, certainly I can’t really find any major differences in the usual suspects of mosfet specs. Nothing that’s jumping out as a fundamental performance difference between the two.
So how do they perform when hooked up to the 7136?

Well the 60N seems to be about the best performer I’ve tested so far with a pretty fast turn on while the 70N is by far the worst. It takes forever to fully turn on and as a result is completely useless in this application.

Here are the graphs of a 0-100% turn on capture.

NTD60N02RG

This fet actually switches on a fair bit faster than the standard small 30YLD fet I tested last time. It has about a 220uS delay from off to fully on compared with the 30YLD’s ~300uS

T70N03G

The 70N on the other hand is complete garbage, it takes so long to ramp up that the graph almost doesn’t fit in to the 100uS horizontal scaling. Just awful…

-
So what’s causing this drastically different behaviour? No clue…
Total and individual gate charges are very close (and very low at only 9.5 and 13.2nC respectively) , they have identical threshold voltages (between 1v - 2v min-max on both) and capacitances are near as well.

The only major difference I found is that the 60N is apparently “Optimized for High Side Switching Requirements in High−Efficiency DC−DC Converters” and looking in the timing values, almost all the same again bar one; Tr or Rise Time.
Here there is a drastic difference between the 60N which has a Tr of 33ns and the 70N which has a staggeringly low Tr of 1.3ns! Haven’t come across a Tr this low up till now so I initially put it down to a datasheet error…but maybe this is really what makes it perform so poorly here?
I’m just speculating here but maybe having an exceptionally fast rise time is actually detrimental when combined with the 7136’s soft start signal output?

-

Just for comparison here is the 30YLD fet again in the same test setup as above.

PSMN3R0-30YLD

And I also tested two other fets.

A philips PHD50N03LT


Looked promising initially but it’s a fairly slow one…strangely that fast initial start and soft ramp up plot actually seems to give a neater PWM trace on some levels. Non of the funky doubling seen on some modes with the “faster” looking fets like the 30YLD @4.7khz. But still nothing usable in terms of mode range.

DinTek DTU06N03

This looks like a variant of the DTU30N02 but with a higher Threshold voltage (1.2v-2.5v vs 0.6v-1.5v of the 30N02) and double the gate charge (total of 74nC vs ~30). Performance isn’t great unfortunately, had my fingers crossed it was all down to the DinTek name! :wink:

Just putting it out there really, see if you guys can make any connections.
I’m thinking an exceptionally low gate threshold voltage is pretty important, seeing as the ramp up of the 7136 keeps below 1.5v for quite a while. A fet like the DTU30N02 that can start opening up at that stage (<0.6v) will have some advantage over these fets which seem to only start cooking in the 1-1.5v range.

Also not sure what exact effect the gate charges have, I suspect at these slow PWM speeds it’s not that critical. Especially as the two fets with the biggest difference in performance both have a very low charge spec.

Also hooking all of these fets up directly to the Attiny 13A pwm pin they seems to slam on and off almost instantly with the PWM signal, nothing at all like the leisurely pace things operate at under the supervision of the 7136.

Here is a trace showing what the “useless” T70N03 does when driven directly from the MCU. Everything else the same.

Nothing at all slow about that :stuck_out_tongue: In comparison you can see what the open 7136 signal is doing in the comparison.

-

My head is spinning…too many datasheets and still non the wiser… :Sp :weary: :arrow: Bed

Good work. I’d rather ask than read the manual: what method are you using to save screenshots?

There’s a manual? :stuck_out_tongue: I’ve just been randomly pressing all the buttons and twiddling the knobs until stuff started working right :slight_smile:

I just plug a usb stick in and the press the little green printer button below “Help” to save a screenshot. I’m sure you could hook the scope to a pc with the USB cable and capture even easier that way.
But this works ok and I haven’t played with the pc software yet.

Thanks! I thought that the little printer button was for screenshot-to-usb-flash-drive, but I hadn’t pressed it yet. My scope is a fair distance away from the PC, so I’d use Ethernet before I used USB-to-PC.

