LTC6655B long term drift

[alex-sh]

Hello all,


I have a small pcb with a voltage vref LTC6655B. Unfortunately, the long-term drift specified only for LS8 is 20ppm/sqr(1000). This means a drift os 81uV. I have MSOP package though. I measured the vref on 11.02.2019 and it was 4.097200v with my 6-1/2 DDM (recently calibrated), which was already outside 0.025% max accuracy
Right now, it is 4.097784V. It is hardly been used in between the dates. This means the vref drifted 584uV. This is a huge drift in my view. Any idea why the LTC6655 drifted that much? Any recommendatins? Maybe I should have burn it for several months for long-term stability?


Thanks
Alex

[ycui7]

I would suggest to run the reference continuously for a few days first, then compare the results.

I have seen this reference drifted quite a bit in the first few days then settled to a few ppm level variation within day (mostly due to room temp change).

An extreme case for me was that a 4.096V reference from Maxim output 5 V the first time it powered on with very high noise, then settles to 4.096V specification and ppm level noise in a few days. Mostly due to my soldering.

[Andreas]

Unfortunately, the long-term drift specified only for LS8 is 20ppm/sqr(1000).

 Maybe I should have burn it for several months for long-term stability?

Hello,

it has a reason why the (typical) drift is only specified for the hermetically package.
The plastic package absorbs humidity which gives stress to the die and thus changes the output voltage.
If you solder the MSOP8 package it gets dry.
The following days/weeks the package absorbs humidity from the air -> output voltage changes.
After a week or two the package stabilizes (see ADC14).

What do you think happens if you burn  in the plastic package: after the burn in. -> it is useless.

By the way: independant of LS8 package or MSOP package: my samples showed a relative large hysteresis so I do not consider the LTC6655 as a "precision" reference. The LS8 package is better, but only if you mount it dead bug style on the PCB so that there is no influence from the PCB.

With best regards

Andreas

[iMo]

I would highly recommend you to clean up your PCBs from the soldering flux (with IPA) carefully..

[Andreas]

The photo was made before soldering the connectors and before cleaning. 

best regards

Andreas

[alex-sh]

Unfortunately, the long-term drift specified only for LS8 is 20ppm/sqr(1000).

 Maybe I should have burn it for several months for long-term stability?

After a week or two the package stabilizes (see ADC14).
What do you think happens if you burn  in the plastic package: after the burn in. -> it is useless.

Sadly, you are correct. I have been running a few tests myself with several LTC6655 and very disappointed with results.
It is a very low noise vref, but I would not call it precise either. The long term drift is horrible.


Would you recommend any better .096V vref in MSOP package please?

[alex-sh]

I would highly recommend you to clean up your PCBs from the soldering flux (with IPA) carefully..

Not sure if it was addressed to me or Andreas, I have cleaned up my PCB several times just to rule out this issue.

[Andreas]

Would you recommend any better .096V vref in MSOP package please?

I see 2 problems in this question.
The smaller (less stiff) the plastic package the more instable the device. (see e.g. ADR293 datasheet).
So a DIP8 device is more stable than a MSOP8

Plastic packages generally have more or less influence from humidity.

So if you need really a stable device over time/humidity there is no way out to use a metal can case or as 2nd best alternative a ceramic case (which has less mechanical decoupling from PCB humidity influences).

4.096V is usually a bandgap reference. The ageing drift of bandgap references tends to be higher than that of buried zener references. (which need a higher power supply voltage since the internal reference is somewhere between 6-7V).

So choosing a optimal reference is always depending on what you really need.

with best regards

Andreas

[alex-sh]

Would you recommend any better .096V vref in MSOP package please?

I see 2 problems in this question.
The smaller (less stiff) the plastic package the more instable the device. (see e.g. ADR293 datasheet).
So a DIP8 device is more stable than a MSOP8

Plastic packages generally have more or less influence from humidity.

So if you need really a stable device over time/humidity there is no way out to use a metal can case or as 2nd best alternative a ceramic case (which has less mechanical decoupling from PCB humidity influences).

4.096V is usually a bandgap reference. The ageing drift of bandgap references tends to be higher than that of buried zener references. (which need a higher power supply voltage since the internal reference is somewhere between 6-7V).

