LT3080 wierdness - dave's power supply (eev224) gone mad

[C]

amspire

I think I under stand, but I could also be blowing smoke.

I think that the problem is Looking at the circuit we are seeing one thing and it's really acting like something else.
In the very quick look at the data sheet examples there is always a resistor connected to the set pin, Except the example on page 16 where they use logic to activate a fet across the set resistor and at the same time the output.

If the resistor is not a resistor, what is it? Looking at the data sheet that resistor is driven with a constant current source.  Constant current source + fixed resistor  equals Voltage reference.
I am thinking that changing the Voltage reference too fast blows the chip.

Make sense?
C
 

 

[amspire]

amspire

I think I under stand, but I could also be blowing smoke.

I think that the problem is Looking at the circuit we are seeing one thing and it's really acting like something else.
In the very quick look at the data sheet examples there is always a resistor connected to the set pin, Except the example on page 16 where they use logic to activate a fet across the set resistor and at the same time the output.

If you need to limit the current into the SET pin with a resistor, then the Absolute specs should state this instead of the +/-0.3V spec.

But the data sheet does have circuits with capacitors on the set pin, and thy behave just like a low impedance voltage source under transient conditions.

If the resistor is not a resistor, what is it? Looking at the data sheet that resistor is driven with a constant current source.  Constant current source + fixed resistor  equals Voltage reference.
I am thinking that changing the Voltage reference too fast blows the chip.

Make sense?
C

I would love to know from the LT designers of the chip what they intended, rather then what the datasheet writer thinks.

I have written back to LT mentioning the forum, and we will see what they come up with. I did mention to the LT contact that I suspect the LT designers of the LT3080 would laugh at the +/- 0.3V maximum spec for the SET to OUT Absolute Maximum. See if he actually checks.

[EEVblog]

I'm starting to get the heebie-jeebies anyway. I'm afraid that I'll make a thousand of these things and XX% will blow the snot out of themselves 
When my one blew, I'm pretty sure I didn't do anything wrong to do that, so that's worrying right there, as these things are supposed to be short circuit proof etc.
Perhaps the old stalwart LM317 might be in order...

[BravoV]

I'm starting to get the heebie-jeebies anyway. I'm afraid that I'll make a thousand of these things and XX% will blow the snot out of themselves 
When my one blew, I'm pretty sure I didn't do anything wrong to do that, so that's worrying right there, as these things are supposed to be short circuit proof etc.
Perhaps the old stalwart LM317 might be in order...

Remembered these new LDO advertising from Linear when they released them .

Not your Dad's ... 

[amspire]

I'm starting to get the heebie-jeebies anyway. I'm afraid that I'll make a thousand of these things and XX% will blow the snot out of themselves 
When my one blew, I'm pretty sure I didn't do anything wrong to do that, so that's worrying right there, as these things are supposed to be short circuit proof etc.
Perhaps the old stalwart LM317 might be in order...

The major fear I have is that if the difference in specs are true, then that would imply they made a mistake with the SET pin circuit in the LT3080 design that they fixed in the LT3083 and LT3085.

Then for some reason, they never redesigned the LT3080.

It would really help to know the full story.

[EEVblog]

The major fear I have is that if the difference in specs are true, then that would imply they made a mistake with the SET pin circuit in the LT3080 design that they fixed in the LT3083 and LT3085.
Then for some reason, they never redesigned the LT3080.
It would really help to know the full story.

And here I am, trying to get a design finalised and tested, with potentially some big issues hanging over the whole thing.
So I have the options:
a) sit back and wait for an official investigation and response from LT
b) sit back and wait and see if someone else sorts it out
c) investigate myself, which I'm sure won't be an easy task, and I'd need stock of expensive LT3080's
d) Abandon the LT3080 now for the LM317 (with extra neg offset circuit)

And options b-c will still require a response from LT before I'd be happy to trust it.
I wouldn't have believed it if I hadn't killed one myself, and maybe two, I'm not sure yet.

[amspire]

Ah! the plot thickens.

The LT3085 does have a modification on the SET pin that the LT3080 does not have.

Take a look at Note: 7 on both datasheets:



It looks like the LT3080 has the SET pin clamped to the output internally with diodes?

Can anyone test this? it would be interesting to see if these are Schottky or conventional diodes, and if there are back to back diodes.

The LT3085 adds 1k resistors (plural).

