Serious adjustable LDO output drift

[Fulcrum]

Hello everyone.

I have a bit of a problem here with a circuit I have designed. It is a board with several rails, one of which is an adjustable LDO (UA723CDR). The output of this LDO can be adjusted with the help of a trimpot (https://eu.mouser.com/ProductDetail/Bourns/TC33X-2-102E?qs=2hAPiALdQ0nRULMRC6GCtg==). It is set to output 15V nominally.

The problem here is that the output of the LDO will creep upwards as time progresses. It is a rather slow creep (around 50mV per 24h), which is why I did not notice it on my prototypes. I now have 16 units (called ROBs) where the 15V rail will invariably drift more or less north, as you can see in the chart I have attached.

Some of the units are more stable than others, and some units are highly unstable. I am at a loss for what could be causing this. My first thought was that it's temperature dependent, but that turned out to be a false lead.
Could it be the potentiometer that is responsible? Some sort of charge buildup? Any suggestions are welcome. I have been unable to find any answers on my own through googling and books, so I turn to the collective experience of this forum.

[Ian.M]

Is the REF voltage drifting? 
Genuine LM723 chips drift is specified as typ. +/-0.05% over 1000 hours.

[GeorgeOfTheJungle]

Could be R48. Try another brand. Piher are top quality.

[Fulcrum]

Is the REF voltage drifting? 
Genuine LM723 chips drift is specified as typ. +/-0.05% over 1000 hours.

Thanks for the suggestion. I will measure the REF voltage over several days and note any change. I purchased them from Digi-key, so I would assume they're the real deal.

Could be R48. Try another brand. Piher are top quality.

Yes, I am also suspecting the potentiometer.

[MK14]

I'm a bit worried about your enable circuit at the top (somewhat quick look at circuit, so I could easily be mistaken).

What is to stop the (P-Channel) Fets gate being blown by overvoltage ?
(its absolute limit seems to be 20V, but an unregulated 19V supply, is taking it very close to the limit)
EDIT:
It is being limited by the 2K and 5K resistor divider. But if the unregulated 19V voltage, goes high enough, it could risk damaging the Fet.

What is limiting the current into the base of T2 ?

[SiliconWizard]

The µA723 IC is quite an antique chip. 
Is there any specific reason you used it? There are many low-noise regulators that will give you 15V in a much more stable way without any trimming needed and with less passive components...

Anyway.
Apart from the pot itself (a multiturn trimpot may be more stable mechanically-wise), I would also suspect the internal VREF. You could check this easily by monitoring the REF pin if you haven't already.

[Fulcrum]

I'm a bit worried about your enable circuit at the top (somewhat quick look at circuit, so I could easily be mistaken).

What is to stop the (P-Channel) Fets gate being blown by overvoltage ?
(its absolute limit seems to be 20V, but an unregulated 19V supply, is taking it very close to the limit)
EDIT:
It is being limited by the 2K and 5K resistor divider. But if the unregulated 19V voltage, goes high enough, it could risk damaging the Fet.

What is limiting the current into the base of T2 ?

Thank you for pointing these things out! The unregulated 19V input is connected to ground through a 20V zener diode. Besides, it is operated at 17V. The 19V moniker is incorrect.
Current into the base of T2 is limited by a 1k resistor elsewhere on the schematic.

The µA723 IC is quite an antique chip. 
Is there any specific reason you used it? There are many low-noise regulators that will give you 15V in a much more stable way without any trimming needed and with less passive components...

Anyway.
Apart from the pot itself (a multiturn trimpot may be more stable mechanically-wise), I would also suspect the internal VREF. You could check this easily by monitoring the REF pin if you haven't already.

The reason for using this particular chip is because I need to be able to adjust the 15V output in order to counteract variation of a particular component from board to board. The boards must all have highly similar characteristics, which means a bit of fine tuning on the 15V is required... I will definitely monitor the REF pin and see how it develops.

[MK14]

Thank you for pointing these things out! The unregulated 19V input is connected to ground through a 20V zener diode. Besides, it is operated at 17V. The 19V moniker is incorrect.
Current into the base of T2 is limited by a 1k resistor elsewhere on the schematic.

You seem to be saying you are giving it ONLY 17V for a 15 volt output.
But a quick glance at the datasheet, seems to indicate it (LM723) needs a minimum of 3 volts (between the input and output). Datasheet says "Input-to-output voltage differential, VC – VO  (min)3V  (max)38V"

That could explain your drift problem.