So choosing a optimal reference is always depending on what you really need.

with best regards

Andreas

The problem is that I have already MSOP8 footprint on the PCB, which I do not want to change.
I can do a daughter board and probably accommodate other packages. I have a choice to either
1) leave existing 4.096V LTC6655B <- this would require more frequent calibration
2) change it for DIP8 / LS8 on the daughter board
3) go for LM399 with some LT5400 to order to derive as close as possible 4.096V <- this would probably be the best in terms of long-term.

I am building a probrammable voltage reference.

Thanks
Alex

[iMo]

What about ref5040? Similar spec..

The plastic package absorbs humidity which gives stress to the die and thus changes the output voltage.
If you solder the MSOP8 package it gets dry.

Once I messed with some expensive 144pin tqfp (epoxy) the storage box contained an instruction how to treat the package such it will not crack upon soldering because of moisture - as I can remember they (ADI) recommended to put the package for maybe 24h into 150C prior soldering..

[Andreas]

The problem is that I have already MSOP8 footprint on the PCB, which I do not want to change.


I am building a probrammable voltage reference.

Hello,

So you are fixed to the LTC6655 because it has a unique pinning (force + sense)
Dont get me wrong: the LTC6655 is not really bad. But you cannot expect 5-6 digits accuracy in MSOP.
Its more like 4-5 digits. (as long as you do not apply force to the PCB).
By the way the LTC6655 has high current consumption and so heats up rather high.
Did you keep air drafts away from it?

That does not explain why you need the 4.1V reference.
(you do not really want to make a 1mV step calibrator with a 12 bit DAC).

The best results (hysteresis + humidity stability) up to now I have with the AD586LQ (5V) and AD587JQ (10V)
But for high demands you have to select for tempco and ageing.

The next best device for 4.1V could be a MAX6341. (if you have >=8V stabilized supply voltage).
But all my MAX6250A and MAX6350 (5V in DIP8) from the same series have around 10 ppm/year ageing drift.
And also on many samples I see popcorn noise of 1-2 ppm near certain temperatures.

What about ref5040? Similar spec..

If you ask me: I have no experience with this part.
(and low interest because it is in a plastic package).

And be carefully with "typical" datasheet specs regarding long term drift and hysteresis.
In most cases they have been done once with the "best package option" long time ago.
(e.g. metal can case which is long obsolete).
They are also never updated when the epoxy which is used to fix the part into the package
is changed (e.g. due to RoHS or other change of supplier).

So if you really want to have exact data you have to test every single device under your individual operating conditions.

with best regards

Andreas

[razvan784]

it was 4.097200v with my 6-1/2 DDM (recently calibrated), which was already outside 0.025% max accuracy

Yes, it is off by 0.029%, assuming your meter is spot-on. I guess the specification applies to the component as delivered, before soldering -- because that's how they are able to test and adjust it.
You have to add the hysteresis due to soldering - some plots (typical, not guaranteed!) are given in the applications section of the datasheet.

This means the vref drifted 584uV. This is a huge drift in my view.

That's around 0.014%, or 140 ppm, or around 70 ppm/month. As others have already said, that's not huge for a MSOP package, which is optimized for a small PCB footprint, at the cost of accuracy.
Look in the datasheet, Figure 10 - Long-Term Drift for MS8 and LS8. The MS8 graph show 4 traces (not statistically significant!) and one of them shows around 100 ppm drift - so 140 is not at all unreasonable. The hermetic LS8 is not much greater (50 instead of 100 ppm, though more samples shown), and it is explicitly stated that it is also humidity-sensitive by influence of the PCB. I particularly like how they emphasize that "proper board materials and layout are essential", without stating what those materials might be Ceramic anyone? (See those Fluke calibrators).
So, IMHO you have two options:
* if you really want stability, use a LTZ1000, LM399, or a CERDIP like the AD586. Redesign the circuit to not need exactly 4096 mV and calibrate it in software. Do an initial cal and repeat once in a while.
* if you really want small size and low power, use the LTC6655, or maybe a LT1461 (very low power, excellent TC), or a TI REF50-something, and calibrate it often.