This almost certainly means the SET pin goes to two 1k resistors - one of which goes to the opamp input, and one goes to the 10uA current source. Then both the opamp and current source circuits are individually diode clamped so they cannot go more then a diode drop negative of the OUT pin. Using two 1K resistors eliminates any loss of regulation from the addition of the resistors, and means the SET pin can now safely go higher or lower then the OUT pin by +/- 10V.

It would be nice if you could add the same circuit externally to the LT3080, but you cannot as you cannot separate the current source from the opamp input.

If the LT3080 does have back to back diodes internally to OUT, then the really important spec becomes the +/- 10mA limit on the SET pin current. If you have, say, 10V on the SET pin and the output gets shorted, your circuit has to ensure that the maximum current into the SET pin is less then 10mA. Same story if the load actively drags the OUT pin up to 20V. Failure to ensure this in the design could lead to chip latchup.

Here is the circuit that LT meant to put into the LT3080 datasheet:



(Note to LT - you have my permission to this image directly into an updated datasheet.)

And the +/- 0.3V ? If they use internal schottky diodes, then I guess +/- 10mA may equal +/- 0.3 volts. Definitely if SET gets to one standard diode drop below the substrate (OUT Pin I guess) on the LT3080 and the LT3080-1, latch up could probably occur. The LT3083 and LT3085 would need the SET pin to be dragged over 10V below OUT for this to be a risk.

A BIG problem with their datasheet is that they allow for a 1uF cap to be put on the SET pin to reduce noise. If the output gets shorted, there is no possible way to keep the SET pin current below 10mA with a 1uf cap directly on the SET pin.  FAIL!!!!!

So can someone do the test on a LT3080 to see what is between SET and OUT?

Richard

[EEVblog]

So can someone do the test on a LT3080 to see what is between SET and OUT?

0.80V both ways on a Fluke 87

[EEVblog]

A BIG problem with their datasheet is that they allow for a 1uF cap to be put on the SET pin to reduce noise. If the output gets shorted, there is no possible way to keep the SET pin current below 10mA with a 1uf cap directly on the SET pin.  FAIL!!!!!

No kidding!
I think it's a short on the output that killed mine. My design has >10uF on the set pin.
Even the extra sneaky 1K series resistor might not be enough to prevent this.

[BravoV]

The forward voltage looks like silicon type instead of schottky diodes. 

Tested both directions between pin 2 (SET) and 3 (OUT) and both have similar 0.7xx forward voltage and tested 3 of them, brand new never used.

[EEVblog]

0.82V both ways on the U1272A, for both SOT223 and TO-220 packages

[amspire]

A BIG problem with their datasheet is that they allow for a 1uF cap to be put on the SET pin to reduce noise. If the output gets shorted, there is no possible way to keep the SET pin current below 10mA with a 1uf cap directly on the SET pin.  FAIL!!!!!

No kidding!
I think it's a short on the output that killed mine. My design has >10uF on the set pin.
Even the extra sneaky 1K series resistor might not be enough to prevent this.

Looks like the only cure is to add external back to back schottky diodes from SET to OUT, and then see how that affects your circuit.

I have passed this on to my new mate Steve at Linear Technology. See what he says.

I have suggested that they need an App note on how to use the LT3080 reliably, and that the LT3080 really does need the LT3085 SET pin mode fix to partially fix the problem.

[amspire]

If anyone feels brave, the next question is what voltages and currents in or out of SET cause latchup? If you put a 10K resistor from the IN/VControl to +5V, OUT to 0V, and then feed positive and negative current into SET, you hopefully can find out when latchup occurs without damaging the IC.

Can it take 10mA into the SET pin, or does latchup happen at a much lower current?

[EEVblog]

Looks like the only cure is to add external back to back schottky diodes from SET to OUT, and then see how that affects your circuit.

That's still not going to fix the 10mA SET pin limit problem though 

[mcbmcb0]

I'm glad the clever people have taken interest in this thread. Weird also how Dave gets 0.76 V Vset-Vout and I got 1.0 and 1.3V. 

When you say 'latch up' Richard are you referring to how when  driving Vset with sudden voltage changes to 0V i then got 2V between Vset and Vout  - at least until Vset was increase to rail and back down again (yikes! was that part of the problem, being >0.3V above Vout?) or switched off then on.

Has anyone else out there found this, or, if agreeing to Richards challenge, willing to test it?