Speaking of op-amps got me thinking of an alternative solution to the intermediary transistor.
I just tried using one as a signal buffer between the 7136 output and the fet gate and it works as expected now with modes going all the way down to below 30pwm. That’s at 4.7khz with the standard 30yld fet!
Not a real solution to the problem obviously, as you say it’s kind of silly to use any additional component like this to fix an inherent issue with an IC… but…I’m pleased I got something working anyway! :slight_smile:

It seems to me like the soft start behavior of the 7136 can be outwitted to an extent this way. The op-amp input is such high impedance as to basically imitate an open output on the 7136. As a result it ramps up much quicker than with something attached. The op-amp output then handles charging the fet gate and it does that admirably, matching the 7136 ramp up behavior exactly. Timing of that ramp up is now down to around 30uS, very similar to the open output tests I did above.


Channels are same as above, except Dark Blue (Ch4) is now showing the gate drive signal coming out of the Op-amp.

30 PWM (Fairly low low, but not what I’d call a MOON mode yet)

60 PWM

100 PWM

180 PWM

Modes are nicely spaced as you’d expect from those traces. No issues at all, of course they’d also work quite well at a faster pwm frequency just with a bit more mode compression. And a faster fet with a lower threshold voltage would also still help in this regard.

Here is a close up of the trace @ 100pwm

Interesting. Can you post a schematic for what you used / did?

I’d probably get the schematic wrong if I tried to draw it properly, it’s very simple though.

I believe what I’m doing is called a Unity-Gain_Buffer config.

Vin is hooked up to the 7136 output, that’s the signal we want to amplify/buffer
Vout is fed to the gate pin on the fet.

Not shown on that diagram of an op-amp there’s also the Vdd and Vss pins, Vdd in my case connects to the postive output of the 5v LDO that also powers the rest of the driver. And Vss is connected to ground/battery negative.

It’s just functioning as a signal amplifier with the gain set to 1. So Vin = Vout. The very high input impedance of an op-amp fools the 7136 into thinking it’s not driving anything, so the ramp up behavior is very quick.
That’s my best stab at understanding what is going on anyway :slight_smile:


Here’s what the “schematic” looks like in reality, not quite so tidy :smiley:

Oh and don’t mind the resistor and the cap in that mess, they’re not connected to the circuit I tested, just there for tinkering.

Man this thing is touchy!

You know how to improve performance even more beyond just the opamp buffer?
Disconnect the scope probe from the 7136 output!! In this configuration the damn 7136 is sensitive even to the 1Mohm impedance of the oscilloscope inputs! Crazy!

Without the probe attached to the output the ramp up is even faster again and the pwm range at 4.7khz is expanded once more. There’s even a mysterious overshoot happening on the op-amp output that actually helps the mosfet turn on even faster in this configuration. Tbh there’s funky stuff happening with the op-amp that I really don’t understand but it’s fun having a play around. I figured out that adding a 10ohm load resistor to the opamp output helped remove most of the oscillations that occasionally set in. (you can still see some of them on the fet output trace in these images) The overshoot may also be as a response to that load resistor.

With probe hooked up to 7136 output.

And without. Ramp up speed is now down to 20us and the mosfet is playing along nicely.

I forgot to say thanks for pointing me to the details on the Unity-Gain_Buffer config, so thanks for that.

As far as the scope interfering - it does that in the normal configuration as well (no op-amp).

Ah interesting, I hadn’t noticed. That’s strange though isn’t it? Seeing as you’d expect the significant load from the fet gate to negate any minimal interference the scope would have?

It’s strange, but it’s not the only place I’ve seen the (as I understand it) extremely high impedance scope probe mess with the operation of something like this.

I really don’t have a good handle on it.

The observed behavior may be due to capacitive loading (as opposed to resistive loading) at the output of the 7136. Googling the subject yields some good tutorials.

Thanks dthoang, I’ll look into it! That sounds legit.

Alrighty so I’ve done a bit more tinkering and I have a version of this driver operational that fits my needs quite well.

Thought I’d outline the parts and setup in case anyone else sees a use for this driver similar to mine.
And no this won’t work particularly well if you just want to install the 17mm driver into a flashlight. You could probably still stick the opamp in there somewhere but my method here relies on having external fets.