[alex-sh]

Look in the datasheet, Figure 10 - Long-Term Drift for MS8 and LS8. The MS8 graph show 4 traces (not statistically significant!) and one of them shows around 100 ppm drift - so 140 is not at all unreasonable. The hermetic LS8 is not much greater (50 instead of 100 ppm, though more samples shown), and it is explicitly stated that it is also humidity-sensitive by influence of the PCB. I particularly like how they emphasize that "proper board materials and layout are essential", without stating what those materials might be  Ceramic anyone? (See those Fluke calibrators).

You are right:
Long-term drift cannot be extrapolated from accelerated high temperature testing. This erroneous technique gives drift numbers that are wildly optimistic. The only way long-term drift can be determined is to measure it over the time interval of interest. The LTC6655 long-term drift data was collected on 80 parts that were soldered into printed circuit boards similar to a real world application. The boards were then placed into a constant temperature oven with a TA = 35°C, their outputs were scanned regularly and measured with an 8.5 digit DVM. Typical long-term drift is illustrated in Figure 10a. The hermetic LS8 package provides additional stability as shown in Figure 10b.
Sadly, the datasheet is giving an optimictic results (for 2.5V vref only):

* if you really want small size and low power, use the LTC6655, or maybe a LT1461 (very low power, excellent TC), or a TI REF50-something, and calibrate it often.

This would be the best option if the calibration was simple. I have 60uV steps and I can calibrate it in 1mV stepts (4096 steps in total).
Very tedious if I was to do it on a regular basis sadly.

[alex-sh]

Hello,

So you are fixed to the LTC6655 because it has a unique pinning (force + sense)
Dont get me wrong: the LTC6655 is not really bad. But you cannot expect 5-6 digits accuracy in MSOP.
Its more like 4-5 digits. (as long as you do not apply force to the PCB).
By the way the LTC6655 has high current consumption and so heats up rather high.
Did you keep air drafts away from it?

That does not explain why you need the 4.1V reference.
(you do not really want to make a 1mV step calibrator with a 12 bit DAC).

The best results (hysteresis + humidity stability) up to now I have with the AD586LQ (5V) and AD587JQ (10V)
But for high demands you have to select for tempco and ageing.

The next best device for 4.1V could be a MAX6341. (if you have >=8V stabilized supply voltage).
But all my MAX6250A and MAX6350 (5V in DIP8) from the same series have around 10 ppm/year ageing drift.
And also on many samples I see popcorn noise of 1-2 ppm near certain temperatures.

Thank you Andreas
My expectation is not meeting reality I suppose. I was hoping for 5-6 digits accuracy, but with LTC6655 I am disappointed. Good initial results, but very poor long term performance.
Yes, the design of the PCB takes in the account that the vref may heat up and it is isolated away from the rest of the components.
I am using AD5541LRZ (16 bit DAC) and have 1mV steps from 1mV to 4096mV with ability of 60uV calibration steps. I'll look into this, but I think 4.096V vref is not critical. The problem is that the vref is fed from +5V power rail and this means using 5V vref would not be possible without modifying or adding a new power rail.
AD586LQ would be a good one if the design was catering for 10.8V ~ 36V input voltage.
I also have MAX6341CPA+-ND in DIP8 - this means I can use it on the daughter board with a few caps. I can probably live with +/- 40uV / year drift.

I suppose I should think what to do next.
Thanks
Alex

[splin]

Hello,

By the way: independant of LS8 package or MSOP package: my samples showed a relative large hysteresis so I do not consider the LTC6655 as a "precision" reference. The LS8 package is better, but only if you mount it dead bug style on the PCB so that there is no influence from the PCB.

Andreas, have you noticed any change in the hysterisis characteristics of these, or other (non-hermetic) references with repeated cycling? I can't image ceramic packaged devices changing much but it may be possible for relaxation effects in plastic packaged parts to show reduced hysterisis after many temperature cycles.

[alex-sh]

Anyone have had any experience with LT1021BCN8-5 (5v vref)?
From the datasheet, I can operate a 5V Reference from 5V Supply - this would be the best way for me to move from bandgap voltage reference to a zener diode one with a very good long-term drift.

[iMo]

The LT1021-5V needs at least 6.9V based on my measurements.
PS: all LT1021s include a 6.2V burried zener plus some electronics, including trimmed divider (-5, -10), the -7 is without the divider and has got the best long term stability.