Meanwhile I've possibly foolishly order some more of these things, tho to ensure a different batch these are SOT223 package from a different supplier that I'll only use with a current boosting transistor. And i might need to get some schottkys.  I did write to element14 about the weird TO220 ones but no reply so far.

looking forward to any more thoughts you all have

Mike

[amspire]

Looks like the only cure is to add external back to back schottky diodes from SET to OUT, and then see how that affects your circuit.

That's still not going to fix the 10mA SET pin limit problem though 

It should. If the internal diodes are conventional diodes, then external back-to-back schottky's from SET to OUT will prevent the internal diodes from turning ON so the maximum SET current will be 10uA. It should make the LT3080 reliable.

If the SET or OUT can ever get above the IN/Vcontrol voltage, you may need an extra Schottky from the SET and OUT back to the IN/Vcontrol.

[airiclenz]

Hello,

I had similar issues with a LT3083. The V-Out was always way above the V-set. I needed to go up to approximately 14V with the set voltage until the V-Out "snapped in" to the V-set. After that I was able to go back down to voltages below 1V.

I found the following in the datasheet (which was also mentioned by Dave in one of the videos in his PSU series):

A fixed current of 10µA flows out of this pin through a single external resistor, which programs the output voltage of the device.

I tested to providing a path from the V-set pin (over some resistors) to GND. This fixed it. I don't see this path in your original schematic (only if Q7 is open).

Could it be that?

I found another problem in the original schematic from Dave. The 100 ohm resistor that sets the current source (R70 in your schematic) might burn when high voltages (>15V out) are applied. I killed 2 of them and changed the design to an op-amp comparator between V-out and V-set which toggles a power transitor to GND over a 10 ohm power-resistor.

Airic

[EEVblog]

It should. If the internal diodes are conventional diodes, then external back-to-back schottky's from SET to OUT will prevent the internal diodes from turning ON so the maximum SET current will be 10uA. It should make the LT3080 reliable.

Of course. Must engage brain before typing...

[mcbmcb0]

Airic
thanks for your post.
i had wondered if the Vset current was too low with the op amp near 0V and trying to sink thru 1100Ohms. I did try a resistor of some value (can't recall what) tho dumped it as i lost too many volts at Vout (as it divided the op amp out).  I also thought  the op amp out even at 350mV/1100Ohm = 30uA-ish, and thru the transistor 100ohms should sink some current. Tho maybe I'm mixing up sinking and sourcing - while the spec sheet says you can actively drive it, its vague on exactly how. Guess i should have tried something bigger - 1Meg-ish - what value did you use? I'll give this another try.

I can't quite follow your comparator bit - any chance you can re-explain?

thanks


Mike

[airiclenz]

Oh, I  did the divider solution too but the used resistor was far away from mega-ohms. If I remember correctly, it was something like 100k. I am aware that this introduces other problems because v-set is not longer the voltage you really want. This could be compensated but it's not optimal though. Anyway - if this fixes your strange behavior, you have a new track to follow and finally, hopefully fix the problem. It might be it but can also be a problem with the LT3080. I thought it's worth a try.

I'll post / send you a schematic of my "comparator-pull-down" later today.
Edit: here it is.

Airic

[C]

I guess I am just thinking different.

How would you create a cheap  to use regulator for on board  use for 1.567 volts?

Now make it a sample manufacture step to make it work for 1.897 volts and 4.568 volts and x.xxx Volts.
O by the way, It would be nice to have it simple to put many regulators all over the board to divide the load and keep the voltage spot on.

-------
 

To get to those numbers you need a cheap voltage reference.
The cheapest part is a resistor.
In the data sheet they talk of resistor Noise causing problems to the voltage reference and the cure a cap.
They talk about a guard ring on the voltage reference as they are using such small currents to create the voltage reference.

If you had one of those High dollar very accurate voltage reference chips, would you connect the OUTPUT of an op amp to the voltage reference's output?

Also, Why would you connect an error feed back loop to a voltage reference? 

The data sheet also talks of the benefits of current sensing this voltage reference.
A Current op amp has very different specifications from a Voltage op amp.

Most of us have used one type of op amp for a long time. See an op amp and tack that Voltage handle on. Treating a current op amp as if it were a voltage op amp is sure to get trouble.