[Andreas]

Hello,

a) a LM723 is not a LDO it needs minimum 3 V headroom between VIN and VOUT
    so dependant on your unstabilized 19V it might be out of spec.

b) according to the data sheet the C52 capacitor is usually not at the VREF pin but at the IN+ pin.

with best regards

Andreas

[David Hess]

1. Use an oscilloscope and make sure nothing is oscillating.

2. I am not sure what the 723 reference output will do with 4.7 microfarads directly attached to it but the application notes contraindicate that.  When input power is lost, I suspect it may cause damage to the Darlington emitter follower buffering the reference through base-emitter breakdown.  As shown in the application notes, connecting that capacitor between ground and the non-inverting input may be acceptable if there is at least 1 kilohm between the reference output and capacitor.

3. Using the wiper of the trim potentiometer in that way is potentially dangerous.  Wipers normally fail open and if that happens, the output voltage will increase.  The end of Linear Technology application note 42 discusses this.

4. Measure if the reference is drifting.  Could you have imposter 723s?

[Zero999]

The reason for using this particular chip is because I need to be able to adjust the 15V output in order to counteract variation of a particular component from board to board. The boards must all have highly similar characteristics, which means a bit of fine tuning on the 15V is required... I will definitely monitor the REF pin and see how it develops.

What component is that?

What is the regulator powering?

Perhaps there's a way to design the circuit so it doesn't need an adjustable power supply.

Trimmers are generally avoided in modern electronics, because it normally works out more expensive than using closer tolerance components, in the first place and it reduces the reliability.

[David Hess]

Do not listen to them; there is nothing wrong with using the 723.  It is one of the better fully integrated solutions.

[Zero999]

Do not listen to them; there is nothing wrong with using the 723.  It is one of the better fully integrated solutions.

I agree. It certainly can be easier to use than some low dropout regulators, which can be unstable and will oscillate, when driving certain loads.

I still question the need for trimming: a properly designed circuit shouldn't need a precision power supply rail. There is usually a better way.

[nctnico]

Do not listen to them; there is nothing wrong with using the 723.  It is one of the better fully integrated solutions.

I agree. It certainly can be easier to use than some low dropout regulators, which can be unstable and will oscillate, when driving certain loads.

I still question the need for trimming: a properly designed circuit shouldn't need a precision power supply rail. There is usually a better way.

Indeed. I would have used a high precision reference IF accuracy is required. Trimming sucks. However in this case I would look at the minimum drop-out voltage first.

[David Hess]

Do not listen to them; there is nothing wrong with using the 723.  It is one of the better fully integrated solutions.

I agree. It certainly can be easier to use than some low dropout regulators, which can be unstable and will oscillate, when driving certain loads.

The 723 is not trivial to use either; with external frequency compensation comes great responsibility.  I saw that on a Cracker-Jack box or in a movie or something.

I still question the need for trimming: a properly designed circuit shouldn't need a precision power supply rail. There is usually a better way.

There are some exceptions like legacy compatibility.  Or trimming a reference can be used in lieu of trimming gain.

There are also other ways to trim a reference or regulator without a potentiometer trimmer or automation.  Check Linear Technology application note 42 for some ideas.

Indeed. I would have used a high precision reference IF accuracy is required. Trimming sucks. However in this case I would look at the minimum drop-out voltage first.

So now the circuit needs a reference, error amplifier, and perhaps pass element when the 723 has all of that available in one package at low cost and the 723 is no slouch for temperature coefficient and noise.  There are some combination reference and error amplifier parts but they cost more, are less available, and most do not perform as well.

In the past I never liked the 723 but most of its shortcomings are due to being designed for the lowest cost and early in the history of analog ICs.  It lacks on-chip trimming and the extra complexity of a true constant current mode but Vbe current limiting is no disadvantage in a regulator or reference.  At least it has a real transconductance amplifier.

[Fulcrum]

Thank you for all the comments. I admire the helpful nature of this forum. You never scoff at assisting those with less experience.
I will attempt to answer all the questions and suggestions you came up with, in chronological order. I will snip the quotes as needed to reduce the length of this post.

You seem to be saying you are giving it ONLY 17V for a 15 volt output.
But a quick glance at the datasheet, seems to indicate it (LM723) needs a minimum of 3 volts (between the input and output). Datasheet says "Input-to-output voltage differential, VC – VO  (min)3V  (max)38V"
That could explain your drift problem.