[alex-sh]

The LT1021-5V needs at least 6.9V based on my measurements.
PS: all LT1021s include a 6.2V burried zener plus some electronics, including trimmed divider (-5, -10), the -7 is without the divider and has got the best long term stability.

Please take a look at the attachement (this is from the LT1021 datasheet).
I have LT1021BCN8-10 but I hate to do a new power rail + INA105 to drop it to 5V

[Andreas]

Andreas, have you noticed any change in the hysterisis characteristics of these, or other (non-hermetic) references with repeated cycling? I can't image ceramic packaged devices changing much but it may be possible for relaxation effects in plastic packaged parts to show reduced hysterisis after many temperature cycles.

Hello,

of course there are differences between repeated cyclings on plastic packages.
Mainly dependant on humidity.
So you cannot expect them getting better and better in a non humidity controlled environment.

Please take a look at the attachement (this is from the LT1021 datasheet).

Yes but practically you want to operate it above 10V (because of better PSRR).
A charge pump introduces additional noise.

By the way: the LT1236A is the enhanced version of the LT1021B(+C).
(low T.C. and low output voltage tolerance at the same time).
So unless you are using the metal can package for the LT1021 does not give any advantage over the LT1236.

with best regards

Andreas

[iMo]

Please take a look at the attachement (this is from the LT1021 datasheet).
I have LT1021BCN8-10 but I hate to do a new power rail + INA105 to drop it to 5V

The LT1021B-10 requires at least 12V.
LT1021B-5 is not easily trimmable by a pot.
The B versions are pretty off without trimming (ie +0.01 to +0.02V in my case) but with lowest tempco.

[alex-sh]

Please take a look at the attachement (this is from the LT1021 datasheet).

Yes but practically you want to operate it above 10V (because of better PSRR).
A charge pump introduces additional noise.

By the way: the LT1236A is the enhanced version of the LT1021B(+C).
(low T.C. and low output voltage tolerance at the same time).
So unless you are using the metal can package for the LT1021 does not give any advantage over the LT1236.

with best regards

Andreas

That's why I am using 4.096V vref - not to introduce another power rail. I have +5V/-5V/+3.3V and trying to take advantage of any of these.
I wonder if I can live with the charge pump noise...

Sadly, LT1021 in H package (metal can) has been discontinued - this is what digikey is telling me.
LT1236 is available in LS8 hermitically sealed.

Did you have any experience with LT1019A?

[alex-sh]

LT1021B-5 is not easily trimmable by a pot.

I would rather avoid using the pot.
If the vref is not initially accurate (+0.01V) I can probably correct it in software.
Though the idea is to find the holy grail - excellent initially accuracy, low long term drift, small package and low power.
However, looks like no way to have a good long term drift without using TO5-3/TO5-8 package.

[iMo]

Though the idea is to find the holy grail - excellent initially accuracy, low long term drift, small package and low power.

We all do as well..

[Andreas]

Though the idea is to find the holy grail - excellent initially accuracy, low long term drift, small package and low power.

We all do as well..

Hello,

I am now trying since 11 years.
Best result up to now AD586LQ / selected (for low T.C.) AD587JQ.
2nd best LT1236DCLS8-5. There is one Idea that I have not done up to now:

If you are not using the trim pin you should be able to use it to measure
the internal chip temperature. (-1.7 mV/K according to data sheet).

With a Resolution of 0.01K or 17uV and a high impedant buffer amplifier this system
could be better than my AD586 solution with a external NTC and 0.1K
(10 Bit ADC resolution) for temperature compensation.
Under same conditions a AD586 based system delivers a factor 2 lower
standard deviation over 1000 Hours daily measurements.

with best regards

Andreas

[Andreas]

Did you have any experience with LT1019A?

Hello,

nearly overlooked the question.
Not the A-Version but the C-Version. (which was easier to get for me).
here a diagram of the drift in my ageing-Box for the last 1.5 years.
The ageing box is heated every day to 50.0 deg C to measure several references.

You clearly can see the difference between a plastic device and the hermetically ones
even when the box is heated up every day for about 1.5 hours.

with best regards

Andreas