Oh and Wight you may want to look away, because I’ve done terrible, nasty things to your beautifully designed board…there’s even some stacking involved! :open_mouth: :wink:


My requirements where something like this…

-It needed to work visually flicker free with at least 4.7khz pwm and have a usable mode range. 1khz pwm flicker does my head in.
(This works great at 4.7khz and moonmode is now around 20pwm with this setup, it will also work at the standard Phase correct 9.5khz but with a more compressed mode range)

-Had to be able to drive a number of externally mounted mosfets. Also had to work with many fets not just the elusive DTU30N02
(All my tested fets work pretty good with this setup, those with lower threshold voltages still start up faster than others but all are now perfectly usable at 4.7khz)

-Had to run off a 5v LDO supply for >6v 2s input


So I used an Op-amp, specifically the MCP6001 as a unity gain buffer (a different op-amp than before but functionally similar to what I tested above). I don’t know too much about op-amps yet but I tested a few and this one worked very well and was available in a small enough form factor (sot-23).
Not much ringing or oscillations either, looks pretty clean on the scope.

Again not entirely sure why the 7136 behaves this way but it looks to me like using the op-amp as a signal buffer tricks the 7136 into ramping up it’s “soft-start” output much quicker than with a fet directly attached. This allows the driver to work fine with far more mosfets at higher PWM frequencies.

I’m not going to question too much how it works, but I do know it works rather well. :stuck_out_tongue:

-

I used the tiny SOT-23 package version of the op-amp and installed it thusly.


Inch thick layer of flux residue and carpet fluff, probably optional… but you never know :stuck_out_tongue:

And some overlays showing how everything is wired up.

And the back showing the 5v LDO location.

It’s really messy but it works a treat. :slight_smile:
Cheers

Heh, that’s a lot of work in order to use this thing LinusHofmann. Thank you (and pilotdog68) for being guinea pigs on this driver, and thank you for all the work you did analyzing it’s behavior.

Good work nonetheless, and I’m glad you’ve got it into a state that may help you!

I’ve marked the thread title with [retired], the driver is not really workable as-is IMO.

To each his own I guess. I like it in my L5, and I prefer it over stacking a bunch of chips.

Yeah it’s a shame it didn’t work in it’s standard form. I hope you return to design another linear mosfet driver at some point though. I think there’s a niche for a low part count simple linear regulator with DD bypass, just based around something less touchy than that 7136 :slight_smile:

For me, with this workaround it works more than well enough to fill that niche. And I certainly learned a lot testing this driver! So thanks for putting it out there.
Cheers

You’re welcome. I’m glad you both got something useful out of it, and I’m certainly not pulling the files. People can continue to build them if they’d like!

I’m sure I’ll take another look at an FET linear driver, but the parts count will certainly go up. An op-amp based solution will require quite a few components: we’ll lose the 7136, but add an op-amp. In addition we’ll have an RC circuit (so two additional passive components, resistor and capacitor) in order to generate an analog output from the MCU. That analog output will be variable unless we use a Zener or LDO. I suppose we can stick with a Zener and see how it works, a 2v or 2.5v Zener should allow the MCU to maintain consistent analog output down to a very low battery level even with only 1s.

Offhand I think that the falling MCU voltage w/out this solution would cause output levels to drop as battery voltage drops (in a 1s config).

The 17mm board was already pretty tight without two more 0805’s. I don’t think it will all fit with the big spring pad. It might not all fit even with a little spring pad and 0603 sized components.

This is not something I intend to attack immediately.

Thank you for all of your hard work on this.

I'm going more and more to FET drivers for anything over 4A with a XM-L2. The newer ones won't pull much over 5A anyways with the best cells, and with average cells only around 3.5A-4A, so really a 5A linear driver is pretty useless now for an XM-L2. I would say that hopefully they would be useful for the triples, etc., but the led4power driver overheats when used with a triple so I suspect that something like this would as well. I see the merit for the 6V LEDs, but again I wonder if heat will end up being a major issue.

I should note, however for all of the newer guys, that a true constant current driver does have some benefits. No PWM (although the 7136 is sort of a hybrid in this regard) and more efficient at lower drive levels. That said, these aren't benefits that most looking for a 5A flashlight really will notice or care about.

It is amazing to see just how much the vF has gone up. It is really evident now that I have my big power supply and can easily measure the vF of different LEDs. The XM-L is quite a bit lower than any XM-L2, but the newer XM-L2s are definitely higher than the older ones. They can't handle as much current either, with a lot of them having a hard time reaching or breaking the 6A mark. That said, the U3s at 5A are pretty dang close to a U2 at 6A, so efficiency has been improved. I wish that we could have had the increased efficiency while retaining the lower vF. The newer XM-L2s of every bin seem to have a higher vF than the older ones, so it's not just the U3s.