Sorry, that was a typo. It it is definitely operated at 18V, plus a few hundred mV to raise it above the 3V minimum (I just measured it - it's 18.4V). So actually we're almost back to that dangerous 19V.

Hello,
a) a LM723 is not a LDO it needs minimum 3 V headroom between VIN and VOUT
    so dependant on your unstabilized 19V it might be out of spec.
b) according to the data sheet the C52 capacitor is usually not at the VREF pin but at the IN+ pin.

a) See reply to post above this one. The LM723 is operated above 15V+3V. (17V was a typo)
b) You are very right! This actually brought to my attention that the suggested high-voltage regulation circuit does not even have a capacitor on REF nor IN+, see attachment. Could this have damaged the reference circuit? Charging a 4.7uF cap might have drawn more current from the REF pin than it's comfortable with. (Open question to all readers. Input welcome.)

1. Use an oscilloscope and make sure nothing is oscillating.
2. I am not sure what the 723 reference output will do with 4.7 microfarads directly attached to it but the application notes contraindicate that.  When input power is lost, I suspect it may cause damage to the Darlington emitter follower buffering the reference through base-emitter breakdown.  As shown in the application notes, connecting that capacitor between ground and the non-inverting input may be acceptable if there is at least 1 kilohm between the reference output and capacitor.
3. Using the wiper of the trim potentiometer in that way is potentially dangerous.  Wipers normally fail open and if that happens, the output voltage will increase.  The end of Linear Technology application note 42 discusses this.
4. Measure if the reference is drifting.  Could you have imposter 723s?

1) Done and done. Everything's nice and stable. Noise is within spec.
2) You are the second one to point out that the 4.7uF cap could be an issue (see quote above this one). I am beginning to suspect it might be the cause.
3) Thanks for this tip! I did not know about it. Will definitely keep this failure mode in mind for future designs.
4) Already on it

What component is that?
What is the regulator powering?
Perhaps there's a way to design the circuit so it doesn't need an adjustable power supply.
Trimmers are generally avoided in modern electronics, because it normally works out more expensive than using closer tolerance components, in the first place and it reduces the reliability.

The regulator is powering the collector of a common base amplifier. Making the 15V adjustable was a bit of an overengineering error on my part. There are simpler ways to equalize the gain among units. My current wish is simply to get the 15V rails stable, and I'll gladly forgo adjustability at this point. Thank you for the input on the trimmer, I will try to steer clear of them in the future where possible.

[nctnico]

Indeed. I would have used a high precision reference IF accuracy is required. Trimming sucks. However in this case I would look at the minimum drop-out voltage first.

So now the circuit needs a reference, error amplifier, and perhaps pass element when the 723 has all of that available in one package at low cost and the 723 is no slouch for temperature coefficient and noise.  There are some combination reference and error amplifier parts but they cost more, are less available, and most do not perform as well.

I assume there is only one point in the circuit actually needs an accurate voltage (and usually such points don't draw a lot of current). The rest of the circuit may even be supplied from the 17V to 19V input directly. BTW some references can sink & source so an external resistor can be used to supply the bulk of the current and the reference just stabilises the voltage within it's current sinking/sourcing abilities.

[David Hess]

I assume there is only one point in the circuit actually needs an accurate voltage (and usually such points don't draw a lot of current). The rest of the circuit may even be supplied from the 17V to 19V input directly. BTW some references can sink & source so an external resistor can be used to supply the bulk of the current and the reference just stabilises the voltage within it's current sinking/sourcing abilities.

There are all kinds of ways to do it.  I am just pointing out that the 723 is one of the more economical options, probably the most economical, and it used to be very common to see it used as what is essentially an integrated power reference.  Usually a zener diode and a couple transistors is sufficient for amplifier biasing but I have seen the 723 used in biasing circuits for RF amplifiers before.

What is the modern replacement for a 723?  Linear Technology has the LT3081 but it is not economical compared to discrete solutions.  There are some low dropout linear controllers like the ADP3310 but they also cost a lot more to use; the UC3836 is incredibly expensive.

[Fulcrum]

On the suggestion of many of you I logged REF while logging Vout on one of the less well-behaved units. I was unfortunately away for most of the logging period, and was therefore only able to get a few datapoints on REF. See attached plot.

It is not easy to say anything conclusive with so few REF datapoints, but the few datapoints I have do seem to follow the variation on Vout. I believe it's therefore possible to conclude that the referance voltage was either dodgy from the very beginning, or got damaged during operation.

I will replace the LDOs and remove the 4.7uF capacitor on the REF pin.