Most of the voltage regulators are built around a Voltage op amp. What would a voltage regulator built around a Current op amp look like, something like this?

C

[SeanB]

I would guess that the 10mA max current does not cause latch up, but that at a current probably in the range of 20 to 100mA ( depends on processing as to the actual gain of the parasitic SCR in the die) it latches up, conecting input, output and set together in a SCR, and staying latched until either the die overheats, a bond wire burns out or the power is cycled. Note that in this case the overtemp protection is not operative, as the device is no longer being controlled, it is only a SCR.

[mcbmcb0]

richard and dave
some more on this - would not adding 2x back to back schottky diodes between set and out mean a large capacitive load could dump a whole lot of current back into whatever is driving the set pin? like the op amp or current limiting transistor?
i guess if they have internal diodes this could happen anyway, tho if adding schottkys perhaps they should have 1k resistors in series to current limit.

i tested my 2 weird LT3080s with a semiconductor tester between set and out. the one that operated with 1V between set and out shows a diode junction of 3.3V with cathode on set. the other (which operated with 1.3V between set and out) shows as no device (and resistance of approx 2Mohm between these pins, compared with 700/1700 Ohms). Note they both still kind of regulate voltage, in between latching up. I'm now wondering if I wounded them by raising set too high above out - set was driven to near rail by the opamp (to about 35V, thru the 1100 Ohms), whereas Vout is obviously at least 1.35V lower. they must be pretty low power diodes in there.

come to think of it, maybe i can bring mine weird ones back to 'normal' by external back to back schottkys??? and maybe i'll just drive it with a pot, like in the olden days.

Good luck Dave if you are going to sell these as kits!

[amspire]

I would guess that the 10mA max current does not cause latch up, but that at a current probably in the range of 20 to 100mA ( depends on processing as to the actual gain of the parasitic SCR in the die) it latches up, conecting input, output and set together in a SCR, and staying latched until either the die overheats, a bond wire burns out or the power is cycled. Note that in this case the overtemp protection is not operative, as the device is no longer being controlled, it is only a SCR.

That is my guess to, but allowing 10mA to flow does violate the absolute maximun SET to OUT voltage rating that LT says is correct..

If the back-to-back diodes between SET and OUT are conventional diodes, and the Absolute Maximum SET to OUT voltage is +/-0.3V that implies you must never let these diodes turn on.

But then they also specify a maximum SET current of +/-10mA. If you have 10mA flowing, the SET to OUT voltage is greater then 0.7V - over double the previous maximum spec.

If there is a risk of latch up, it will be when SET is more negative then OUT, but probably not when it is more positive. They do not say this anywhere though.

I have asked LT for clarification of their previous email that I couldn't understand, but no replies yet. In their email reply, they said the +/- 0.3V Absolute Maximum was correct, and then in the next line, they said the Ground Clamp circuits which lets SET get below OUT by one 1N4148 voltage drop is also correct. ??

So for now, the only safe way to stay in spec is to have two big back-to-back schottkys from SET to OUT and then see how that affects the rest of the design.

I am certain that the LT designers actually considered this problem pretty carefully, but whatever safe operational rules they built into the design never made it to the datasheet. At least one of the back to back diodes from SET to OUT was deliberately added (the other diode may be the substrate diode) and so it was intended to be used. To me that implies that the SET can go to 0.7V above the OUT pin which would mean the +/- 0.3V Absolute Spec is wrong.  But this is just my guess. I do not have a LT3080 myself at the moment to do some tests.

To me, the safer tactic is to use either the LT3083 or LT3085 and for the moment stay away from the LT3080 totally.

With the LT3083 and LT3085, at least you know you have to keep SET to OUT to less then 10V, and the datasheets do make total sense. It is pretty clear that the designers got the LT3080 SET pin design wrong and they fixed it in the LT3083 and LT3085, but at the moment, there is no indication that they ever will fix the LT3080 design. If they did, it would probably have to be assigned a new part number.

[EEVblog]

To me, the safer tactic is to use either the LT3083 or LT3085 and for the moment stay away from the LT3080 totally.

Yes, that appears to be the only conclusion I can come to as well.
I only wish one of mine had failed way back, and I would have investigated this much earlier.
But I trusted the example circuit, and it tested ok (by luck I guess), so it looked like a goer.

Fo me, the LT3083 is too expensive, and the LT3085 is too low current. So it's back to the LM317 I go I guess